[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-satellite-image-deep-learning--datasets":3,"tool-satellite-image-deep-learning--datasets":62},[4,18,26,35,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",108322,2,"2026-04-10T11:39:34",[14,15,13],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":32,"last_commit_at":41,"category_tags":42,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[43,13,15,14],"插件",{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":10,"last_commit_at":50,"category_tags":51,"status":17},4487,"LLMs-from-scratch","rasbt\u002FLLMs-from-scratch","LLMs-from-scratch 是一个基于 PyTorch 的开源教育项目，旨在引导用户从零开始一步步构建一个类似 ChatGPT 的大型语言模型（LLM）。它不仅是同名技术著作的官方代码库，更提供了一套完整的实践方案，涵盖模型开发、预训练及微调的全过程。\n\n该项目主要解决了大模型领域“黑盒化”的学习痛点。许多开发者虽能调用现成模型，却难以深入理解其内部架构与训练机制。通过亲手编写每一行核心代码，用户能够透彻掌握 Transformer 架构、注意力机制等关键原理，从而真正理解大模型是如何“思考”的。此外，项目还包含了加载大型预训练权重进行微调的代码，帮助用户将理论知识延伸至实际应用。\n\nLLMs-from-scratch 特别适合希望深入底层原理的 AI 开发者、研究人员以及计算机专业的学生。对于不满足于仅使用 API，而是渴望探究模型构建细节的技术人员而言，这是极佳的学习资源。其独特的技术亮点在于“循序渐进”的教学设计：将复杂的系统工程拆解为清晰的步骤，配合详细的图表与示例，让构建一个虽小但功能完备的大模型变得触手可及。无论你是想夯实理论基础，还是为未来研发更大规模的模型做准备",90106,"2026-04-06T11:19:32",[52,15,13,14],"语言模型",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":10,"last_commit_at":59,"category_tags":60,"status":17},4292,"Deep-Live-Cam","hacksider\u002FDeep-Live-Cam","Deep-Live-Cam 是一款专注于实时换脸与视频生成的开源工具，用户仅需一张静态照片，即可通过“一键操作”实现摄像头画面的即时变脸或制作深度伪造视频。它有效解决了传统换脸技术流程繁琐、对硬件配置要求极高以及难以实时预览的痛点，让高质量的数字内容创作变得触手可及。\n\n这款工具不仅适合开发者和技术研究人员探索算法边界，更因其极简的操作逻辑（仅需三步：选脸、选摄像头、启动），广泛适用于普通用户、内容创作者、设计师及直播主播。无论是为了动画角色定制、服装展示模特替换，还是制作趣味短视频和直播互动，Deep-Live-Cam 都能提供流畅的支持。\n\n其核心技术亮点在于强大的实时处理能力，支持口型遮罩（Mouth Mask）以保留使用者原始的嘴部动作，确保表情自然精准；同时具备“人脸映射”功能，可同时对画面中的多个主体应用不同面孔。此外，项目内置了严格的内容安全过滤机制，自动拦截涉及裸露、暴力等不当素材，并倡导用户在获得授权及明确标注的前提下合规使用，体现了技术发展与伦理责任的平衡。",88924,"2026-04-06T03:28:53",[14,15,13,61],"视频",{"id":63,"github_repo":64,"name":65,"description_en":66,"description_zh":67,"ai_summary_zh":67,"readme_en":68,"readme_zh":69,"quickstart_zh":70,"use_case_zh":71,"hero_image_url":72,"owner_login":73,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":76,"owner_email":77,"owner_twitter":76,"owner_website":78,"owner_url":79,"languages":76,"stars":80,"forks":81,"last_commit_at":82,"license":76,"difficulty_score":83,"env_os":75,"env_gpu":84,"env_ram":84,"env_deps":85,"category_tags":88,"github_topics":89,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":95,"updated_at":96,"faqs":97,"releases":108},8936,"satellite-image-deep-learning\u002Fdatasets","datasets","Datasets for deep learning with satellite & aerial imagery","datasets 是一个专为深度学习打造的卫星与航空影像数据资源库，旨在解决遥感领域高质量标注数据分散、难以查找的痛点。它并非单一的数据集，而是一份精心整理的“导航地图”，汇集了全球多个权威开源项目、基准测试集及数据枢纽链接。\n\n无论是需要训练模型的研究人员，还是开发地理空间应用的工程师，都能在这里快速定位所需资源。内容覆盖广泛，包括用于洪水监测的 Sentinel-1 雷达数据、用于地物分类的 Sentinel-2 光学影像，以及作物产量预测、超分辨率算法训练等特定场景的专业数据集。此外，它还整合了 AWS、Google Earth Engine 和 Microsoft Planetary Computer 等主流云平台的数据目录，方便用户直接访问海量云端数据。\n\n其独特亮点在于极强的针对性与实用性：不仅按传感器类型（如 SAR 与光学）和应用场景（如变化检测、时间序列分析）进行了细致分类，还提供了从原始数据下载到代码示例（如 Python\u002FKeras 加载教程）的一站式指引。如果你正在寻找可靠的遥感数据来启动或优化你的 AI 项目，datasets 将是不可或缺的起点。","\u003Cdiv align=\"center\">\n  \u003Cp>\n    \u003Ca href=\"https:\u002F\u002Fwww.satellite-image-deep-learning.com\u002F\">\n        \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsatellite-image-deep-learning_datasets_readme_6558fd4f23db.png\" width=\"700\">\n    \u003C\u002Fa>\n\u003C\u002Fp>\n  \u003Ch2>Datasets for deep learning applied to satellite and aerial imagery.\u003C\u002Fh2>\n\n# 👉 [satellite-image-deep-learning.com](https:\u002F\u002Fwww.satellite-image-deep-learning.com\u002F) 👈\n\n\u003C\u002Fdiv>\n\n**How to use this repository:** if you know exactly what you are looking for (e.g. you have the paper name) you can `Control+F` to search for it in this page (or search in the raw markdown).\n\n# Lists of datasets\n\u003C!-- markdown-link-check-disable -->\n* [Earth Observation Database](https:\u002F\u002Feod-grss-ieee.com\u002F)\n\u003C!-- markdown-link-check-enable -->\n* [awesome-satellite-imagery-datasets](https:\u002F\u002Fgithub.com\u002Fchrieke\u002Fawesome-satellite-imagery-datasets)\n* [Awesome_Satellite_Benchmark_Datasets](https:\u002F\u002Fgithub.com\u002FSeyed-Ali-Ahmadi\u002FAwesome_Satellite_Benchmark_Datasets)\n* [awesome-remote-sensing-change-detection](https:\u002F\u002Fgithub.com\u002Fwenhwu\u002Fawesome-remote-sensing-change-detection) -> dedicated to change detection\n* [Callisto-Dataset-Collection](https:\u002F\u002Fgithub.com\u002FAgri-Hub\u002FCallisto-Dataset-Collection) -> datasets that use Copernicus\u002Fsentinel data\n* [geospatial-data-catalogs](https:\u002F\u002Fgithub.com\u002Fgiswqs\u002Fgeospatial-data-catalogs) -> A list of open geospatial datasets available on AWS, Earth Engine, Planetary Computer, and STAC Index\n* [BED4RS](https:\u002F\u002Fcaptain-whu.github.io\u002FBED4RS\u002F)\n* [Satellite-Image-Time-Series-Datasets](https:\u002F\u002Fgithub.com\u002Fcorentin-dfg\u002FSatellite-Image-Time-Series-Datasets)\n\n# Remote sensing dataset hubs\n* [Radiant MLHub](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241213015814\u002Fhttps:\u002F\u002Fmlhub.earth\u002F) -> both datasets and models\n* [Registry of Open Data on AWS](https:\u002F\u002Fregistry.opendata.aws)\n* [Microsoft Planetary Computer data catalog](https:\u002F\u002Fplanetarycomputer.microsoft.com\u002Fcatalog)\n* [Google Earth Engine Data Catalog](https:\u002F\u002Fdevelopers.google.com\u002Fearth-engine\u002Fdatasets)\n\n## Sentinel\nAs part of the [EU Copernicus program](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCopernicus_Programme), multiple Sentinel satellites are capturing imagery -> see [wikipedia](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCopernicus_Programme#Sentinel_missions)\n\n### Sentinel-1 (SAR)\n* [Xarray backend to Copernicus Sentinel-1 satellite data products](https:\u002F\u002Fgithub.com\u002Fbopen\u002Fxarray-sentinel)\n* [mmflood](https:\u002F\u002Fgithub.com\u002Fedornd\u002Fmmflood) -> Flood delineation from Sentinel-1 SAR imagery, with [paper](https:\u002F\u002Fieeexplore.ieee.org\u002Fabstract\u002Fdocument\u002F9882096)\n* [Sentinel-1 for Science Amazonas](https:\u002F\u002Fsen4ama.gisat.cz\u002F) -> forest lost time series dataset\n* [CYCleSS](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41597-025-06528-x) -> A comprehensive UK crop yield dataset incorporating satellite, weather, and soil type informations\n\n### Sentinel-2 (Optical)\n* [Sentinel-2 Cloud-Optimized GeoTIFFs](https:\u002F\u002Fregistry.opendata.aws\u002Fsentinel-2-l2a-cogs\u002F) and [Sentinel-2 L2A 120m Mosaic](https:\u002F\u002Fregistry.opendata.aws\u002Fsentinel-s2-l2a-mosaic-120\u002F)\n* [Open access data on GCP](https:\u002F\u002Fconsole.cloud.google.com\u002Fstorage\u002Fbrowser\u002Fgcp-public-data-sentinel-2?prefix=tiles%2F31%2FT%2FCJ%2F)\n* [Example loading sentinel data in a notebook](https:\u002F\u002Fgithub.com\u002Fbinder-examples\u002Fgetting-data\u002Fblob\u002Fmaster\u002FSentinel2.ipynb)\n* [Analyzing Sentinel-2 satellite data in Python with Keras](https:\u002F\u002Fgithub.com\u002Fjensleitloff\u002FCNN-Sentinel)\n* [SEN2VENµS](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6514159#.YoRxM5PMK3I) -> a dataset for the training of Sentinel-2 super-resolution algorithms\n* [M3LEO](https:\u002F\u002Fhuggingface.co\u002FM3LEO) -> [Github](https:\u002F\u002Fgithub.com\u002Fspaceml-org\u002FM3LEO). A very large scale georeferenced dataset of Sentinel 1\u002F2 imagery plus interferometric SAR products and auxiliary datasets such as Land cover, Biomass and Digital Elevation Models.\n* [SEN12MS](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSEN12MS) -> A Curated Dataset of Georeferenced Multi-spectral Sentinel-1\u002F2 Imagery for Deep Learning and Data Fusion. Checkout [SEN12MS toolbox](https:\u002F\u002Fgithub.com\u002Fschmitt-muc\u002FSEN12MS) and many referenced uses on [paperswithcode.com](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fsen12ms)\n* [SEN2NAIP](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Ftacofoundation\u002FSEN2NAIPv2) -> Spatially and spectrally harmonized Sen-2 + NAIP dataset for 4x RGB-NIR super-resolution.\n* [Sen4AgriNet](https:\u002F\u002Fgithub.com\u002FOrion-AI-Lab\u002FS4A) -> A Sentinel-2 multi-year, multi-country benchmark dataset for crop classification and segmentation with deep learning, with and [models](https:\u002F\u002Fgithub.com\u002FOrion-AI-Lab\u002FS4A-Models)\n* [sentinel2tools](https:\u002F\u002Fgithub.com\u002FQuantuMobileSoftware\u002Fsentinel2tools) -> downloading & basic processing of Sentinel 2 imagesry. Read [Sentinel2tools: simple lib for downloading Sentinel-2 satellite images](https:\u002F\u002Fmedium.com\u002Fgeekculture\u002Fsentinel2tools-simple-lib-for-downloading-sentinel-2-satellite-images-f8a6be3ee894)\n* [open-sentinel-map](https:\u002F\u002Fgithub.com\u002FVisionSystemsInc\u002Fopen-sentinel-map) -> The OpenSentinelMap dataset contains Sentinel-2 imagery and per-pixel semantic label masks derived from OpenStreetMap\n* [Canadian-cropland-dataset](https:\u002F\u002Fgithub.com\u002FbioinfoUQAM\u002FCanadian-cropland-dataset) -> a novel patch-based dataset compiled using optical satellite images of Canadian agricultural croplands retrieved from Sentinel-2\n* [Sentinel-2 Cloud Cover Segmentation Dataset](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20230606184945\u002Fhttps:\u002F\u002Fmlhub.earth\u002Fdata\u002Fref_cloud_cover_detection_challenge_v1) on Radiant mlhub\n* [The Azavea Cloud Dataset](https:\u002F\u002Fwww.azavea.com\u002Fblog\u002F2021\u002F08\u002F02\u002Fthe-azavea-cloud-dataset\u002F) which is used to train this [cloud-model](https:\u002F\u002Fgithub.com\u002Fazavea\u002Fcloud-model)\n* [fMoW-Sentinel](https:\u002F\u002Fpurl.stanford.edu\u002Fvg497cb6002) -> The Functional Map of the World - Sentinel-2 corresponding images (fMoW-Sentinel) dataset consists of image time series collected by the Sentinel-2 satellite, corresponding to locations from the Functional Map of the World (fMoW) dataset across several different times. Used in [SatMAE](https:\u002F\u002Fgithub.com\u002Fsustainlab-group\u002FSatMAE)\n* [Earth Surface Water Dataset](https:\u002F\u002Fzenodo.org\u002Frecord\u002F5205674#.Y4iEFezP1hE) -> a dataset for deep learning of surface water features on Sentinel-2 satellite images. See [this ref using it in torchgeo](https:\u002F\u002Ftowardsdatascience.com\u002Fartificial-intelligence-for-geospatial-analysis-with-pytorchs-torchgeo-part-1-52d17e409f09)\n* [Ship-S2-AIS dataset](https:\u002F\u002Fzenodo.org\u002Frecord\u002F7229756#.Y5GsgOzP1hE) -> 13k tiles extracted from 29 free Sentinel-2 products. 2k images showing ships in Denmark sovereign waters: one may detect cargos, fishing, or container ships\n* [Amazon Rainforest dataset for semantic segmentation](https:\u002F\u002Fzenodo.org\u002Frecord\u002F3233081#.Y6LPLOzP1hE) -> Sentinel 2 images\n* [MATTER](https:\u002F\u002Fgithub.com\u002Fperiakiva\u002FMATTER) -> a Sentinel 2 dataset for Self-Supervised Training\n* [S2GLC](https:\u002F\u002Fs2glc.cbk.waw.pl\u002F) -> High resolution Land Cover Map of Europe\n* [Generating Imperviousness Maps from Multispectral Sentinel-2 Satellite Imagery](https:\u002F\u002Fzenodo.org\u002Frecord\u002F7058860#.ZDrAeuzMLdo)\n* [Sentinel-2 Water Edges Dataset (SWED)](https:\u002F\u002Fopenmldata.ukho.gov.uk\u002F)\n* [Sentinel2 Munich480](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fartelabsuper\u002Fsentinel2-munich480) -> dataset for crop mapping by exploiting the time series of Sentinel-2 satellite\n* [Meadows vs Orchards](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbaptistel\u002Fmeadows-vs-orchards) -> a pixel time series dataset\n* [Sentinel-2 Image Time Series for Crop Mapping](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fignazio\u002Fsentinel2-crop-mapping) -> data for the Lombardy region in Italy\n* [Deforestation in Ukraine from Sentinel2 data](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fisaienkov\u002Fdeforestation-in-ukraine)\n* [satellite-change-events](https:\u002F\u002Fwww.cs.cornell.edu\u002Fprojects\u002Fsatellite-change-events\u002F) -> CaiRoad & CalFire change detection Sentinel 2 datasets\n* [Sentinel-2 dataset for ship detection](https:\u002F\u002Fzenodo.org\u002Frecords\u002F3923841), also edited and redistributed as [VDS2RAW](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7982468#.ZIiLxS8QOo4)\n* [MineSegSAT](https:\u002F\u002Fgithub.com\u002Fmacdonaldezra\u002FMineSegSAT) -> dataset for paper: AN AUTOMATED SYSTEM TO EVALUATE MINING DISTURBED AREA EXTENTS FROM SENTINEL-2 IMAGERY\n* [CaBuAr](https:\u002F\u002Fgithub.com\u002FDarthReca\u002FCaBuAr) -> California Burned Areas dataset for delineation\n* [sen12mscr](https:\u002F\u002Fpatricktum.github.io\u002Fcloud_removal\u002Fsen12mscr\u002F) -> Multimodal Cloud Removal\n* [Greenearthnet](https:\u002F\u002Fgithub.com\u002Fvitusbenson\u002Fgreenearthnet\u002Ftree\u002Fmain) -> dataset specifically designed for high-resolution vegetation forecasting\n* [Floating-Marine-Debris-Data](https:\u002F\u002Fgithub.com\u002Fmiguelmendesduarte\u002FFloating-Marine-Debris-Data) -> floating marine debris, with annotations for six debris classes, including plastic, driftwood, seaweed, pumice, sea snot, and sea foam.\n* [Sen2Fire](https:\u002F\u002Fzenodo.org\u002Frecords\u002F10881058) -> A Challenging Benchmark Dataset for Wildfire Detection using Sentinel Data\n* [L1BSR](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7826696) -> 3740 pairs of overlapping image crops extracted from two L1B products\n* [GloSoFarID](https:\u002F\u002Fgithub.com\u002Fyzyly1992\u002FGloSoFarID) -> Global multispectral dataset for Solar Farm IDentification\n* [MARIDA](https:\u002F\u002Fmarine-debris.github.io\u002Findex.html) -> Marine Debris detection from Sentinel-2\n* [MADOS](https:\u002F\u002Fgithub.com\u002Fgkakogeorgiou\u002Fmados) -> Marine Debris and Oil Spill from Sentinel-2\n* [Sentinel-2 dataset for ship detection and characterization](https:\u002F\u002Fzenodo.org\u002Frecords\u002F10418786) -> RGB\n* [S2-SHIPS](https:\u002F\u002Fgithub.com\u002Falina2204\u002Fcontrastive_SSL_ship_detection) -> all 12 channels\n* [ChatEarthNet](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FChatEarthNet) -> A Global-Scale Image-Text Dataset Empowering Vision-Language Geo-Foundation Models, utilizes Sentinel-2 data with captions generated by ChatGPT\n* [UKFields](https:\u002F\u002Fgithub.com\u002FSpiruel\u002FUKFields) -> over 2.3 million automatically delineated field boundaries spanning England, Wales, Scotland, and Northern Ireland\n* [ShipWakes](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7947694) -> Keypoints Method for Recognition of Ship Wake Components in Sentinel-2 Images by Deep Learning\n* [TimeSen2Crop](https:\u002F\u002Fzenodo.org\u002Frecords\u002F4715631) -> a Million Labeled Samples Dataset of Sentinel 2 Image Time Series for Crop Type Classification\n* [AgriSen-COG](https:\u002F\u002Fgithub.com\u002Ftselea\u002Fagrisen-cog) -> a Multicountry, Multitemporal Large-Scale Sentinel-2 Benchmark Dataset for Crop Mapping: includes an anomaly detection preprocessing step\n* [MagicBathyNet](https:\u002F\u002Fwww.magicbathy.eu\u002Fmagicbathynet.html) -> a new multimodal benchmark dataset made up of image patches of Sentinel-2, SPOT-6 and aerial imagery, bathymetry in raster format and seabed classes annotations\n* [MuS2: A Benchmark for Sentinel-2 Multi-Image Super-Resolution](https:\u002F\u002Fdataverse.harvard.edu\u002Fdataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2F1JMRAT)\n* [Sen4Map](https:\u002F\u002Fdatapub.fz-juelich.de\u002Fsen4map\u002F) -> Sentinel-2 time series images, covering over 335,125 geo-tagged locations across the European Union. These geo-tagged locations are associated with detailed landcover and land-use information\n* [CloudSEN12Plus](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fisp-uv-es\u002FCloudSEN12Plus) -> the largest cloud detection dataset to date for Sentinel-2\n* [mayrajeo S2 ship detection](https:\u002F\u002Fgithub.com\u002Fmayrajeo\u002Fship-detection) -> labels for Detecting marine vessels from Sentinel-2 imagery with YOLOv8\n* [Fields of The World](https:\u002F\u002Ffieldsofthe.world\u002F) -> instance segmentation of agricultural field boundaries\n* [ai4boundaries](https:\u002F\u002Fgithub.com\u002Fwaldnerf\u002Fai4boundaries) -> field boundaries with Sentinel-2 and aerial photography\n* [California Wildfire GeoImaging Dataset - CWGID](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.16380) -> Development and Application of a Sentinel-2 Satellite Imagery Dataset for Deep-Learning Driven Forest Wildfire Detection\n* [substation-seg](https:\u002F\u002Fgithub.com\u002FLindsay-Lab\u002Fsubstation-seg) -> segmenting substations dataset\n* [PhilEO-downstream](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FPhilEO-community\u002FPhilEO-downstream) -> a 400GB Sentinel-2 dataset for building density estimation, road segmentation, and land cover classification.\n* [PhilEO-pretrain](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FPhilEO-community\u002FPhilEO-pretrain) -> a 500GB global dataset of Sentinel-2 images for model pre-training.\n* [KappaSet: Sentinel-2 KappaZeta Cloud and Cloud Shadow Masks](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7100327)\n* [AllClear](https:\u002F\u002Fallclear.cs.cornell.edu\u002F) A Comprehensive Dataset and Benchmark for Cloud Removal in Satellite Imagery\n* [Sentinel-2 reference cloud masks generated by an active learning method](https:\u002F\u002Fzenodo.org\u002Frecords\u002F1460961)\n* [Cloud gap-filling with deep learning for improved grassland monitoring](https:\u002F\u002Fzenodo.org\u002Frecords\u002F11651601)\n* [Remote Sensing Ship Wake Dataset](https:\u002F\u002Fgithub.com\u002Fzjze\u002FRSSW_Dateset)\n* [ERAS-dataset](https:\u002F\u002Fgithub.com\u002Fcscribano\u002FERAS-dataset) -> Emilia-Romagna Agri Seg (ERAS) field segmentation dataset\n* [Sentinel 2 super-resolved data cubes - 92 scenes over 2 regions in Switzerland spanning 5 years](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241003205022\u002Fhttps:\u002F\u002Fieee-dataport.org\u002Fdocuments\u002Fsentinel-2-super-resolved-data-cubes-92-scenes-over-2-regions-switzerland-spanning-5-years)\n* [SeasoNet](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6979994) -> A Seasonal Scene Classification, Segmentation and Retrieval Dataset for Satellite Imagery over Germany. Land cover classes based on the CORINE Land Cover database (CLC) 2018\n* [EuroCropsML](https:\u002F\u002Fgithub.com\u002Fdida-do\u002Feurocropsml) -> a ready-to-use benchmark dataset for few-shot crop type classification using Sentinel-2 imagery\n* [CanadaFireSat](https:\u002F\u002Fgithub.com\u002Feceo-epfl\u002FCanadaFireSat-Data) -> Sentinel-2 Level-1C time series\n* [ssl4eco](https:\u002F\u002Fgithub.com\u002FPlekhanovaElena\u002Fssl4eco) -> a recipe for building pretraining sets that capture the geographical and phenological diversity of ecosystems across the globe\n* [IRRISIGHT](https:\u002F\u002Fgithub.com\u002FNibir088\u002FIRRISIGHT) -> a large-scale, multimodal remote sensing dataset for irrigation classification, soil-water mapping, and agricultural monitoring.\n* [SentinelKilnDB](https:\u002F\u002Fsustainability-lab.github.io\u002Fsentinelkilndb\u002F) -> Sentinel-2 dataset for monitoring brick kiln emissions in South Asia\n* [MSSWD - Multi-Spectral Ship Wake Dataset](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13870226)\n* [MOSAIC-SEN2-CC](https:\u002F\u002Fgithub.com\u002FChangeCapsInRS\u002FMOSAIC-SEN2-CC) -> A Multispectral Dataset and Adaptation Framework for Remote Sensing Change Captioning\n* [PLUTo](https:\u002F\u002Fzenodo.org\u002Frecords\u002F15629667) -> post-deforestation land uses across the tropics\n* [SentinelKilnDB](https:\u002F\u002Fgithub.com\u002Frishabh-mondal\u002FSENTINELKILNDB_NeurIPS_2025) -> A Large-Scale Dataset and Benchmark for Oriented Bounding Box (OBB) Brick Kiln Detection in South Asia Using Satellite Imagery\n* [GSDD](https:\u002F\u002Fzenodo.org\u002Frecords\u002F17161810) -> Global Supraglacial Debris Dataset\n* [MT4AFE](https:\u002F\u002Fzenodo.org\u002Frecords\u002F15395167) -> Multi-Task Learning for Agricultural Field Extraction\n* [agripotential](https:\u002F\u002Fgithub.com\u002FMohammadElSakka\u002Fagripotential) ->  a Satellite Image Time Series (STIS) of 34 Sentinel-2 timeframes with 5 classes of agriculutural potential\n* [YieldSAT](https:\u002F\u002Fyieldsat.github.io\u002F) -> A Multimodal Benchmark Dataset for High-Resolution Crop Yield Prediction\n\n### Combined Sentinel\n* [awesome-sentinel](https:\u002F\u002Fgithub.com\u002FFernerkundung\u002Fawesome-sentinel) -> a curated list of awesome tools, tutorials and APIs related to data from the Copernicus Sentinel Satellites.\n* Paid access to Sentinel & Landsat data via [sentinel-hub](https:\u002F\u002Fwww.sentinel-hub.com\u002F) and [python-api](https:\u002F\u002Fgithub.com\u002Fsentinel-hub\u002Fsentinelhub-py)\n* [Jupyter Notebooks for working with Sentinel-5P Level 2 data stored on S3](https:\u002F\u002Fgithub.com\u002FSentinel-5P\u002Fdata-on-s3). The data can be browsed [here](https:\u002F\u002Fmeeo-s5p.s3.amazonaws.com\u002Findex.html#\u002F?t=catalogs)\n* [Sentinel NetCDF data](https:\u002F\u002Fgithub.com\u002Facgeospatial\u002FSentinel-5P\u002Fblob\u002Fmaster\u002FSentinel_5P.ipynb)\n* [earthspy](https:\u002F\u002Fgithub.com\u002FAdrienWehrle\u002Fearthspy) -> Monitor and study any place on Earth and in Near Real-Time (NRT) using the Sentinel Hub services developed by the EO research team at Sinergise\n* [Gold Mining and clandestine airstrips datasets](https:\u002F\u002Fgithub.com\u002Fearthrise-media\u002Fmining-detector)\n* [Industrial Smoke Plumes](https:\u002F\u002Fzenodo.org\u002Frecord\u002F4250706)\n* [MARIDA: Marine Debris Archive](https:\u002F\u002Fgithub.com\u002Fmarine-debris\u002Fmarine-debris.github.io)\n* [OMS2CD](https:\u002F\u002Fgithub.com\u002FDibz15\u002FOpenMineChangeDetection) -> hand-labelled images for change-detection in open-pit mining areas\n* [coal power plants' emissions](https:\u002F\u002Ftransitionzero.medium.com\u002Festimating-coal-power-plant-operation-from-satellite-images-with-computer-vision-b966af56919e) -> a dataset of coal power plants' emissions, including images, metadata and labels.\n* [RapidAI4EO](https:\u002F\u002Frapidai4eo.radiant.earth\u002F) -> dense time series satellite imagery sampled at 500,000 locations across Europe, comprising S2 & Planet imagery, with CORINE Land Cover multiclass labels for 2018\n* [MS-HS-BCD-dataset](https:\u002F\u002Fgithub.com\u002Farcgislearner\u002FMS-HS-BCD-dataset) -> multisource change detection dataset used in paper: Building Change Detection with Deep Learning by Fusing Spectral and Texture Features of Multisource Remote Sensing Images: A GF-1 and Sentinel 2B Data Case\n* [CropNet: An Open Large-Scale Dataset with Multiple Modalities for Climate Change-aware Crop Yield Predictions](https:\u002F\u002Fgithub.com\u002Ffudong03\u002FCropNet) -> terabyte-sized, publicly available, and multi-modal dataset for climate change-aware crop yield predictions\n* [Tiny CropNet dataset](https:\u002F\u002Fgithub.com\u002Ffudong03\u002FMMST-ViT)\n* [Multitask Learning for Estimating Power Plant Greenhouse Gas Emissions from Satellite Imagery](https:\u002F\u002Fzenodo.org\u002Frecord\u002F5644746)\n* [METER-ML: A Multi-sensor Earth Observation Benchmark for Automated Methane Source Mapping](https:\u002F\u002Fstanfordmlgroup.github.io\u002Fprojects\u002Fmeter-ml\u002F) -> data [on Zenodo](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6911013)\n* [MultiSenGE](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6375466) -> large-scale multimodal and multitemporal benchmark dataset\n* [SEN12MS](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSEN12MS) -> A Curated Dataset of Georeferenced Multi-spectral Sentinel-1\u002F2 Imagery for Deep Learning and Data Fusion. Checkout [SEN12MS toolbox](https:\u002F\u002Fgithub.com\u002Fschmitt-muc\u002FSEN12MS) and many referenced uses on [paperswithcode.com](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fsen12ms)\n* [Space2Ground](https:\u002F\u002Fgithub.com\u002FAgri-Hub\u002FSpace2Ground) -> dataset with Space (Sentinel-1\u002F2) and Ground (street-level images) components, annotated with crop-type labels for agriculture monitoring.\n* [MSCDUnet](https:\u002F\u002Fgithub.com\u002FLihy256\u002FMSCDUnet) -> change detection datasets containing VHR, multispectral (Sentinel-2) and SAR (Sentinel-1)\n* [OMBRIA](https:\u002F\u002Fgithub.com\u002Fgeodrak\u002FOMBRIA) -> Sentinel-1 & 2 dataset for adressing the flood mapping problem\n* [Satellite Burned Area Dataset](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6597139#.Y9ufiezP1hE) -> segmentation dataset containing several satellite acquisitions related to past forest wildfires. It contains 73 acquisitions from Sentinel-2 and Sentinel-1 (Copernicus).\n* [SEN12_GUM](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6914898) -> SEN12 Global Urban Mapping Dataset\n* [Sentinel-1&2 Image Pairs (SAR & Optical)](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Frequiemonk\u002Fsentinel12-image-pairs-segregated-by-terrain)\n* [MSOSCD](https:\u002F\u002Fgithub.com\u002FLihy256\u002FMSCDUnet) -> change detection datasets containing VHR, multispectral (Sentinel-2) and SAR (Sentinel-1)\n* [SICKLE](https:\u002F\u002Fgithub.com\u002FDepanshu-Sani\u002FSICKLE) -> A Multi-Sensor Satellite Imagery Dataset Annotated with Multiple Key Cropping Parameters. Multi-resolution time-series images from Landsat-8, Sentinel-1, and Sentinel-2\n* [Sentinel-1 and Sentinel-2 Vessel Detection](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fvessel-detection-sentinels)\n* [TreeSatAI](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6780578) -> Sentinel-1, Sentinel-2\n* [AI2-S2-NAIP](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fallenai\u002Fs2-naip) -> aligned NAIP, Sentinel-2, Sentinel-1, and Landsat images spanning the entire continental US\n* [POPCORN: High-resolution Population Maps Derived from Sentinel-1 and Sentinel-2](https:\u002F\u002Fpopcorn-population.github.io\u002F)\n* [CropClimateX](https:\u002F\u002Fgithub.com\u002Fdrnhhl\u002FCropClimateX) -> A large-scale Multitask, Multisensory Dataset for Crop Monitoring under Climate Extremes\n* [SmallMinesDS](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fellaampy\u002FSmallMinesDS) -> A Multimodal Dataset for Mapping Artisanal and Small-Scale Gold Mines. Imagery reused in [CocoaMiningDS](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fellaampy\u002FCocoaMiningDS)\n* [Hoss-ReID](https:\u002F\u002Fgithub.com\u002FAlioth2000\u002FHoss-ReID) -> Cross-modal Ship Re-Identification via Optical and SAR Imagery\n*  [IDEABench Benchmark Dataset](https:\u002F\u002Fgithub.com\u002FIDEAtlas\u002Fai-dua-mapping) -> Mapping and Benchmarking Urban Deprivation for a Global Sample of Cities\n* [ImpactMesh](https:\u002F\u002Fgithub.com\u002FIBM\u002FImpactMesh) -> a large-scale multimodal, multitemporal dataset for flood and wildfire mapping\n* [Sen12Landslides](https:\u002F\u002Fgithub.com\u002FPaulH97\u002FSen12Landslides) -> Spatio-Temporal Landslide & Anomaly Detection Dataset\n* [Cryo-Bench](https:\u002F\u002Fgithub.com\u002FSk-2103\u002FCryo-Bench) -> A Benchmark for Evaluating Geospatial Foundation Models on Cryosphere Applications\n* [BigEarthNet.txt](https:\u002F\u002Ftxt.bigearth.net\u002F) -> A Large-Scale Multi-Sensor Image-Text Dataset and Benchmark for Earth Observation\n* [Borneo_Forest_Disturbance_Dataset](https:\u002F\u002Fgithub.com\u002FColmKeyes\u002FBorneo_Forest_Disturbance_Dataset) -> Forest Disturbance Dataset utilising Sentinel-2 data and RADD alert disturbances.\n\n## Landsat\nLong running US program -> see [Wikipedia](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FLandsat_program)\n* 8 bands, 15 to 60 meters, 185km swath, the temporal resolution is 16 days\n* [Landsat 4, 5, 7, and 8 imagery on Google](https:\u002F\u002Fcloud.google.com\u002Fstorage\u002Fdocs\u002Fpublic-datasets\u002Flandsat), see [the GCP bucket here](https:\u002F\u002Fconsole.cloud.google.com\u002Fstorage\u002Fbrowser\u002Fgcp-public-data-landsat\u002F), with Landsat 8 imagery in COG format analysed in [this notebook](https:\u002F\u002Fgithub.com\u002Fpangeo-data\u002Fpangeo-example-notebooks\u002Fblob\u002Fmaster\u002Flandsat8-cog-ndvi.ipynb)\n* [Landsat 8 imagery on AWS](https:\u002F\u002Fregistry.opendata.aws\u002Flandsat-8\u002F), with many tutorials and tools listed\n* https:\u002F\u002Fgithub.com\u002Fkylebarron\u002Flandsat-mosaic-latest -> Auto-updating cloudless Landsat 8 mosaic from AWS SNS notifications\n* [Visualise landsat imagery using Datashader](https:\u002F\u002Fexamples.pyviz.org\u002Flandsat\u002Flandsat.html#landsat-gallery-landsat)\n* [Landsat-mosaic-tiler](https:\u002F\u002Fgithub.com\u002Fkylebarron\u002Flandsat-mosaic-tiler) -> This repo hosts all the code for landsatlive.live website and APIs.\n* [LandsatSCD](https:\u002F\u002Fgithub.com\u002FggsDing\u002FSCanNet\u002Ftree\u002Fmain) -> a change detection dataset, it consists of 8468 pairs of images, each having the spatial resolution of 416 × 416\n* [The Landsat Irish Coastal Segmentation Dataset](https:\u002F\u002Fzenodo.org\u002Frecords\u002F8414665)\n* [Wildfire-Spread-Dataset](https:\u002F\u002Fgithub.com\u002FBEEILAB\u002FWildfire-Spread-Dataset) -> ABNextFire: A Multi-Source Deep Learning Based Dataset for Wildfire Spread Prediction\n\n## VENμS\nVegetation and Environment monitoring on a New Micro-Satellite ([VENμS](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FVEN%CE%BCS))\n* [VENUS L2A Cloud-Optimized GeoTIFFs](https:\u002F\u002Fregistry.opendata.aws\u002Fvenus-l2a-cogs\u002F)\n* [VENuS cloud mask training dataset](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7040177)\n* [Sen2Venµs](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6514159) -> a dataset for the training of Sentinel-2 super-resolution algorithms\n* [sen2venus-pytorch-dataset](https:\u002F\u002Fgithub.com\u002Fpiclem\u002Fsen2venus-pytorch-dataset) -> torch dataloader and other utilities\n\n## Vantor\nSatellites owned by Vantor (formerly Maxar & DigitalGlobe) include [GeoEye-1](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FGeoEye-1), [WorldView-2](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWorldView-2), [3](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWorldView-3) & [4](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWorldView-4)\n* [Maxar Open Data Program](https:\u002F\u002Fgithub.com\u002Fopengeos\u002Fmaxar-open-data) provides pre and post-event high-resolution satellite imagery in support of emergency planning, response, damage assessment, and recovery\n* [WorldView-2 European Cities](https:\u002F\u002Fearth.esa.int\u002Feogateway\u002Fcatalog\u002Fworldview-2-european-cities) -> dataset covering the most populated areas in Europe at 40 cm resolution\n\n## Planet\nAlso see Spacenet-7 and the Kaggle ship and plane classifications datasets later in this page\n* [Planet’s high-resolution, analysis-ready mosaics of the world’s tropics](https:\u002F\u002Fwww.planet.com\u002Fnicfi\u002F), supported through Norway’s International Climate & Forests Initiative. [BBC coverage](https:\u002F\u002Fwww.bbc.co.uk\u002Fnews\u002Fscience-environment-54651453)\n* Planet have made imagery available via kaggle competitions\n* [Alberta Wells Dataset](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13743323) -> Pinpointing Oil and Gas Wells from Satellite Imagery\n* [ARGO ship classification dataset](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6058710) -> 1750 labelled images from PlanetScope-4-Band satelites. Created [here](https:\u002F\u002Fgithub.com\u002Felizamanelli\u002Fship_dataset\u002Ftree\u002Fmain)\n* [Marine Debris Dataset for Object Detection in Planetscope Imagery](https:\u002F\u002Fcmr.earthdata.nasa.gov\u002Fsearch\u002Fconcepts\u002FC2781412735-MLHUB.html)\n* [LitterLines](https:\u002F\u002Fgithub.com\u002FgeoJoost\u002FLitterLines) -> An Annotated Dataset for Detection of Marine Litter Windrows in PlanetScope Imagery\n* [FloodPlanet Inundation Dataset](https:\u002F\u002Fzenodo.org\u002Frecords\u002F15238572) -> multi-sensor co-registered dataset labeled based on 3m PlanetScope data and spatially overlapping, temporally near Sentinel-1, Sentinel-2, and Landsat-8 data\n* [Zhijie_FloodPlanet_2023](https:\u002F\u002Fdatacommons.cyverse.org\u002Fbrowse\u002Fiplant\u002Fhome\u002Fshared\u002Fcommons_repo\u002Fcurated\u002FZhijie_FloodPlanet_2023) -> 19 flood events that occurred between 2017 and 2020\n\n## UC Merced\nLand use classification dataset with 21 classes and 100 RGB TIFF images for each class. Each image measures 256x256 pixels with a pixel resolution of 1 foot\n* http:\u002F\u002Fweegee.vision.ucmerced.edu\u002Fdatasets\u002Flanduse.html\n* Also [available as a multi-label dataset](https:\u002F\u002Ftowardsdatascience.com\u002Fmulti-label-land-cover-classification-with-deep-learning-d39ce2944a3d)\n* Read [Vision Transformers for Remote Sensing Image Classification](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F13\u002F3\u002F516\u002Fhtm) where a Vision Transformer classifier achieves 98.49% classification accuracy on Merced\n\n## EuroSAT\nLand use classification dataset of Sentinel-2 satellite images covering 13 spectral bands and consisting of 10 classes with 27000 labeled and geo-referenced samples. Available in RGB and 13 band versions\n* [EuroSAT: Land Use and Land Cover Classification with Sentinel-2](https:\u002F\u002Fgithub.com\u002Fphelber\u002FEuroSAT) -> publication where a CNN achieves a classification accuracy 98.57%\n* Repos using fastai [here](https:\u002F\u002Fgithub.com\u002Fshakasom\u002FDeep-Learning-for-Satellite-Imagery) and [here](https:\u002F\u002Fwww.luigiselmi.eu\u002Feo\u002Flulc-classification-deeplearning.html)\n* [evolved_channel_selection](http:\u002F\u002Fmatpalm.com\u002Fblog\u002Fevolved_channel_selection\u002F) -> explores the trade off between mixed resolutions and whether to use a channel at all, with [repo](https:\u002F\u002Fgithub.com\u002Fmatpalm\u002Fevolved_channel_selection)\n* RGB version available as [dataset in pytorch](https:\u002F\u002Fpytorch.org\u002Fvision\u002Fstable\u002Fgenerated\u002Ftorchvision.datasets.EuroSAT.html#torchvision.datasets.EuroSAT) with the 13 band version [in torchgeo](https:\u002F\u002Ftorchgeo.readthedocs.io\u002Fen\u002Flatest\u002Fapi\u002Fdatasets.html#eurosat). Checkout the tutorial on [data augmentation with this dataset](https:\u002F\u002Ftorchgeo.readthedocs.io\u002Fen\u002Flatest\u002Ftutorials\u002Ftransforms.html)\n* [EuroSAT-SAR](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fwangyi111\u002FEuroSAT-SAR) -> matched each Sentinel-2 image in EuroSAT with one Sentinel-1 patch according to the geospatial coordinates\n\n## PatternNet\nLand use classification dataset with 38 classes and 800 RGB JPG images for each class\n* https:\u002F\u002Fsites.google.com\u002Fview\u002Fzhouwx\u002Fdataset?authuser=0\n* Publication: [PatternNet: A Benchmark Dataset for Performance Evaluation of Remote Sensing Image Retrieval](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.03424)\n\n## Gaofen Image Dataset (GID) for classification\n- https:\u002F\u002Fcaptain-whu.github.io\u002FGID\u002F\n- a large-scale classification set and a fine land-cover classification set\n\n## Million-AID\nA large-scale benchmark dataset containing million instances for RS scene classification, 51 scene categories organized by the hierarchical category\n* https:\u002F\u002Fcaptain-whu.github.io\u002FDiRS\u002F\n* [Pretrained models](https:\u002F\u002Fgithub.com\u002FViTAE-Transformer\u002FViTAE-Transformer-Remote-Sensing)\n* Also see [AID](https:\u002F\u002Fcaptain-whu.github.io\u002FAID\u002F), [AID-Multilabel-Dataset](https:\u002F\u002Fgithub.com\u002FHua-YS\u002FAID-Multilabel-Dataset) & [DFC15-multilabel-dataset](https:\u002F\u002Fgithub.com\u002FHua-YS\u002FDFC15-Multilabel-Dataset)\n\n## DIOR object detection dataset\nA large-scale benchmark dataset for object detection in optical remote sensing images, which consists of 23,463 images and 192,518 object instances annotated with horizontal bounding boxes\n* https:\u002F\u002Fgcheng-nwpu.github.io\u002F\n* https:\u002F\u002Farxiv.org\u002Fabs\u002F1909.00133\n* [ors-detection](https:\u002F\u002Fgithub.com\u002FVlad15lav\u002Fors-detection) -> Object Detection on the DIOR dataset using YOLOv3\n* [dior_detect](https:\u002F\u002Fgithub.com\u002Fhm-better\u002Fdior_detect) -> benchmarks for object detection on DIOR dataset\n* [Tools](https:\u002F\u002Fgithub.com\u002FCrazyStoneonRoad\u002FTools) -> for dealing with the DIOR\n* [Object_Detection_Satellite_Imagery_Yolov8_DIOR](https:\u002F\u002Fgithub.com\u002FJohnPPinto\u002FObject_Detection_Satellite_Imagery_Yolov8_DIOR)\n\n## Multiscene\nMultiScene dataset aims at two tasks: Developing algorithms for multi-scene recognition & Network learning with noisy labels\n* https:\u002F\u002Fmultiscene.github.io\u002F & https:\u002F\u002Fgithub.com\u002FHua-YS\u002FMulti-Scene-Recognition\n\n## FAIR1M object detection dataset\nA Benchmark Dataset for Fine-grained Object Recognition in High-Resolution Remote Sensing Imagery\n* [arxiv papr](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.05569)\n* Download at gaofen-challenge.com\n* [2020Gaofen](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002F2020Gaofen) -> 2020 Gaofen Challenge data, baselines, and metrics\n\n## DOTA object detection dataset\nA Large-Scale Benchmark and Challenges for Object Detection in Aerial Images. Segmentation annotations available in iSAID dataset\n* https:\u002F\u002Fcaptain-whu.github.io\u002FDOTA\u002Findex.html\n* [DOTA_devkit](https:\u002F\u002Fgithub.com\u002FCAPTAIN-WHU\u002FDOTA_devkit) for loading dataset\n* [Arxiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1711.10398)\n* [Pretrained models in mmrotate](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmrotate)\n* [DOTA2VOCtools](https:\u002F\u002Fgithub.com\u002FComplicateddd\u002FDOTA2VOCtools) -> dataset split and transform to voc format\n* [dotatron](https:\u002F\u002Fgithub.com\u002Fnaivelogic\u002Fdotatron) -> 2021 Learning to Understand Aerial Images Challenge on DOTA dataset\n\n## iSAID instance segmentation dataset\nA Large-scale Dataset for Instance Segmentation in Aerial Images\n* https:\u002F\u002Fcaptain-whu.github.io\u002FiSAID\u002Fdataset.html\n* Uses images from the DOTA dataset\n\n## HRSC RGB ship object detection dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fguofeng\u002Fhrsc2016\n* [Pretrained models in mmrotate](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmrotate)\n* [Rotation-RetinaNet-PyTorch](https:\u002F\u002Fgithub.com\u002FHsLOL\u002FRotation-RetinaNet-PyTorch)\n\n## SAR Ship Detection Dataset (SSDD)\n* https:\u002F\u002Fgithub.com\u002FTianwenZhang0825\u002FOfficial-SSDD\n* [Rotation-RetinaNet-PyTorch](https:\u002F\u002Fgithub.com\u002FHsLOL\u002FRotation-RetinaNet-PyTorch)\n\n## High-Resolution SAR Rotation Ship Detection Dataset (SRSDD)\n* [Github](https:\u002F\u002Fgithub.com\u002FHeuristicLU\u002FSRSDD-V1.0)\n* [A Lightweight Model for Ship Detection and Recognition in Complex-Scene SAR Images](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F14\u002F23\u002F6053)\n\n## LEVIR ship dataset\nA dataset for tiny ship detection under medium-resolution remote sensing images. Annotations in bounding box format\n* [LEVIR-Ship](https:\u002F\u002Fgithub.com\u002FWindVChen\u002FLEVIR-Ship)\n\u003C!-- markdown-link-check-disable -->\n* Hosted on [Nucleus](https:\u002F\u002Fdashboard.scale.com\u002Fnucleus\u002Fds_cbsghny30nf00b1x3w7g?utm_source=open_dataset&utm_medium=github&utm_campaign=levir_ships)\n\u003C!-- markdown-link-check-enable -->\n\n## SAR Aircraft Detection Dataset\n2966 non-overlapped 224×224 slices are collected with 7835 aircraft targets\n* https:\u002F\u002Fgithub.com\u002Fhust-rslab\u002FSAR-aircraft-data\n\n## xView1: Objects in context for overhead imagery\nA fine-grained object detection dataset with 60 object classes along an ontology of 8 class types. Over 1,000,000 objects across over 1,400 km^2 of 0.3m resolution imagery. Annotations in bounding box format\n* [Official website](http:\u002F\u002Fxviewdataset.org\u002F)\n* [arXiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.07856).\n* [paperswithcode](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fxview)\n* [Satellite_Imagery_Detection_YOLOV7](https:\u002F\u002Fgithub.com\u002FRadhika-Keni\u002FSatellite_Imagery_Detection_YOLOV7) -> YOLOV7 applied to xView1\n\n## xView2: xBD building damage assessment\nAnnotated high-resolution satellite imagery for building damage assessment, precise segmentation masks and damage labels on a four-level spectrum, 0.3m resolution imagery\n* [Official website](https:\u002F\u002Fxview2.org\u002F)\n* [arXiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1911.09296)\n* [paperswithcode](https:\u002F\u002Fpaperswithcode.com\u002Fpaper\u002Fxbd-a-dataset-for-assessing-building-damage)\n* [xView2_baseline](https:\u002F\u002Fgithub.com\u002FDIUx-xView\u002FxView2_baseline) -> baseline solution in tensorflow\n* [metadamagenet](https:\u002F\u002Fgithub.com\u002Fnimaafshar\u002Fmetadamagenet) -> pytorch solution\n* [U-Net models from michal2409](https:\u002F\u002Fgithub.com\u002Fmichal2409\u002FxView2)\n* [DAHiTra](https:\u002F\u002Fgithub.com\u002Fnka77\u002FDAHiTra) -> code for 2022 [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2208.02205): Large-scale Building Damage Assessment using a Novel Hierarchical Transformer Architecture on Satellite Images. Uses xView2 xBD dataset\n* [Damage assessment using Amazon SageMaker geospatial capabilities and custom SageMaker models](https:\u002F\u002Faws.amazon.com\u002Fblogs\u002Fmachine-learning\u002Fdamage-assessment-using-amazon-sagemaker-geospatial-capabilities-and-custom-sagemaker-models\u002F)\n* [Xview2_Strong_Baseline](https:\u002F\u002Fgithub.com\u002FPaulBorneP\u002FXview2_Strong_Baseline) -> a simple implementation of a strong baseline\n\n## xView3: Detecting dark vessels in SAR\nDetecting dark vessels engaged in illegal, unreported, and unregulated (IUU) fishing activities on synthetic aperture radar (SAR) imagery. With human and algorithm annotated instances of vessels and fixed infrastructure across 43,200,000 km^2 of Sentinel-1 imagery, this multi-modal dataset enables algorithms to detect and classify dark vessels\n* [Official website](https:\u002F\u002Fiuu.xview.us\u002F)\n* [arXiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2206.00897)\n* [Github](https:\u002F\u002Fgithub.com\u002FDIUx-xView) -> all reference code, dataset processing utilities, and winning model codes + weights\n* [paperswithcode](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fxview3-sar)\n* [xview3_ship_detection](https:\u002F\u002Fgithub.com\u002Fnaivelogic\u002Fxview3_ship_detection)\n\n## Vehicle Detection in Aerial Imagery (VEDAI)\nVehicle Detection in Aerial Imagery. Bounding box annotations\n* https:\u002F\u002Fdownloads.greyc.fr\u002Fvedai\u002F\n* [pytorch-vedai](https:\u002F\u002Fgithub.com\u002FMichelHalmes\u002Fpytorch-vedai)\n\n## Cars Overhead With Context (COWC)\nLarge set of annotated cars from overhead. Established baseline for object detection and counting tasks. Annotations in bounding box format\n* http:\u002F\u002Fgdo152.ucllnl.org\u002Fcowc\u002F\n* https:\u002F\u002Fgithub.com\u002FLLNL\u002Fcowc\n* [Detecting cars from aerial imagery for the NATO Innovation Challenge](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241110114250\u002Fhttps:\u002F\u002Farthurdouillard.com\u002Fpost\u002Fnato-challenge\u002F)\n* [LINZ and UGRC](https:\u002F\u002Fgithub.com\u002Fhumansensinglab\u002FAGenDA\u002Ftree\u002Fmain\u002FData)\n\n## AI-TOD & AI-TOD-v2 - tiny object detection\nThe mean size of objects in AI-TOD is about 12.8 pixels, which is much smaller than other datasets. Annotations in bounding box format. V2 is a meticulous relabelling of the v1 dataset\n* https:\u002F\u002Fgithub.com\u002Fjwwangchn\u002FAI-TOD\n* https:\u002F\u002Fchasel-tsui.github.io\u002FAI-TOD-v2\u002F\n* [NWD](https:\u002F\u002Fgithub.com\u002Fjwwangchn\u002FNWD) -> code for 2021 [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2110.13389): A Normalized Gaussian Wasserstein Distance for Tiny Object Detection. Uses AI-TOD dataset\n* [ORFENet](https:\u002F\u002Fgithub.com\u002Fdyl96\u002FORFENet) -> Tiny Object Detection in Remote Sensing Images Based on Object Reconstruction and Multiple Receptive Field Adaptive Feature Enhancement. Uses LEVIR-ship & AI-TOD-v2\n\n## RarePlanes\n* [RarePlanes](https:\u002F\u002Fregistry.opendata.aws\u002Frareplanes\u002F) -> incorporates both real and synthetically generated satellite imagery including aircraft. Read the [arxiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2006.02963) and checkout [this repo](https:\u002F\u002Fgithub.com\u002Fjdc08161063\u002FRarePlanes). Note the dataset is available through the AWS Open-Data Program for free download\n* [Understanding the RarePlanes Dataset and Building an Aircraft Detection Model](https:\u002F\u002Fencord.com\u002Fblog\u002Frareplane-dataset-aircraft-detection-model\u002F) -> blog post\n* Read [this article from NVIDIA](https:\u002F\u002Fdeveloper.nvidia.com\u002Fblog\u002Fpreparing-models-for-object-detection-with-real-and-synthetic-data-and-tao-toolkit\u002F) which discusses fine tuning a model pre-trained on synthetic data (Rareplanes) with 10% real data, then pruning the model to reduce its size, before quantizing the model to improve inference speed\n* [yoltv4](https:\u002F\u002Fgithub.com\u002Favanetten\u002Fyoltv4) includes examples on the [RarePlanes dataset](https:\u002F\u002Fregistry.opendata.aws\u002Frareplanes\u002F)\n* [rareplanes-yolov5](https:\u002F\u002Fgithub.com\u002Fjeffaudi\u002Frareplanes-yolov5) -> using YOLOv5 and the RarePlanes dataset to detect and classify sub-characteristics of aircraft, with [article](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Fdetecting-aircrafts-on-airbus-pleiades-imagery-with-yolov5-5f3d464b75ad)\n\n## Counting from Sky\nA Large-scale Dataset for Remote Sensing Object Counting and A Benchmark Method\n* https:\u002F\u002Fgithub.com\u002Fgaoguangshuai\u002FCounting-from-Sky-A-Large-scale-Dataset-for-Remote-Sensing-Object-Counting-and-A-Benchmark-Method\n\n## AIRS (Aerial Imagery for Roof Segmentation)\nPublic dataset for roof segmentation from very-high-resolution aerial imagery (7.5cm). Covers almost the full area of Christchurch, the largest city in the South Island of New Zealand.\n* [On Kaggle](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fatilol\u002Faerialimageryforroofsegmentation)\n* [Rooftop-Instance-Segmentation](https:\u002F\u002Fgithub.com\u002FMasterSkepticista\u002FRooftop-Instance-Segmentation) -> VGG-16, Instance Segmentation, uses the Airs dataset\n\n## Inria building\u002Fnot building segmentation dataset\nRGB GeoTIFF at spatial resolution of 0.3 m. Data covering Austin, Chicago, Kitsap County, Western & Easter Tyrol, Innsbruck, San Francisco & Vienna\n* https:\u002F\u002Fproject.inria.fr\u002Faerialimagelabeling\u002Fcontest\u002F\n* [SemSegBuildings](https:\u002F\u002Fgithub.com\u002FSharpestProjects\u002FSemSegBuildings) -> Project using fast.ai framework for semantic segmentation on Inria building segmentation dataset\n* [UNet_keras_for_RSimage](https:\u002F\u002Fgithub.com\u002Floveswine\u002FUNet_keras_for_RSimage) -> keras code for binary semantic segmentation\n\n## AICrowd Mapping Challenge: building segmentation dataset\n300x300 pixel RGB images with annotations in COCO format. Imagery appears to be global but with significant fraction from North America\n* Dataset release as part of the [mapping-challenge](https:\u002F\u002Fwww.aicrowd.com\u002Fchallenges\u002Fmapping-challenge)\n* Winning solution published by neptune.ai [here](https:\u002F\u002Fgithub.com\u002Fneptune-ai\u002Fopen-solution-mapping-challenge), achieved precision 0.943 and recall 0.954 using Unet with Resnet.\n* [mappingchallenge](https:\u002F\u002Fgithub.com\u002Fkrishanr\u002Fmappingchallenge) -> YOLOv5 applied to the AICrowd Mapping Challenge dataset\n\n## BONAI - building footprint dataset\nBONAI (Buildings in Off-Nadir Aerial Images) is a dataset for building footprint extraction (BFE) in off-nadir aerial images\n* https:\u002F\u002Fgithub.com\u002Fjwwangchn\u002FBONAI\n\n## LEVIR-CD building change detection dataset\n* https:\u002F\u002Fjustchenhao.github.io\u002FLEVIR\u002F\n* [FCCDN_pytorch](https:\u002F\u002Fgithub.com\u002Fchenpan0615\u002FFCCDN_pytorch) -> pytorch implemention of FCCDN for change detection task\n* [RSICC](https:\u002F\u002Fgithub.com\u002FChen-Yang-Liu\u002FRSICC) -> the Remote Sensing Image Change Captioning dataset uses LEVIR-CD imagery\n\n## Onera (OSCD) Sentinel-2 change detection dataset\nIt comprises 24 pairs of multispectral images taken from the Sentinel-2 satellites between 2015 and 2018. \n* [Onera Satellite Change Detection Dataset](https:\u002F\u002Fieee-dataport.org\u002Fopen-access\u002Foscd-onera-satellite-change-detection) comprises 24 pairs of multispectral images taken from the Sentinel-2 satellites between 2015 and 2018\n* [Website](https:\u002F\u002Frcdaudt.github.io\u002Foscd\u002F)\n* [change_detection_onera_baselines](https:\u002F\u002Fgithub.com\u002Fprevitus\u002Fchange_detection_onera_baselines) -> Siamese version of U-Net baseline model\n* [Urban Change Detection for Multispectral Earth Observation Using Convolutional Neural Networks](https:\u002F\u002Fgithub.com\u002Frcdaudt\u002Fpatch_based_change_detection) -> with [paper](https:\u002F\u002Fieeexplore.ieee.org\u002Fabstract\u002Fdocument\u002F8518015)\n* [DS_UNet](https:\u002F\u002Fgithub.com\u002FSebastianHafner\u002FDS_UNet) -> code for 2021 paper: Sentinel-1 and Sentinel-2 Data Fusion for Urban Change Detection using a Dual Stream U-Net, uses Onera Satellite Change Detection dataset\n* [ChangeDetection_wOnera](https:\u002F\u002Fgithub.com\u002Ftonydp03\u002FChangeDetection_wOnera)\n* [OSCD + additional Dates](https:\u002F\u002Fgithub.com\u002Fgranularai\u002Ffabric) -> extended with three different dates\n* [MSOSCD](https:\u002F\u002Fgithub.com\u002FLihy256\u002FMSCDUnet) -> change detection datasets containing VHR, multispectral (Sentinel-2) and SAR (Sentinel-1)\n\n## SECOND - semantic change detection\n* https:\u002F\u002Fcaptain-whu.github.io\u002FSCD\u002F\n* Change detection at the pixel level\n\n## Amazon and Atlantic Forest dataset\nFor semantic segmentation with Sentinel 2\n* [Amazon and Atlantic Forest image datasets for semantic segmentation](https:\u002F\u002Fzenodo.org\u002Frecord\u002F4498086#.Y6LPLuzP1hE)\n* [attention-mechanism-unet](https:\u002F\u002Fgithub.com\u002Fdavej23\u002Fattention-mechanism-unet) -> An attention-based U-Net for detecting deforestation within satellite sensor imagery\n* [TransUNetplus2](https:\u002F\u002Fgithub.com\u002Faj1365\u002FTransUNetplus2) -> Rethinking attention gated TransU-Net for deforestation mapping\n\n## Functional Map of the World ( fMoW)\n* https:\u002F\u002Fgithub.com\u002FfMoW\u002Fdataset\n* RGB & multispectral variants\n* High resolution, chip classification dataset\n* Purpose: predicting the functional purpose of buildings and land use from temporal sequences of satellite images and a rich set of metadata features\n\n## HRSCD change detection\n* https:\u002F\u002Frcdaudt.github.io\u002Fhrscd\u002F\n* 291 coregistered image pairs of high resolution RGB aerial images\n* Pixel-level change and land cover annotations are provided\n\n## MiniFrance-DFC22 - semi-supervised semantic segmentation\n* The [MiniFrance-DFC22 (MF-DFC22) dataset](https:\u002F\u002Fieee-dataport.org\u002Fcompetitions\u002Fdata-fusion-contest-2022-dfc2022) extends and modifies the [MiniFrance dataset](https:\u002F\u002Fieee-dataport.org\u002Fopen-access\u002Fminifrance) for training semi-supervised semantic segmentation models for land use\u002Fland cover mapping\n* [dfc2022-baseline](https:\u002F\u002Fgithub.com\u002Fisaaccorley\u002Fdfc2022-baseline) -> baseline solution to the 2022 IEEE GRSS Data Fusion Contest (DFC2022) using TorchGeo, PyTorch Lightning, and Segmentation Models PyTorch to train a U-Net with a ResNet-18 backbone and a loss function of Focal + Dice loss to perform semantic segmentation on the DFC2022 dataset\n* https:\u002F\u002Fgithub.com\u002Fmveo\u002Fmveo-challenge\n\n## FLAIR\nSemantic segmentation and domain adaptation challenge proposed by the French National Institute of Geographical and Forest Information (IGN). Uses a dataset composed of over 70,000 aerial imagery patches with pixel-based annotations and 50,000 Sentinel-2 satellite acquisitions.\n* [Challenge on codalab](https:\u002F\u002Fcodalab.lisn.upsaclay.fr\u002Fcompetitions\u002F13447)\n* [FLAIR-2 github](https:\u002F\u002Fgithub.com\u002FIGNF\u002FFLAIR-2)\n* [flair-2 8th place solution](https:\u002F\u002Fgithub.com\u002Fassociation-rosia\u002Fflair-2)\n* [IGNF HuggingFace](https:\u002F\u002Fhuggingface.co\u002FIGNF)\n\n## ISPRS\nSemantic segmentation dataset. 38 patches of 6000x6000 pixels, each consisting of a true orthophoto (TOP) extracted from a larger TOP mosaic, and a DSM. Resolution 5 cm\n* https:\u002F\u002Fwww.isprs.org\u002Fresources\u002Fdatasets\u002Fbenchmarks\u002FUrbanSemLab\u002F2d-sem-label-potsdam.aspx\n\n## SpaceNet\nSpaceNet is a series of competitions with datasets and utilities provided. The challenges covered are: (1 & 2) building segmentation, (3) road segmentation, (4) off-nadir buildings, (5) road network extraction, (6) multi-senor mapping, (7) multi-temporal urban change, (8) Flood Detection Challenge Using Multiclass Segmentation\n* [spacenet.ai](https:\u002F\u002Fspacenet.ai\u002F) is an online hub for data, challenges, algorithms, and tools\n* [The SpaceNet 7 Multi-Temporal Urban Development Challenge: Dataset Release](https:\u002F\u002Fmedium.com\u002Fthe-downlinq\u002Fthe-spacenet-7-multi-temporal-urban-development-challenge-dataset-release-9e6e5f65c8d5)\n* [spacenet-three-topcoder](https:\u002F\u002Fgithub.com\u002Fsnakers4\u002Fspacenet-three-topcoder) solution\n* [official utilities](https:\u002F\u002Fgithub.com\u002FSpaceNetChallenge\u002Futilities) -> Packages intended to assist in the preprocessing of SpaceNet satellite imagery dataset to a format that is consumable by machine learning algorithms\n* [andraugust spacenet-utils](https:\u002F\u002Fgithub.com\u002Fandraugust\u002Fspacenet-utils) -> Display geotiff image with building-polygon overlay & label buildings using kNN on the pixel spectra\n* [Spacenet-Building-Detection](https:\u002F\u002Fgithub.com\u002FIdanC1s2\u002FSpacenet-Building-Detection) -> uses keras and [Spacenet 1 dataset](https:\u002F\u002Fspacenet.ai\u002Fspacenet-buildings-dataset-v1\u002F)\n* [Spacenet 8 winners blog post](https:\u002F\u002Fmedium.com\u002F@SpaceNet_Project\u002Fspacenet-8-a-closer-look-at-the-winning-approaches-75ff4033bf53)\n\n## WorldStrat Dataset\nNearly 10,000 km² of free high-resolution satellite imagery of unique locations which ensure stratified representation of all types of land-use across the world: from agriculture to ice caps, from forests to multiple urbanization densities.\n* https:\u002F\u002Fgithub.com\u002Fworldstrat\u002Fworldstrat\n* [Quick tour of the WorldStrat Dataset](https:\u002F\u002Fmedium.com\u002F@robmarkcole\u002Fquick-tour-of-the-worldstrat-dataset-b2d1c2d435db)\n* Each high-resolution image (1.5 m\u002Fpixel) comes with multiple temporally-matched low-resolution images from the freely accessible lower-resolution Sentinel-2 satellites (10 m\u002Fpixel)\n* Several super-resolution benchmark models trained on it\n\n## Satlas Pretrain\nSatlasPretrain is a large-scale pre-training dataset for tasks that involve understanding satellite images. Regularly-updated satellite data is publicly available for much of the Earth through sources such as Sentinel-2 and NAIP, and can inform numerous applications from tackling illegal deforestation to monitoring marine infrastructure. \n* [Website](https:\u002F\u002Fsatlas-pretrain.allen.ai\u002F)\n* [Code](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fsatlas)\n\n## FLAIR 1 & 2 Segmentation datasets\n* https:\u002F\u002Fignf.github.io\u002FFLAIR\u002F\n* The FLAIR #1 semantic segmentation dataset consists of 77,412 high resolution patches (512x512 at 0.2 m spatial resolution) with 19 semantic classes\n* FLAIR #2 includes an expanded dataset of Sentinel-2 time series for multi-modal semantic segmentation\n\n## Five Billion Pixels segmentation dataset\n* https:\u002F\u002Fx-ytong.github.io\u002Fproject\u002FFive-Billion-Pixels.html\n* 4m Gaofen-2 imagery over China\n* 24 land cover classes\n* Paper and code demonstrating domain adaptation to Sentinel-2 and Planetscope imagery\n* Extends the [GID15 large scale semantic segmentation dataset](https:\u002F\u002Fcaptain-whu.github.io\u002FGID15\u002F)\n* [GID](https:\u002F\u002Fx-ytong.github.io\u002Fproject\u002FGID.html) -> the Gaofen Image Dataset is a large-scale land-cover dataset with Gaofen-2 (GF-2) satellite images\n* [MM-5B Dataset](https:\u002F\u002Fgithub.com\u002FAI-Tianlong\u002FHieraRS) -> Multi-Modal Five-Billion-Pixels is a large-scale, multi-modal, hierarchical Land Cover and Land Use (LCLU) dataset, built upon the Five-Billion-Pixels foundation.\n\n## RF100 object detection benchmark\nRF100 is compiled from 100 real world datasets that straddle a range of domains. The aim is that performance evaluation on this dataset will enable a more nuanced guide of how a model will perform in different domains. Contains 10k aerial images\n* https:\u002F\u002Fwww.rf100.org\u002F\n* https:\u002F\u002Fgithub.com\u002Froboflow-ai\u002Froboflow-100-benchmark\n\n## SATIN (SATellite ImageNet)\nSATIN is a multi-task remote sensing classification metadataset consisting of 27 datasets grouped into 6 tasks. The imagery spans 5 orders of magnitude of resolution, over 250 distinct class labels, and many field of view sizes. The overall SATIN benchmark, as well as each of the 27 constituent datasets, are released via HuggingFace. A public leaderboard is provided to guide and track the progress of vision-language models on SATIN. \n* [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.11619)\n* [Website](https:\u002F\u002Fsatinbenchmark.github.io\u002F)\n* [Data](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fjonathan-roberts1\u002FSATIN)\n\n## SODA-A rotated bounding boxes\n* https:\u002F\u002Fshaunyuan22.github.io\u002FSODA\u002F\n* SODA-A comprises 2513 high-resolution images of aerial scenes, which has 872069 instances annotated with oriented rectangle box annotations over 9 classes\n* https:\u002F\u002Fgithub.com\u002Fshaunyuan22\u002FCFINet\n\n## EarthView from Satellogic\n* https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fsatellogic\u002FEarthView\n* Dataset for foundational models, with Sentinel 1 & 2 and 1m RGB\n\n## Microsoft datasets\n* [US Building Footprints](https:\u002F\u002Fgithub.com\u002FMicrosoft\u002FUSBuildingFootprints) -> building footprints in all 50 US states, GeoJSON format, generated using semantic segmentation. Also [Australia](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FAustraliaBuildingFootprints), [Canadian](https:\u002F\u002Fgithub.com\u002FMicrosoft\u002FCanadianBuildingFootprints), [Uganda-Tanzania](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FUganda-Tanzania-Building-Footprints), [Kenya-Nigeria](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FKenyaNigeriaBuildingFootprints) and [GlobalMLBuildingFootprints](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FGlobalMLBuildingFootprints) are available. Checkout [RasterizingBuildingFootprints](https:\u002F\u002Fgithub.com\u002Fmehdiheris\u002FRasterizingBuildingFootprints) to convert vector shapefiles to raster layers\n* [Microsoft Planetary Computer](https:\u002F\u002Fplanetarycomputer.microsoft.com\u002F) is a Dask-Gateway enabled JupyterHub deployment focused on supporting scalable geospatial analysis, [source repo](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fplanetary-computer-hub)\n* [landcover-orinoquia](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Flandcover-orinoquia) -> Land cover mapping of the Orinoquía region in Colombia, in collaboration with Wildlife Conservation Society Colombia. An #AIforEarth project\n* [RoadDetections dataset by Microsoft](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRoadDetections)\n\n## Google datasets\n* [open-buildings](https:\u002F\u002Fsites.research.google\u002Fopen-buildings\u002F) -> A dataset of building footprints to support social good applications covering 64% of the African continent. Read [Mapping Africa’s Buildings with Satellite Imagery](https:\u002F\u002Fai.googleblog.com\u002F2021\u002F07\u002Fmapping-africas-buildings-with.html)\n\n## Google Earth Engine (GEE)\nSince there is a whole community around GEE I will not reproduce it here but list very select references. Get started at https:\u002F\u002Fdevelopers.google.com\u002Fearth-engine\u002F\n* Various imagery and climate datasets, including Landsat & Sentinel imagery\n* Supports large scale processing with classical algorithms, e.g. clustering for land use. For deep learning, you export datasets from GEE as tfrecords, train on your preferred GPU platform, then upload inference results back to GEE\n* [awesome-google-earth-engine](https:\u002F\u002Fgithub.com\u002Fgee-community\u002Fawesome-google-earth-engine)\n* [Awesome-GEE](https:\u002F\u002Fgithub.com\u002Fgiswqs\u002FAwesome-GEE)\n* [awesome-earth-engine-apps](https:\u002F\u002Fgithub.com\u002Fphilippgaertner\u002Fawesome-earth-engine-apps)\n* [How to Use Google Earth Engine and Python API to Export Images to Roboflow](https:\u002F\u002Fblog.roboflow.com\u002Fhow-to-use-google-earth-engine-with-roboflow\u002F) -> to acquire training data\n* [ee-fastapi](https:\u002F\u002Fgithub.com\u002Fcsaybar\u002Fee-fastapi) is a simple FastAPI web application for performing flood detection using Google Earth Engine in the backend.\n* [How to Download High-Resolution Satellite Data for Anywhere on Earth](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241003151941\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fhow-to-download-high-resolution-satellite-data-for-anywhere-on-earth-5e6dddee2803)\n* [wxee](https:\u002F\u002Fgithub.com\u002Faazuspan\u002Fwxee) -> Export data from GEE to xarray using wxee then train with pytorch or tensorflow models. Useful since GEE only suports tfrecord export natively\n\n## Image captioning datasets\n* [RSICD](https:\u002F\u002Fgithub.com\u002F201528014227051\u002FRSICD_optimal) -> 10921 images with five sentences descriptions per image. Used in  [Fine tuning CLIP with Remote Sensing (Satellite) images and captions](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Ffine-tune-clip-rsicd), models at [this repo](https:\u002F\u002Fgithub.com\u002Farampacha\u002FCLIP-rsicd)\n* [RSICC](https:\u002F\u002Fgithub.com\u002FChen-Yang-Liu\u002FRSICC) -> the Remote Sensing Image Change Captioning dataset contains 10077 pairs of bi-temporal remote sensing images and 50385 sentences describing the differences between images. Uses LEVIR-CD imagery\n* [ChatEarthNet](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FChatEarthNet) -> A Global-Scale Image-Text Dataset Empowering Vision-Language Geo-Foundation Models, utilizes Sentinel-2 data with captions generated by ChatGPT\n\n## Weather Datasets\n* NASA (make request and emailed when ready) -> https:\u002F\u002Fsearch.earthdata.nasa.gov\n* NOAA (requires BigQuery) -> https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fnoaa\u002Fgoes16\u002Fhome\n* Time series weather data for several US cities -> https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fselfishgene\u002Fhistorical-hourly-weather-data\n* [DeepWeather](https:\u002F\u002Fgithub.com\u002Fadamhazimeh\u002FDeepWeather) -> improve weather forecasting accuracy by analyzing satellite images\n\n## Cloud datasets\n* [Planet-CR](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FPlanet-CR) -> A Multi-Modal and Multi-Resolution Dataset for Cloud Removal in High Resolution Optical Remote Sensing Imagery, 3m resolution, with [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2301.03432)\n* [The Azavea Cloud Dataset](https:\u002F\u002Fwww.azavea.com\u002Fblog\u002F2021\u002F08\u002F02\u002Fthe-azavea-cloud-dataset\u002F) which is used to train this [cloud-model](https:\u002F\u002Fgithub.com\u002Fazavea\u002Fcloud-model)\n* [Sentinel-2 Cloud Cover Segmentation Dataset](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20230606184945\u002Fhttps:\u002F\u002Fmlhub.earth\u002Fdata\u002Fref_cloud_cover_detection_challenge_v1) on Radiant mlhub\n* [cloudsen12](https:\u002F\u002Fcloudsen12.github.io\u002F) -> see [video](https:\u002F\u002Fyoutu.be\u002FGhQwnVhJ1wo)\n* [WHUS2-CD+](https:\u002F\u002Fzenodo.org\u002Frecord\u002F5511793) -> 36 manually labeled cloud masks at 10m resolution and corresponding Sentinel-2 images evenly distributed over China mainland, and used to train CD-FM3SF [cloud-model](https:\u002F\u002Fgithub.com\u002FNeooolee\u002FWHUS2-CD)\n* [HRC_WHU](https:\u002F\u002Fgithub.com\u002Fdr-lizhiwei\u002FHRC_WHU) -> High-Resolution Cloud Detection Dataset comprising 150 RGB images and a resolution varying from 0.5 to 15 m in different global regions\n* [AIR-CD](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FAIR-CD) -> a challenging cloud detection data set called AIR-CD, with higher spatial resolution and more representative landcover types\n* [Landsat 8 Cloud Cover Assessment Validation Data](https:\u002F\u002Flandsat.usgs.gov\u002Flandsat-8-cloud-cover-assessment-validation-data)\n\n## Forest datasets\n* [OpenForest](https:\u002F\u002Fgithub.com\u002FRolnickLab\u002FOpenForest) -> A catalogue of open access forest datasets\n* [awesome-forests](https:\u002F\u002Fgithub.com\u002Fblutjens\u002Fawesome-forests) -> A curated list of ground-truth forest datasets for the machine learning and forestry community\n* [ReforesTree](https:\u002F\u002Fgithub.com\u002Fgyrrei\u002FReforesTree) -> A dataset for estimating tropical forest biomass based on drone and field data\n* [yosemite-tree-dataset](https:\u002F\u002Fgithub.com\u002Fnightonion\u002Fyosemite-tree-dataset) -> a benchmark dataset for tree counting from aerial images\n* [Amazon Rainforest dataset for semantic segmentation](https:\u002F\u002Fzenodo.org\u002Frecord\u002F3233081#.Y6LPLOzP1hE) -> Sentinel 2 images. Used in the paper 'An attention-based U-Net for detecting deforestation within satellite sensor imagery'\n* [Amazon and Atlantic Forest image datasets for semantic segmentation](https:\u002F\u002Fzenodo.org\u002Frecord\u002F4498086#.Y6LPLuzP1hE) -> Sentinel 2 images. Used in paper 'An attention-based U-Net for detecting deforestation within satellite sensor imagery'\n* [TreeSatAI](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6780578) -> Sentinel-1, Sentinel-2\n* [PureForest](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FIGNF\u002FPureForest) -> VHR RGB + Near-Infrared & lidar, each patch represents a monospecific forest\n\n## Geospatial datasets\n* [Resource Watch](https:\u002F\u002Fresourcewatch.org\u002Fdata\u002Fexplore) provides a wide range of geospatial datasets and a UI to visualise them\n\n## Time series & change detection datasets\n* [BreizhCrops](https:\u002F\u002Fgithub.com\u002Fdl4sits\u002FBreizhCrops) -> A Time Series Dataset for Crop Type Mapping\n* The SeCo dataset contains image patches from Sentinel-2 tiles captured at different timestamps at each geographical location. [Download SeCo here](https:\u002F\u002Fgithub.com\u002FElementAI\u002Fseasonal-contrast)\n* [SYSU-CD](https:\u002F\u002Fgithub.com\u002Fliumency\u002FSYSU-CD) -> The dataset contains 20000 pairs of 0.5-m aerial images of size 256×256 taken between the years 2007 and 2014 in Hong Kong\n\n### DEM (digital elevation maps)\n* Shuttle Radar Topography Mission, search online at usgs.gov\n* Copernicus Digital Elevation Model (DEM) on S3, represents the surface of the Earth including buildings, infrastructure and vegetation. Data is provided as Cloud Optimized GeoTIFFs. [link](https:\u002F\u002Fregistry.opendata.aws\u002Fcopernicus-dem\u002F)\n* [Awesome-DEM](https:\u002F\u002Fgithub.com\u002FDahnJ\u002FAwesome-DEM)\n\n## UAV & Drone datasets\n* Many on https:\u002F\u002Fwww.visualdata.io\n* [AU-AIR dataset](https:\u002F\u002Farxiv.org\u002Fabs\u002F2104.06781) -> a multi-modal UAV dataset for object detection.\n* [ERA](https:\u002F\u002Flcmou.github.io\u002FERA_Dataset\u002F) ->  A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos.\n* [Aerial Maritime Drone Dataset](https:\u002F\u002Fpublic.roboflow.ai\u002Fobject-detection\u002Faerial-maritime) -> bounding boxes\n* [RetinaNet for pedestrian detection](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241002023333\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fpedestrian-detection-in-aerial-images-using-retinanet-9053e8a72c6) -> bounding boxes\n* [BIRDSAI: A Dataset for Detection and Tracking in Aerial Thermal Infrared Videos](https:\u002F\u002Fgithub.com\u002Fexb7900\u002FBIRDSAI) -> Thermal IR videos of humans and animals\n* [ERA: A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos](https:\u002F\u002Flcmou.github.io\u002FERA_Dataset\u002F)\n* [DroneVehicle](https:\u002F\u002Fgithub.com\u002FVisDrone\u002FDroneVehicle) -> Drone-based RGB-Infrared Cross-Modality Vehicle Detection via Uncertainty-Aware Learning. Annotations are rotated bounding boxes. With [Github repo](https:\u002F\u002Fgithub.com\u002FSunYM2020\u002FUA-CMDet)\n* [UAVOD10](https:\u002F\u002Fgithub.com\u002Fweihancug\u002F10-category-UAV-small-weak-object-detection-dataset-UAVOD10) -> 10 class of objects at 15 cm resolution. Classes are; building, ship, vehicle, prefabricated house, well, cable tower, pool, landslide, cultivation mesh cage, and quarry. Bounding boxes\n* [Busy-parking-lot-dataset---vehicle-detection-in-UAV-video](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FBusy-parking-lot-dataset---vehicle-detection-in-UAV-video) -> Vehicle instance segmentation. Unsure format of annotations, possible Matlab specific\n* [dd-ml-segmentation-benchmark](https:\u002F\u002Fgithub.com\u002Fdronedeploy\u002Fdd-ml-segmentation-benchmark) -> DroneDeploy Machine Learning Segmentation Benchmark\n* [SeaDronesSee](https:\u002F\u002Fgithub.com\u002FBen93kie\u002FSeaDronesSee) -> Vision Benchmark for Maritime Search and Rescue. Bounding box object detection, single-object tracking and multi-object tracking annotations\n* [aeroscapes](https:\u002F\u002Fgithub.com\u002Fishann\u002Faeroscapes) -> semantic segmentation benchmark comprises of images captured using a commercial drone from an altitude range of 5 to 50 metres.\n* [ALTO](https:\u002F\u002Fgithub.com\u002FMetaSLAM\u002FALTO) -> Aerial-view Large-scale Terrain-Oriented. For deep learning based UAV visual place recognition and localization tasks.\n* [HIT-UAV-Infrared-Thermal-Dataset](https:\u002F\u002Fgithub.com\u002Fsuojiashun\u002FHIT-UAV-Infrared-Thermal-Dataset) -> A High-altitude Infrared Thermal Object Detection Dataset for Unmanned Aerial Vehicles\n* [caltech-aerial-rgbt-dataset](https:\u002F\u002Fgithub.com\u002Faerorobotics\u002Fcaltech-aerial-rgbt-dataset) -> synchronized RGB, thermal, GPS, and IMU data\n* [Leafy Spurge Dataset](https:\u002F\u002Fleafy-spurge-dataset.github.io\u002F) -> Real-world Weed Classification Within Aerial Drone Imagery\n* [Agriculture-Vision 2021 Dataset](https:\u002F\u002Fwww.agriculture-vision.com\u002Fagriculture-vision-2021\u002Fdataset-2021)\n* [UAV-HSI-Crop-Dataset](https:\u002F\u002Fgithub.com\u002FMrSuperNiu\u002FUAV-HSI-Crop-Dataset) -> dataset for \"HSI-TransUNet: A Transformer based semantic segmentation model for crop mapping from UAV hyperspectral imagery\"\n* [UAVVaste](https:\u002F\u002Fgithub.com\u002FPUTvision\u002FUAVVaste) -> COCO-like dataset and effective waste detection in aerial images\n* [BSB-Aerial-Dataset](https:\u002F\u002Fgithub.com\u002Fosmarluiz\u002FBSB-Aerial-Dataset) -> A panoptic segmentation dataset of aerial imagery from Brasilia, Brazil.\n\n## Other datasets\n\n### Object Detection & Classification\n* [RSOD-Dataset](https:\u002F\u002Fgithub.com\u002FRSIA-LIESMARS-WHU\u002FRSOD-Dataset-) -> dataset for object detection in PASCAL VOC format. Aircraft, playgrounds, overpasses & oiltanks\n* [VHR-10_dataset_coco](https:\u002F\u002Fgithub.com\u002Fchaozhong2010\u002FVHR-10_dataset_coco) -> Object detection and instance segmentation dataset based on NWPU VHR-10 dataset. RGB & SAR\n* [MAR20](https:\u002F\u002Fgcheng-nwpu.github.io\u002F) -> Military Aircraft Recognition dataset\n* [RSAPS-ASD](https:\u002F\u002Fgithub.com\u002FSKLSEIIT\u002FRSAPS-ASD) -> A Remote Sensing Airport Panoptic Segmentation with Airplane States dataset, constructed in \"Airplane State Discrimination from Single-temporal High-Resolution Remote Sensing Images\"\n* [Sewage-Treatment-Plant-Dataset](https:\u002F\u002Fgithub.com\u002Fpeijinwang\u002FSewage-Treatment-Plant-Dataset) -> object detection\n* [TGRS-HRRSD-Dataset](https:\u002F\u002Fgithub.com\u002FCrazyStoneonRoad\u002FTGRS-HRRSD-Dataset) -> High Resolution Remote Sensing Detection (HRRSD)\n* [OGST](https:\u002F\u002Fdata.mendeley.com\u002Fdatasets\u002Fbkxj8z84m9\u002F3) -> Oil and Gas Tank Dataset\n* [SearchAndRescueNet](https:\u002F\u002Fgithub.com\u002Fmichaelthoreau\u002FSearchAndRescueNet) -> Satellite Imagery for Search And Rescue Dataset, with example Faster R-CNN model\n* [UBC-dataset](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FUBC-dataset) -> a dataset for building detection and classification from very high-resolution satellite imagery with the focus on object-level interpretation of individual buildings\n* [Building_Dataset](https:\u002F\u002Fgithub.com\u002FQiaoWenfan\u002FBuilding_Dataset) -> High-speed Rail Line Building Dataset Display\n* [RID](https:\u002F\u002Fgithub.com\u002FTUMFTM\u002FRID) -> Roof Information Dataset for CV-Based Photovoltaic Potential Assessment. With [paper](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F14\u002F10\u002F2299)\n* [APKLOT](https:\u002F\u002Fgithub.com\u002Flangheran\u002FAPKLOT) -> A dataset for aerial parking block segmentation\n* [SAR-ACD](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FSAR-ACD) -> SAR-ACD consists of 4322 aircraft clips with 6 civil aircraft categories and 14 other aircraft categories\n* [SODA](https:\u002F\u002Fshaunyuan22.github.io\u002FSODA\u002F) -> A large-scale Small Object Detection dataset. SODA-A comprises 2510 high-resolution images of aerial scenes, which has 800203 instances annotated with oriented rectangle box annotations over 9 classes.\n* [urban-tree-detection-data](https:\u002F\u002Fgithub.com\u002Fjonathanventura\u002Furban-tree-detection-data) -> Dataset for training and evaluating tree detectors in urban environments with aerial imagery\n* [Satellite imagery datasets containing ships](https:\u002F\u002Fgithub.com\u002FNaLiu613\u002FSatellite-Imagery-Datasets-Containing-Ships) -> A list of radar and optical satellite datasets for ship detection, classification, semantic segmentation and instance segmentation tasks\n* [Roofline-Extraction](https:\u002F\u002Fgithub.com\u002Floosgagnet\u002FRoofline-Extraction) -> dataset for paper 'Knowledge-Based 3D Building Reconstruction (3DBR) Using Single Aerial Images and Convolutional Neural Networks (CNNs)'\n* [Building-detection-and-roof-type-recognition](https:\u002F\u002Fgithub.com\u002Floosgagnet\u002FBuilding-detection-and-roof-type-recognition) -> datasets for the paper 'A CNN-Based Approach for Automatic Building Detection and Recognition of Roof Types Using a Single Aerial Image'\n* [OnlyPlanes](https:\u002F\u002Fgithub.com\u002Fnaivelogic\u002FOnlyPlanes) -> Synthetic dataset and pretrained models for Detectron2\n* [SV248S](https:\u002F\u002Fgithub.com\u002Fxdai-dlgvv\u002FSV248S) -> Single Object Tracking Dataset, tracking Vehicle, Large-Vehicle, Ship and Airplane\n* [NWPU-MOC](https:\u002F\u002Fgithub.com\u002Flyongo\u002FNWPU-MOC) -> A Benchmark for Fine-grained Multi-category Object Counting in Aerial Images\n* [Vehicle Perception from Satellite](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.00703) -> a large-scale benchmark for traffic monitoring from satellite\n* [SARDet-100K](https:\u002F\u002Fgithub.com\u002Fzcablii\u002FSARDet_100K) -> Large-Scale Synthetic Aperture Radar (SAR) Object Detection\n* [Urban Vehicle Segmentation Dataset (UV6K)](https:\u002F\u002Fzenodo.org\u002Frecords\u002F8404754)\n* [ShipRSImageNet](https:\u002F\u002Fgithub.com\u002Fzzndream\u002FShipRSImageNet) -> A Large-scale Fine-Grained Dataset for Ship Detection in High-Resolution Optical Remote Sensing Images\n* [VME: A Satellite Imagery Dataset and Benchmark for Detecting Vehicles in the Middle East and Beyond](https:\u002F\u002Fgithub.com\u002Fnalemadi\u002FVME_CDSI_dataset_benchmark)\n* [VHRV: Very High-Resolution Benchmark Dataset for Vessel Detection](https:\u002F\u002Fgithub.com\u002Fbuyukkanber\u002Fvhrv)\n\n### Land Use & Land Cover\n* [land-use-land-cover-datasets](https:\u002F\u002Fgithub.com\u002Fr-wenger\u002Fland-use-land-cover-datasets)\n* [RSD46-WHU](https:\u002F\u002Fgithub.com\u002FRSIA-LIESMARS-WHU\u002FRSD46-WHU) -> 46 scene classes for image classification, free for education, research and commercial use\n* [RSSCN7](https:\u002F\u002Fgithub.com\u002Fpalewithout\u002FRSSCN7) -> Dataset of the article \"Deep Learning Based Feature Selection for Remote Sensing Scene Classification\"\n* [geonrw](https:\u002F\u002Fieee-dataport.org\u002Fopen-access\u002Fgeonrw) -> orthorectified aerial photographs, LiDAR derived digital elevation models and segmentation maps with 10 classes. With [repo](https:\u002F\u002Fgithub.com\u002Fgbaier\u002Fgeonrw)\n* [Attribute-Cooperated-Classification-Datasets](https:\u002F\u002Fgithub.com\u002FCrazyStoneonRoad\u002FAttribute-Cooperated-Classification-Datasets) -> Three datasets based on AID, UCM, and Sydney. For each image, there is a label of scene classification and a label vector of attribute items.\n* [open_earth_map](https:\u002F\u002Fgithub.com\u002Fbao18\u002Fopen_earth_map) -> a benchmark dataset for global high-resolution land cover mapping\n* [Mumbai-Semantic-Segmentation-Dataset](https:\u002F\u002Fgithub.com\u002FGeoAI-Research-Lab\u002FMumbai-Semantic-Segmentation-Dataset)\n* [GAMUS](https:\u002F\u002Fgithub.com\u002FEarthNets\u002FRSI-MMSegmentation) ->  A Geometry-aware Multi-modal Semantic Segmentation Benchmark for Remote Sensing Data\n* [openWUSU](https:\u002F\u002Fgithub.com\u002FAngieNikki\u002FopenWUSU) -> WUSU is a semantic understanding dataset focusing on urban structure and the urbanization process in Wuhan\n* [RSE_Cross-city](https:\u002F\u002Fgithub.com\u002Fdanfenghong\u002FRSE_Cross-city) -> Cross-City Matters: A Multimodal Remote Sensing Benchmark Dataset for Cross-City Semantic Segmentation using High-Resolution Domain Adaptation Networks\n* [AErial Lane](https:\u002F\u002Fgithub.com\u002FAidasDir\u002FAerialLaneNet) -> AErial Lane (AEL) Dataset is a first large-scale aerial image dataset built for lane detection, with high-quality polyline lane annotations on high-resolution images of around 80 kilometers of road\n* [Chesapeake Roads Spatial Context (RSC)](https:\u002F\u002Fgithub.com\u002Fisaaccorley\u002FChesapeakeRSC)\n* [So2Sat-POP-DL](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSo2Sat-POP-DL) -> Dataset discovery: So2Sat Population dataset covering 98 EU cities\n* [HouseTS](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fshengkunwang\u002Fhousets-dataset) -> Long-term, Multimodal Housing Dataset Across 30 U.S. Metropolitan Area. Uses NAIP. [With paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.00765)\n* [10,000 Crop Field Boundaries across India](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7315090) -> using Airbus SPOT\n* [OpenEarthMap-SAR](https:\u002F\u002Fgithub.com\u002Fcliffbb\u002FOpenEarthMap-SAR) -> VHR SAR used in the 2025 IEEE GRSS Data Fusion Contest Track 1: All-Weather Land Cover Mapping. Utilises data from Umbra and Capella Space\n* [Tokyo Land Use Land Cover Dataset](https:\u002F\u002Fgithub.com\u002FTusaifei\u002FTokyo_dataset) ->  0.5-m resolution images, two kinds of 10-m resolution LCPs, and two kinds of 30-m resolution LCPs\n\n### Change Detection\n* [S2Looking](https:\u002F\u002Fgithub.com\u002FS2Looking\u002FDataset) -> A Satellite Side-Looking Dataset for Building Change Detection, [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2107.09244)\n* [Haiming-Z\u002FMtS-WH-reference-map](https:\u002F\u002Fgithub.com\u002FHaiming-Z\u002FMtS-WH-reference-map) -> a reference map for change detection based on MtS-WH\n* [MtS-WH-Dataset](https:\u002F\u002Fgithub.com\u002Frulixiang\u002FMtS-WH-Dataset) -> Multi-temporal Scene WuHan (MtS-WH) Dataset\n* [SZTAKI](http:\u002F\u002Fweb.eee.sztaki.hu\u002Fremotesensing\u002Fairchange_benchmark.html) -> A Ground truth collection for change detection in optical aerial images taken with several years time differences\n* [DSIFN](https:\u002F\u002Fgithub.com\u002FGeoZcx\u002FA-deeply-supervised-image-fusion-network-for-change-detection-in-remote-sensing-images\u002Ftree\u002Fmaster\u002Fdataset) -> change detection dataset, it consists of six large bi-temporal high resolution images covering six cities in China\n* [Road-Change-Detection-Dataset](https:\u002F\u002Fgithub.com\u002FfightingMinty\u002FRoad-Change-Detection-Dataset)\n* [3DCD](https:\u002F\u002Fsites.google.com\u002Funiroma1.it\u002F3dchangedetection\u002Fhome-page) -> infer 3D CD maps using only remote sensing optical bitemporal images as input without the need of Digital Elevation Models (DEMs)\n* [TUE-CD](https:\u002F\u002Fgithub.com\u002FRSMagneto\u002FMSI-Net) -> A change detection detection for building damage estimation after earthquake\n* [Hi-UCD](https:\u002F\u002Fgithub.com\u002FDaisy-7\u002FHi-UCD-S) -> ultra-High Urban Change Detection for urban semantic change detection\n* [LEVIR-CC-Dataset](https:\u002F\u002Fgithub.com\u002FChen-Yang-Liu\u002FLEVIR-CC-Dataset) -> A Large Dataset for Remote Sensing Image Change Captioning\n* [GDCLD](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13612636) -> A globally distributed dataset of coseismic landslide mapping via multi-source high-resolution remote sensing images\n* [BANet change dataset - RS image to cadastral map](https:\u002F\u002Fgithub.com\u002Flqycrystal\u002FBANet)\n* [Indian Cities Change Detection (ICCD) Dataset](https:\u002F\u002Fieee-dataport.org\u002Fdocuments\u002Findian-cities-change-detection-iccd-dataset)\n\n### SAR-Specific Datasets\n* [HRSID](https:\u002F\u002Fgithub.com\u002Fchaozhong2010\u002FHRSID) -> high resolution sar images dataset for ship detection, semantic segmentation, and instance segmentation tasks\n* [LS-SSDD-v1.0-OPEN](https:\u002F\u002Fgithub.com\u002FTianwenZhang0825\u002FLS-SSDD-v1.0-OPEN) -> Large-Scale SAR Ship Detection Dataset\n* [WHU-SEN-City](https:\u002F\u002Fgithub.com\u002Fwhu-csl\u002FWHU-SEN-City) -> A paired SAR-to-optical image translation dataset which covers 34 big cities of China\n* [SAR_vehicle_detection_dataset](https:\u002F\u002Fgithub.com\u002Fwhu-csl\u002FSAR_vehicle_detection_dataset) -> 104 SAR images for vehicle detection, collected from Sandia MiniSAR\u002FFARAD SAR images and MSTAR images\n* [AIR-PolSAR-Seg](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FAIR-PolSAR-Seg) -> a challenging PolSAR terrain segmentation dataset\n* [QXS-SAROPT](https:\u002F\u002Fgithub.com\u002Fyaoxu008\u002FQXS-SAROPT) -> Optical and SAR pairing dataset from the [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.08259): The QXS-SAROPT Dataset for Deep Learning in SAR-Optical Data Fusion\n* [SynthWakeSAR](https:\u002F\u002Fdata.bris.ac.uk\u002Fdata\u002Fdataset\u002F30kvuvmatwzij2mz1573zqumfx) -> A Synthetic SAR Dataset for Deep Learning Classification of Ships at Sea, with [paper](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F14\u002F16\u002F3999)\n* [SAR2Opt-Heterogeneous-Dataset](https:\u002F\u002Fgithub.com\u002FMarsZhaoYT\u002FSAR2Opt-Heterogeneous-Dataset) -> SAR-optical images to be used as a benchmark in change detection and image transaltion on remote sensing images\n* [OpenSARWake](https:\u002F\u002Fgithub.com\u002Flibzzluo\u002FOpenSARWake) -> A SAR ship wake rotation detection benchmark dataset.\n\n### Specialized Applications\n* [MUSIC4HA](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FMUSIC4HA) -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Hot Area\n* [MUSIC4GC](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FMUSIC4GC) -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Golf Course\n* [MUSIC4P3](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FMUSIC4P3) -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Photovoltaic Power Plants (solar panels)\n* [ABCDdataset](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FABCDdataset) -> damage detection dataset to identify whether buildings have been washed-away by tsunami\n* [Thermal power plans dataset](https:\u002F\u002Fgithub.com\u002FwenxinYin\u002FAIR-TPPDD)\n* [SolarDK](https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.01260) -> A high-resolution urban solar panel image classification and localization dataset\n* [Oil and Gas Infrastructure Mapping (OGIM) database](https:\u002F\u002Fzenodo.org\u002Frecord\u002F7922117) -> includes locations and facility attributes of oil and gas infrastructure types that are important sources of methane emissions\n* [Overhead Wind Turbine Dataset - NAIP](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7385227#.Y419qezMLdr)\n* [CloudTracks: A Dataset for Localizing Ship Tracks in Satellite Images of Clouds](https:\u002F\u002Fzenodo.org\u002Frecords\u002F10042922) -> the dataset consists of 1,780 MODIS satellite images hand-labeled for the presence of more than 12,000 ship tracks.\n* [Digital Typhoon Dataset](https:\u002F\u002Fgithub.com\u002Fkitamoto-lab\u002Fdigital-typhoon\u002F) -> aimed at benchmarking machine learning models for long-term spatio-temporal data\n* [BirdSAT](https:\u002F\u002Fgithub.com\u002Fmvrl\u002FBirdSAT) -> Cross-View iNAT Birds 2021: This cross-view birds species dataset consists of paired ground-level bird images and satellite images, along with meta-information associated with the iNaturalist-2021 dataset.\n* [RSHaze+](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13837162) -> remote sensing dehazing datasets in PhDnet: A novel physic-aware dehazing network for remote sensing images\n* [GMSEUS](https:\u002F\u002Fgithub.com\u002Fstidjaco\u002FGMSEUS) -> A comprehensive ground-mounted solar energy dataset with sub-array design metadata in the United States\n* [MultiviewRS](https:\u002F\u002Fgithub.com\u002Ffmenat\u002FmultiviewRS-datasets) -> List of remote sensing (RS) multi-view datasets for exploring multi-view learning\n* [SatDepth](https:\u002F\u002Fsatdepth.pythonanywhere.com\u002F) -> A Novel Dataset for Satellite Image Matching and Depth Estimation\n* [OpenSatMap](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fz-hb\u002FOpenSatMap) -> for large-scale map construction and downstream tasks like autonomous driving\n\n### Agricultural & Environmental\n* [Hyperspectral Change Detection Dataset Irrigated Agricultural Area](https:\u002F\u002Fgithub.com\u002FSicongLiuRS\u002FHyperspectral-Change-Detection-Dataset-Irrigated-Agricultural-Area)\n* [CNN-RNN-Yield-Prediction](https:\u002F\u002Fgithub.com\u002Fsaeedkhaki92\u002FCNN-RNN-Yield-Prediction) -> soybean dataset\n* [FireRisk](https:\u002F\u002Fgithub.com\u002FCharmonyShen\u002FFireRisk) -> A Remote Sensing Dataset for Fire Risk Assessment with Benchmarks Using Supervised and Self-supervised Learning\n* [TimeMatch](https:\u002F\u002Fzenodo.org\u002Frecords\u002F5636422) -> dataset for cross-region adaptation for crop identification from SITS in four different regions in Europe\n* [Landsat 8 Cloud Cover Assessment Validation Data](https:\u002F\u002Flandsat.usgs.gov\u002Flandsat-8-cloud-cover-assessment-validation-data)\n* [Remote Sensing Satellite Video Dataset for Super-resolution](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6969604#.ZCBd-OzMJhE)\n* [SpatioTemporalYield](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fellaampy\u002FSpatioTemporalYield) -> covers the USA’s top five corn-producing states: Iowa, Illinois, Indiana, Nebraska, and Minnesota.\n* [Palm Tree Dataset](https:\u002F\u002Fgithub.com\u002FNourO93\u002FPalm-Tree-Dataset\u002Ftree\u002Fmain)\n* [ts-satfire](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fz789456sx\u002Fts-satfire) -> A Multi-Task Satellite Image Time-Series Dataset for Wildfire Detection and Prediction\n* [GTPBD](https:\u002F\u002Fgithub.com\u002FZ-ZW-WXQ\u002FGTPBD\u002F) -> Global Terraced Parcel and Boundary Dataset\n\n### Hyperspectral & Multi-modal\n* [AeroRIT](https:\u002F\u002Fgithub.com\u002Faneesh3108\u002FAeroRIT) -> A New Scene for Hyperspectral Image Analysis\n* [Data-CSHSI](https:\u002F\u002Fgithub.com\u002FYuxiangZhang-BIT\u002FData-CSHSI) -> Open source datasets for Cross-Scene Hyperspectral Image Classification, includes Houston, Pavia & HyRank datasets\n* [HySpecNet-11k](https:\u002F\u002Fhyspecnet.rsim.berlin\u002F) -> a large-scale hyperspectral benchmark dataset\n* [STARCOP dataset: Semantic Segmentation of Methane Plumes with Hyperspectral Machine Learning Models](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7863343)\n* [Toulouse Hyperspectral Data Set](https:\u002F\u002Fwww.toulouse-hyperspectral-data-set.com\u002F)\n* [Toulouse Hyperspectral Data Set](https:\u002F\u002Fgithub.com\u002FRomain3Ch216\u002FTlseHypDataSet)\n* [Multi-modality-image-matching](https:\u002F\u002Fgithub.com\u002FStaRainJ\u002FMulti-modality-image-matching-database-metrics-methods) -> image matching dataset including several remote sensing modalities\n* [PanCollection](https:\u002F\u002Fgithub.com\u002Fliangjiandeng\u002FPanCollection) -> Pansharpening Datasets from WorldView 2, WorldView 3, QuickBird, Gaofen 2 sensors\n\n### Benchmark & Foundation Models\n* [EORSSD-dataset](https:\u002F\u002Fgithub.com\u002Frmcong\u002FEORSSD-dataset) -> Extended Optical Remote Sensing Saliency Detection (EORSSD) Dataset\n* [ERA-DATASET](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FERA-DATASET) -> A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos\n* [SSL4EO-S12](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSSL4EO-S12) -> a large-scale dataset for self-supervised learning in Earth observation\n* [AIR-CD](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FAIR-CD) -> a challenging cloud detection data set called AIR-CD, with higher spatial resolution and more representative landcover types\n* [HRC_WHU](https:\u002F\u002Fgithub.com\u002Fdr-lizhiwei\u002FHRC_WHU) -> High-Resolution Cloud Detection Dataset comprising 150 RGB images and a resolution varying from 0.5 to 15 m in different global regions\n* [University1652-Baseline](https:\u002F\u002Fgithub.com\u002Flayumi\u002FUniversity1652-Baseline) -> A Multi-view Multi-source Benchmark for Drone-based Geo-localization\n* [benchmark_ISPRS2021](https:\u002F\u002Fgithub.com\u002Fwhuwuteng\u002Fbenchmark_ISPRS2021) -> A new stereo dense matching benchmark dataset for deep learning\n* [WHU-Stereo](https:\u002F\u002Fgithub.com\u002FSheng029\u002FWHU-Stereo) -> A Challenging Benchmark for Stereo Matching of High-Resolution Satellite Images\n* [GeoPile pretraining dataset](https:\u002F\u002Fgithub.com\u002Fmmendiet\u002FGFM) -> compiles imagery from other datasets including RSD46-WHU, MLRSNet and RESISC45 for pretraining of Foundational models\n* [pangaea-bench](https:\u002F\u002Fgithub.com\u002Fyurujaja\u002Fpangaea-bench) -> A Global and Inclusive Benchmark for Geospatial Foundation Models\n* [VRSBench: A Versatile Vision-Language Benchmark Dataset for Remote Sensing Image Understanding](https:\u002F\u002Fvrsbench.github.io\u002F)\n* [SeeFar](https:\u002F\u002Fcoastalcarbon.ai\u002Fseefar) -> Satellite Agnostic Multi-Resolution Dataset for Geospatial Foundation Models\n* [dynnet](https:\u002F\u002Fgithub.com\u002Faysim\u002Fdynnet) -> DynamicEarthNet: Daily Multi-Spectral Satellite Dataset for Semantic Change Segmentation\n* [Awesome-Remote-Sensing-Relative-Radiometric-Normalization-Datasets](https:\u002F\u002Fgithub.com\u002FArminMoghimi\u002FAwesome-Remote-Sensing-Relative-Radiometric-Normalization-Datasets)\n* [AISD](https:\u002F\u002Fgithub.com\u002FRSrscoder\u002FAISD) -> Aerial Imagery dataset for Shadow Detection## Kaggle\nKaggle hosts over > 200 satellite image datasets, [search results here](https:\u002F\u002Fwww.kaggle.com\u002Fsearch?q=satellite+image+in%3Adatasets).\nThe [kaggle blog](http:\u002F\u002Fblog.kaggle.com) is an interesting read.\n\n### Kaggle - Amazon from space - classification challenge\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fplanet-understanding-the-amazon-from-space\u002Fdata\n* 3-5 meter resolution GeoTIFF images from planet Dove satellite constellation\n* 12 classes including - **cloudy, primary + waterway** etc\n* [1st place winner interview - used 11 custom CNN](http:\u002F\u002Fblog.kaggle.com\u002F2017\u002F10\u002F17\u002Fplanet-understanding-the-amazon-from-space-1st-place-winners-interview\u002F)\n* [FastAI Multi-label image classification](https:\u002F\u002Ftowardsdatascience.com\u002Ffastai-multi-label-image-classification-8034be646e95)\n* [Multi-Label Classification of Satellite Photos of the Amazon Rainforest](https:\u002F\u002Fmachinelearningmastery.com\u002Fhow-to-develop-a-convolutional-neural-network-to-classify-satellite-photos-of-the-amazon-rainforest\u002F)\n* [Understanding the Amazon Rainforest with Multi-Label Classification + VGG-19, Inceptionv3, AlexNet & Transfer Learning](https:\u002F\u002Ftowardsdatascience.com\u002Funderstanding-the-amazon-rainforest-with-multi-label-classification-vgg-19-inceptionv3-5084544fb655)\n* [amazon-classifier](https:\u002F\u002Fgithub.com\u002Fmikeskaug\u002Famazon-classifier) -> compares random forest with CNN\n* [multilabel-classification](https:\u002F\u002Fgithub.com\u002Fmuneeb706\u002Fmultilabel-classification) -> compares various CNN architecutres\n* [Planet-Amazon-Kaggle](https:\u002F\u002Fgithub.com\u002FSkumarr53\u002FPlanet-Amazon-Kaggle) -> uses fast.ai\n* [deforestation_deep_learning](https:\u002F\u002Fgithub.com\u002Fschumanzhang\u002Fdeforestation_deep_learning)\n* [Track-Human-Footprint-in-Amazon-using-Deep-Learning](https:\u002F\u002Fgithub.com\u002Fsahanasub\u002FTrack-Human-Footprint-in-Amazon-using-Deep-Learning)\n* [Amazon-Rainforest-CNN](https:\u002F\u002Fgithub.com\u002Fcldowdy\u002FAmazon-Rainforest-CNN) -> uses a 3-layer CNN in Tensorflow\n* [rainforest-tagging](https:\u002F\u002Fgithub.com\u002Fminggli\u002Frainforest-tagging) -> Convolutional Neural Net and Recurrent Neural Net in Tensorflow for satellite images multi-label classification\n* [satellite-deforestation](https:\u002F\u002Fgithub.com\u002Fdrewhibbard\u002Fsatellite-deforestation) -> Using Satellite Imagery to Identify the Leading Indicators of Deforestation, applied to the Kaggle Challenge Understanding the Amazon from Space\n\n### Kaggle - DSTL segmentation challenge\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fdstl-satellite-imagery-feature-detection\n* Rating - medium, many good examples (see the Discussion as well as kernels), but as this competition was run a couple of years ago many examples use python 2\n* WorldView 3 - 45 satellite images covering 1km x 1km in both 3 (i.e. RGB) and 16-band (400nm - SWIR) images\n* 10 Labelled classes include - **Buildings, Road, Trees, Crops, Waterway, Vehicles**\n* [Interview with 1st place winner who used segmentation networks](http:\u002F\u002Fblog.kaggle.com\u002F2017\u002F04\u002F26\u002Fdstl-satellite-imagery-competition-1st-place-winners-interview-kyle-lee\u002F) - 40+ models, each tweaked for particular target (e.g. roads, trees)\n* [ZF_UNET_224_Pretrained_Model 2nd place solution](https:\u002F\u002Fgithub.com\u002FZFTurbo\u002FZF_UNET_224_Pretrained_Model) ->\n* [3rd place soluton](https:\u002F\u002Fgithub.com\u002Fosin-vladimir\u002Fkaggle-satellite-imagery-feature-detection) -> which explored pansharpening & calculating reflectance indices, with [arxiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.06169) \n* [Deepsense 4th place solution](https:\u002F\u002Fdeepsense.ai\u002Fdeep-learning-for-satellite-imagery-via-image-segmentation\u002F)\n* [Entry by lopuhin](https:\u002F\u002Fgithub.com\u002Flopuhin\u002Fkaggle-dstl) using UNet with batch-normalization\n* [Multi-class semantic segmentation of satellite images using U-Net](https:\u002F\u002Fgithub.com\u002Frogerxujiang\u002Fdstl_unet) using DSTL dataset, tensorflow 1 & python 2.7. Accompanying [article](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20240930001745\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fdstl-satellite-imagery-contest-on-kaggle-2f3ef7b8ac40)\n* [Deep-Satellite-Image-Segmentation](https:\u002F\u002Fgithub.com\u002Fantoine-spahr\u002FDeep-Satellite-Image-Segmentation)\n* [Dstl-Satellite-Imagery-Feature-Detection-Improved](https:\u002F\u002Fgithub.com\u002Fcuulee\u002FDstl-Satellite-Imagery-Feature-Detection-Improved)\n* [Satellite-imagery-feature-detection](https:\u002F\u002Fgithub.com\u002FArangurenAndres\u002FSatellite-imagery-feature-detection)\n* [Satellite_Image_Classification](https:\u002F\u002Fgithub.com\u002Faditya-sawadh\u002FSatellite_Image_Classification) -> using XGBoost and ensemble classification methods\n* [Unet-for-Satellite](https:\u002F\u002Fgithub.com\u002Fjustinishikawa\u002FUnet-for-Satellite)\n* [building-segmentation](https:\u002F\u002Fgithub.com\u002Fjimpala\u002Fbuilding-segmentation) -> TensorFlow U-Net implementation trained to segment buildings in satellite imagery\n\n### Kaggle - DeepSat land cover classification\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat4 & https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat6\n* [DeepSat-Kaggle](https:\u002F\u002Fgithub.com\u002Fathulsudheesh\u002FDeepSat-Kaggle) -> uses Julia\n* [deepsat-aws-emr-pyspark](https:\u002F\u002Fgithub.com\u002Fhellosaumil\u002Fdeepsat-aws-emr-pyspark) -> Using PySpark for Image Classification on Satellite Imagery of Agricultural Terrains\n\n### Kaggle - Airbus ship detection challenge\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fairbus-ship-detection\u002Foverview\n* Rating - medium, most solutions using deep-learning, many kernels, [good example kernel](https:\u002F\u002Fwww.kaggle.com\u002Fkmader\u002Fbaseline-u-net-model-part-1)\n* [Detecting ships in satellite imagery: five years later…](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Fdetecting-ships-in-satellite-imagery-five-years-later-28df2e83f987)\n* I believe there was a problem with this dataset, which led to many complaints that the competition was ruined\n* [Lessons Learned from Kaggle’s Airbus Challenge](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241002082610\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Flessons-learned-from-kaggles-airbus-challenge-252e25c5efac)\n* [Airbus-Ship-Detection](https:\u002F\u002Fgithub.com\u002Fkheyer\u002FAirbus-Ship-Detection) -> This solution scored 139 out of 884 for the competition, combines ResNeXt50 based classifier and a U-net segmentation model\n* [Ship-Detection-Project](https:\u002F\u002Fgithub.com\u002FZTong1201\u002FShip-Detection-Project) -> uses Mask R-CNN and UNet model\n* [Airbus_SDC](https:\u002F\u002Fgithub.com\u002FWillieMaddox\u002FAirbus_SDC)\n* [Airbus_SDC_dup](https:\u002F\u002Fgithub.com\u002FWillieMaddox\u002FAirbus_SDC_dup) -> Project focused on detecting duplicate regions of overlapping satellite imagery. Applied to Airbus ship detection dataset\n* [airbus-ship-detection](https:\u002F\u002Fgithub.com\u002Fjancervenka\u002Fairbus-ship-detection) -> CNN with REST API\n* [Ship-Detection-from-Satellite-Images-using-YOLOV4](https:\u002F\u002Fgithub.com\u002Fdebasis-dotcom\u002FShip-Detection-from-Satellite-Images-using-YOLOV4) -> uses Kaggle Airbus Ship Detection dataset\n* [Image Segmentation: Kaggle experience](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20240929194243\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fimage-segmentation-kaggle-experience-9a41cb8924f0) -> Medium article by gold medal winner Vlad Shmyhlo\n\n### Kaggle - Ships in Google Earth\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ftomluther\u002Fships-in-google-earth\n* 794 jpegs showing various sized ships in satellite imagery, annotations in Pascal VOC format for object detection models\n* [\u002Fkaggle-ships-in-satellite-imagery-with-YOLOv8](https:\u002F\u002Fgithub.com\u002Frobmarkcole\u002Fkaggle-ships-in-satellite-imagery-with-YOLOv8)\n\n### Kaggle - Classify Ships in San Franciso Bay using Planet satellite imagery\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Frhammell\u002Fships-in-satellite-imagery\n* 4000 80x80 RGB images labeled with either a \"ship\" or \"no-ship\" classification, 3 meter pixel size\n* [shipsnet-detector](https:\u002F\u002Fgithub.com\u002Frhammell\u002Fshipsnet-detector) -> Detect container ships in Planet imagery using machine learning\n* [DeepLearningShipDetection](https:\u002F\u002Fgithub.com\u002FPenguinDan\u002FDeepLearningShipDetection)\n* [Ship-Detection-Using-Satellite-Imagery](https:\u002F\u002Fgithub.com\u002FDhruvisha29\u002FShip-Detection-Using-Satellite-Imagery)\n\n### Kaggle - Planesnet classification dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Frhammell\u002Fplanesnet -> Detect aircraft in Planet satellite image chips\n* 20x20 RGB images, the \"plane\" class includes 8000 images and the \"no-plane\" class includes 24000 images\n* [Dataset repo](https:\u002F\u002Fgithub.com\u002Frhammell\u002Fplanesnet) and [planesnet-detector](https:\u002F\u002Fgithub.com\u002Frhammell\u002Fplanesnet-detector) demonstrates a small CNN classifier on this dataset\n* [ergo-planes-detector](https:\u002F\u002Fgithub.com\u002Fevilsocket\u002Fergo-planes-detector) -> An ergo based project that relies on a convolutional neural network to detect airplanes from satellite imagery, uses the PlanesNet dataset\n* [Using AWS SageMaker\u002FPlanesNet to process Satellite Imagery](https:\u002F\u002Fgithub.com\u002Fkskalvar\u002Faws-sagemaker-planesnet-imagery)\n* [Airplane-in-Planet-Image](https:\u002F\u002Fgithub.com\u002FMaxLenormand\u002FAirplane-in-Planet-Image) -> pytorch model\n\n### Kaggle - CGI Planes in Satellite Imagery w\u002F BBoxes\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Faceofspades914\u002Fcgi-planes-in-satellite-imagery-w-bboxes\n* 500 computer generated satellite images of planes\n* [Faster RCNN to detect airplanes](https:\u002F\u002Fgithub.com\u002FShubhankarRawat\u002FAirplane-Detection-for-Satellites)\n* [aircraft-detection-from-satellite-images-yolov3](https:\u002F\u002Fgithub.com\u002Femrekrtorun\u002Faircraft-detection-from-satellite-images-yolov3)\n\n### Kaggle - Swimming pool and car detection using satellite imagery\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fkbhartiya83\u002Fswimming-pool-and-car-detection\n* 3750 satellite images of residential areas with annotation data for swimming pools and cars\n* [Object detection on Satellite Imagery using RetinaNet](https:\u002F\u002Fmedium.com\u002F@ije_good\u002Fobject-detection-on-satellite-imagery-using-retinanet-part-1-training-e589975afbd5)\n\n### Kaggle - Draper challenge to place images in order of time\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fdraper-satellite-image-chronology\u002Fdata\n* Rating - hard. Not many useful kernels.\n* Images are grouped into sets of five, each of which have the same setId. Each image in a set was taken on a different day (but not necessarily at the same time each day). The images for each set cover approximately the same area but are not exactly aligned.\n* Kaggle interviews for entrants who [used XGBOOST](http:\u002F\u002Fblog.kaggle.com\u002F2016\u002F09\u002F15\u002Fdraper-satellite-image-chronology-machine-learning-solution-vicens-gaitan\u002F) and a [hybrid human\u002FML approach](http:\u002F\u002Fblog.kaggle.com\u002F2016\u002F09\u002F08\u002Fdraper-satellite-image-chronology-damien-soukhavong\u002F)\n* [deep-cnn-sat-image-time-series](https:\u002F\u002Fgithub.com\u002FMickyDowns\u002Fdeep-cnn-sat-image-time-series) -> uses LSTM\n\n### Kaggle - Dubai segmentation\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fhumansintheloop\u002Fsemantic-segmentation-of-aerial-imagery\n* 72 satellite images of Dubai, the UAE, and is segmented into 6 classes\n* [dubai-satellite-imagery-segmentation](https:\u002F\u002Fgithub.com\u002Fayushdabra\u002Fdubai-satellite-imagery-segmentation) -> due to the small dataset, image augmentation was used\n* [U-Net for Semantic Segmentation on Unbalanced Aerial Imagery](https:\u002F\u002Ftowardsdatascience.com\u002Fu-net-for-semantic-segmentation-on-unbalanced-aerial-imagery-3474fa1d3e56) -> using the Dubai dataset\n* [Semantic-Segmentation-using-U-Net](https:\u002F\u002Fgithub.com\u002FAnay21110\u002FSemantic-Segmentation-using-U-Net) -> uses keras\n* [unet_satelite_image_segmentation](https:\u002F\u002Fgithub.com\u002Fnassimaliou\u002Funet_satelite_image_segmentation)\n\n### Kaggle - Massachusetts Roads & Buildings Datasets - segmentation\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fmassachusetts-roads-dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fmassachusetts-buildings-dataset\n* [Official published dataset](https:\u002F\u002Fwww.cs.toronto.edu\u002F~vmnih\u002Fdata\u002F)\n* [Road_seg_dataset](https:\u002F\u002Fgithub.com\u002Fparth1620\u002FRoad_seg_dataset) -> subset of the roads dataset containing only 200 images and masks\n* [Road and Building Semantic Segmentation in Satellite Imagery](https:\u002F\u002Fgithub.com\u002FPaulymorphous\u002FRoad-Segmentation) uses U-Net on the Massachusetts Roads Dataset & keras\n* [Semantic-segmentation repo by fuweifu-vtoo](https:\u002F\u002Fgithub.com\u002Ffuweifu-vtoo\u002FSemantic-segmentation) -> uses pytorch and the [Massachusetts Buildings & Roads Datasets](https:\u002F\u002Fwww.cs.toronto.edu\u002F~vmnih\u002Fdata\u002F)\n* [ssai-cnn](https:\u002F\u002Fgithub.com\u002Fmitmul\u002Fssai-cnn) -> This is an implementation of Volodymyr Mnih's dissertation methods on his Massachusetts road & building dataset\n* [building-footprint-segmentation](https:\u002F\u002Fgithub.com\u002Ffuzailpalnak\u002Fbuilding-footprint-segmentation) -> pip installable library to train building footprint segmentation on satellite and aerial imagery, applied to Massachusetts Buildings Dataset and Inria Aerial Image Labeling Dataset\n* [Road detection using semantic segmentation and albumentations for data augmention](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20240929191243\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Froad-detection-using-segmentation-models-and-albumentations-libraries-on-keras-d5434eaf73a8) using the Massachusetts Roads Dataset, U-net & Keras\n* [Image-Segmentation)](https:\u002F\u002Fgithub.com\u002Fmschulz\u002FImage-Segmentation) -> using Massachusetts Road dataset and fast.ai\n\n### Kaggle - Deepsat classification challenge\nNot satellite but airborne imagery. Each sample image is 28x28 pixels and consists of 4 bands - red, green, blue and near infrared. The training and test labels are one-hot encoded 1x6 vectors. Each image patch is size normalized to 28x28 pixels. Data in `.mat` Matlab format. JPEG?\n* [Sat4](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat4) 500,000 image patches covering four broad land cover classes - **barren land, trees, grassland and a class that consists of all land cover classes other than the above three**\n* [Sat6](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat6) 405,000 image patches each of size 28x28 and covering 6 landcover classes - **barren land, trees, grassland, roads, buildings and water bodies.**\n\n### Kaggle - High resolution ship collections 2016 (HRSC2016)\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fguofeng\u002Fhrsc2016\n* Ship images harvested from Google Earth\n* [HRSC2016_SOTA](https:\u002F\u002Fgithub.com\u002Fming71\u002FHRSC2016_SOTA) -> Fair comparison of different algorithms on the HRSC2016 dataset\n\n### Kaggle - SWIM-Ship Wake Imagery Mass\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Flilitopia\u002Fswimship-wake-imagery-mass\n* An optical ship wake detection benchmark dataset built for deep learning\n* [WakeNet](https:\u002F\u002Fgithub.com\u002FLilytopia\u002FWakeNet) -> A CNN-based optical image ship wake detector, code for 2021 paper: Rethinking Automatic Ship Wake Detection: State-of-the-Art CNN-based Wake Detection via Optical Images\n\n### Kaggle - Understanding Clouds from Satellite Images\nIn this challenge, you will build a model to classify cloud organization patterns from satellite images.\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Funderstanding_cloud_organization\u002F\n* [3rd place solution on Github by naivelamb](https:\u002F\u002Fgithub.com\u002Fnaivelamb\u002Fkaggle-cloud-organization)\n* [15th place solution on Github by Soongja](https:\u002F\u002Fgithub.com\u002FSoongja\u002Fkaggle-clouds)\n* [69th place solution on Github by yukkyo](https:\u002F\u002Fgithub.com\u002Fyukkyo\u002FKaggle-Understanding-Clouds-69th-solution)\n* [161st place solution on Github by michal-nahlik](https:\u002F\u002Fgithub.com\u002Fmichal-nahlik\u002Fkaggle-clouds-2019)\n* [Solution by yurayli](https:\u002F\u002Fgithub.com\u002Fyurayli\u002Fsatellite-cloud-segmentation)\n* [Solution by HazelMartindale](https:\u002F\u002Fgithub.com\u002FHazelMartindale\u002Fkaggle_understanding_clouds_learning_project) uses 3 versions of U-net architecture\n* [Solution by khornlund](https:\u002F\u002Fgithub.com\u002Fkhornlund\u002Funderstanding-cloud-organization)\n* [Solution by Diyago](https:\u002F\u002Fgithub.com\u002FDiyago\u002FUnderstanding-Clouds-from-Satellite-Images)\n* [Solution by tanishqgautam](https:\u002F\u002Fgithub.com\u002Ftanishqgautam\u002FMulti-Label-Segmentation-With-FastAI)\n\n### Kaggle - 38-Cloud Cloud Segmentation\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsorour\u002F38cloud-cloud-segmentation-in-satellite-images\n* Contains 38 Landsat 8 images and manually extracted pixel-level ground truths\n* [38-Cloud Github repository](https:\u002F\u002Fgithub.com\u002FSorourMo\u002F38-Cloud-A-Cloud-Segmentation-Dataset) and follow up [95-Cloud](https:\u002F\u002Fgithub.com\u002FSorourMo\u002F95-Cloud-An-Extension-to-38-Cloud-Dataset) dataset\n* [How to create a custom Dataset \u002F Loader in PyTorch, from Scratch, for multi-band Satellite Images Dataset from Kaggle](https:\u002F\u002Fmedium.com\u002Fanalytics-vidhya\u002Fhow-to-create-a-custom-dataset-loader-in-pytorch-from-scratch-for-multi-band-satellite-images-c5924e908edf)\n* [Cloud-Net: A semantic segmentation CNN for cloud detection](https:\u002F\u002Fgithub.com\u002FSorourMo\u002FCloud-Net-A-semantic-segmentation-CNN-for-cloud-detection) -> an end-to-end cloud detection algorithm for Landsat 8 imagery, trained on 38-Cloud Training Set\n* [Segmentation of Clouds in Satellite Images Using Deep Learning](https:\u002F\u002Fmedium.com\u002Fswlh\u002Fsegmentation-of-clouds-in-satellite-images-using-deep-learning-a9f56e0aa83d) -> semantic segmentation using a Unet on the Kaggle 38-Cloud dataset\n\n### Kaggle - Airbus Aircraft Detection Dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fairbusgeo\u002Fairbus-aircrafts-sample-dataset\n* One hundred civilian airports and over 3000 annotated commercial aircrafts\n* [detecting-aircrafts-on-airbus-pleiades-imagery-with-yolov5](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Fdetecting-aircrafts-on-airbus-pleiades-imagery-with-yolov5-5f3d464b75ad)\n* [pytorch-remote-sensing](https:\u002F\u002Fgithub.com\u002Fmiko7879\u002Fpytorch-remote-sensing) -> Aircraft detection using the 'Airbus Aircraft Detection' dataset and Faster-RCNN with ResNet-50 backbone in pytorch\n\n### Kaggle - Airbus oil storage detection dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fairbusgeo\u002Fairbus-oil-storage-detection-dataset\n* [Oil-Storage Tank Instance Segmentation with Mask R-CNN](https:\u002F\u002Fgithub.com\u002Fgeorgiosouzounis\u002Finstance-segmentation-mask-rcnn\u002Fblob\u002Fmain\u002Fmask_rcnn_oiltanks_gpu.ipynb) with [accompanying article](https:\u002F\u002Fmedium.com\u002F@georgios.ouzounis\u002Foil-storage-tank-instance-segmentation-with-mask-r-cnn-77c94433045f)\n* [Oil Storage Detection on Airbus Imagery with YOLOX](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Foil-storage-detection-on-airbus-imagery-with-yolox-9e38eb6f7e62) -> uses the Kaggle Airbus Oil Storage Detection dataset\n* [Oil-Storage-Tanks-Data-Preparation-YOLO-Format](https:\u002F\u002Fgithub.com\u002Fshah0nawaz\u002FOil-Storage-Tanks-Data-Preparation-YOLO-Format)\n\n### Kaggle - Satellite images of hurricane damage\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fkmader\u002Fsatellite-images-of-hurricane-damage\n* https:\u002F\u002Fgithub.com\u002Fdbuscombe-usgs\u002FHurricaneHarvey_buildingdamage\n\n### Kaggle - Austin Zoning Satellite Images\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ffranchenstein\u002Faustin-zoning-satellite-images\n* classify a images of Austin into one of its zones, such as residential, industrial, etc. 3667 satellite images\n\n### Kaggle - Statoil\u002FC-CORE Iceberg Classifier Challenge\nClassify the target in a SAR image chip as either a ship or an iceberg. The dataset for the competition included 5000 images extracted from multichannel SAR data collected by the Sentinel-1 satellite. Top entries used ensembles to boost prediction accuracy from about 92% to 97%.\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fstatoil-iceberg-classifier-challenge\u002Fdata\n* [An interview with David Austin: 1st place winner](https:\u002F\u002Fpyimagesearch.com\u002F2018\u002F03\u002F26\u002Finterview-david-austin-1st-place-25000-kaggles-popular-competition\u002F)\n* [radar-image-recognition](https:\u002F\u002Fgithub.com\u002Fsiarez\u002Fradar-image-recognition)\n* [Iceberg-Classification-Using-Deep-Learning](https:\u002F\u002Fgithub.com\u002Fmankadronit\u002FIceberg-Classification-Using-Deep-Learning) -> uses keras\n* [Deep-Learning-Project](https:\u002F\u002Fgithub.com\u002Fsingh-shakti94\u002FDeep-Learning-Project) -> uses keras\n* [iceberg-classifier-challenge solution by ShehabSunny](https:\u002F\u002Fgithub.com\u002FShehabSunny\u002Ficeberg-classifier-challenge) -> uses keras\n* [Analyzing Satellite Radar Imagery with Deep Learning](https:\u002F\u002Fuk.mathworks.com\u002Fcompany\u002Fnewsletters\u002Farticles\u002Fanalyzing-satellite-radar-imagery-with-deep-learning.html) -> by Matlab, uses ensemble with greedy search\n* [16th place solution](https:\u002F\u002Fgithub.com\u002Fsergeyshilin\u002Fkaggle-statoil-iceberg-classifier-challenge)\n* [fastai solution](https:\u002F\u002Fgithub.com\u002Fsmarkochev\u002Fds_notebooks\u002Fblob\u002Fmaster\u002FStatoil_Kaggle_competition_google_colab_notebook.ipynb)\n\n### Kaggle - Land Cover Classification Dataset from DeepGlobe Challenge - segmentation\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fdeepglobe-land-cover-classification-dataset\n* [Satellite Imagery Semantic Segmentation with CNN](https:\u002F\u002Fjoshting.medium.com\u002Fsatellite-imagery-segmentation-with-convolutional-neural-networks-f9254de3b907) -> 7 different segmentation classes, DeepGlobe Land Cover Classification Challenge dataset, with [repo](https:\u002F\u002Fgithub.com\u002Fjustjoshtings\u002Fsatellite_image_segmentation)\n* [Land Cover Classification with U-Net](https:\u002F\u002Fbaratam-tarunkumar.medium.com\u002Fland-cover-classification-with-u-net-aa618ea64a1b) -> Satellite Image Multi-Class Semantic Segmentation Task with PyTorch Implementation of U-Net, uses DeepGlobe Land Cover Segmentation dataset, with [code](https:\u002F\u002Fgithub.com\u002FTarunKumar1995-glitch\u002Fland_cover_classification_unet)\n* [DeepGlobe Land Cover Classification Challenge solution](https:\u002F\u002Fgithub.com\u002FGeneralLi95\u002Fdeepglobe_land_cover_classification_with_deeplabv3plus)\n\n### Kaggle - Next Day Wildfire Spread\nA Data Set to Predict Wildfire Spreading from Remote-Sensing Data\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ffantineh\u002Fnext-day-wildfire-spread\n* https:\u002F\u002Farxiv.org\u002Fabs\u002F2112.02447\n\n### Kaggle - Satellite Next Day Wildfire Spread\nInspired by the above dataset, using different data sources\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsatellitevu\u002Fsatellite-next-day-wildfire-spread\n* https:\u002F\u002Fgithub.com\u002FSatelliteVu\u002FSatelliteVu-AWS-Disaster-Response-Hackathon\n\n## Kaggle - Spacenet 7 Multi-Temporal Urban Change Detection\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Famerii\u002Fspacenet-7-multitemporal-urban-development\n* [SatFootprint](https:\u002F\u002Fgithub.com\u002FPriyanK7n\u002FSatFootprint) -> building segmentation on the Spacenet 7 dataset\n\n## Kaggle - Satellite Images to predict poverty in Africa\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsandeshbhat\u002Fsatellite-images-to-predict-povertyafrica\n* Uses satellite imagery and nightlights data to predict poverty levels at a local level\n* [Predicting-Poverty](https:\u002F\u002Fgithub.com\u002Fjmather625\u002Fpredicting-poverty-replication) -> Combining satellite imagery and machine learning to predict poverty, in PyTorch\n\n## Kaggle - NOAA Fisheries Steller Sea Lion Population Count\n* https:\u002F\u002Fwww.kaggle.com\u002Fcompetitions\u002Fnoaa-fisheries-steller-sea-lion-population-count -> count sea lions from aerial images\n* [Sealion-counting](https:\u002F\u002Fgithub.com\u002Fbabyformula\u002FSealion-counting)\n* [Sealion_Detection_Classification](https:\u002F\u002Fgithub.com\u002Fyyc9268\u002FSealion_Detection_Classification)\n\n## Kaggle - Arctic Sea Ice Image Masking\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Falexandersylvester\u002Farctic-sea-ice-image-masking\n* [sea_ice_remote_sensing](https:\u002F\u002Fgithub.com\u002Fsum1lim\u002Fsea_ice_remote_sensing)\n\n## Kaggle - Overhead-MNIST\n* A Benchmark Satellite Dataset as Drop-In Replacement for MNIST\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fdatamunge\u002Foverheadmnist -> kaggle\n* https:\u002F\u002Farxiv.org\u002Fabs\u002F2102.04266 -> paper\n* https:\u002F\u002Fgithub.com\u002Freveondivad\u002Fov-mnist -> github\n\n## Kaggle - Satellite Image Classification\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fmahmoudreda55\u002Fsatellite-image-classification\n* [satellite-image-classification-pytorch](https:\u002F\u002Fgithub.com\u002Fdilaraozdemir\u002Fsatellite-image-classification-pytorch)\n\n## Kaggle - EuroSAT - Sentinel-2 Dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fraoofnaushad\u002Feurosat-sentinel2-dataset\n* RGB Land Cover and Land Use Classification using Sentinel-2 Satellite\n* Used in paper [Image Augmentation for Satellite Images](https:\u002F\u002Farxiv.org\u002Fabs\u002F2207.14580)\n\n## Kaggle - Satellite Images of Water Bodies\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ffranciscoescobar\u002Fsatellite-images-of-water-bodies\n* [pytorch-waterbody-segmentation](https:\u002F\u002Fgithub.com\u002Fgauthamk02\u002Fpytorch-waterbody-segmentation) -> UNET model trained on the Satellite Images of Water Bodies dataset from Kaggle. The model is deployed on Hugging Face Spaces\n\n## Kaggle - NOAA sea lion count\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fnoaa-fisheries-steller-sea-lion-population-count\n* [noaa](https:\u002F\u002Fgithub.com\u002Fdarraghdog\u002Fnoaa) -> UNET, object detection and image level regression approaches\n\n### Kaggle - miscellaneous\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Freubencpereira\u002Fspatial-data-repo -> Satellite + loan data\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ftowardsentropy\u002Foil-storage-tanks -> Image data of industrial oil tanks with bounding box annotations, estimate tank fill % from shadows\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fairbusgeo\u002Fairbus-wind-turbines-patches -> Airbus SPOT satellites images over wind turbines for classification\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Faceofspades914\u002Fcgi-planes-in-satellite-imagery-w-bboxes -> CGI planes object detection dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fatilol\u002Faerialimageryforroofsegmentation -> Aerial Imagery for Roof Segmentation\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fandrewmvd\u002Fship-detection -> 621 images of boats and ships\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Falpereniek\u002Fvehicle-detection-from-satellite-images-data-set\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsergiishchus\u002Fmaxar-satellite-data -> Example Maxar data at 15 cm resolution\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcici118\u002Fswimming-pool-detection-algarves-landscape\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fdonkroco\u002Fsolar-panel-module -> object detection for solar panels\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fdeepglobe-road-extraction-dataset -> segment roads\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ftowardsentropy\u002Foil-storage-tanks -> Image data of industrial Oil Storage Tanks with bounding box annotations\n* https:\u002F\u002Fwww.kaggle.com\u002Fcompetitions\u002Fwidsdatathon2019\u002F -> Palm oil plantations\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsiddharthkumarsah\u002Fships-in-aerial-images -> Ships\u002FVessels in Aerial Images\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fjangsienicajzkowy\u002Fafo-aerial-dataset-of-floating-objects -> Aerial dataset for maritime Search and Rescue applications\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fyaroslavnaychuk\u002Fsatelliteimagesegmentation -> Segmentation on Gaofen Satellite Image, extracted from GID-15 dataset\n\n# Competitions\nCompetitions are an excellent source for accessing clean, ready-to-use satellite datasets and model benchmarks.  \n\n* https:\u002F\u002Fcodalab.lisn.upsaclay.fr\u002Fcompetitions\u002F9603 -> object detection from diversified satellite imagery\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F143\u002Ftick-tick-bloom\u002F -> detect and classify algal bloom\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F81\u002Fdetect-flood-water\u002F -> map floodwater from radar imagery\n* https:\u002F\u002Fplatform.ai4eo.eu\u002Fenhanced-sentinel2-agriculture -> map cultivated land using Sentinel imagery\n* https:\u002F\u002Fwww.diu.mil\u002Fai-xview-challenge -> multiple challenges ranging from detecting fishing vessals to estimating building damages\n* https:\u002F\u002Fcompetitions.codalab.org\u002Fcompetitions\u002F30440 -> flood detection\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F83\u002Fcloud-cover\u002F -> cloud cover detection\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F78\u002Foverhead-geopose-challenge\u002Fpage\u002F372\u002F -> predicts geocentric pose from single-view oblique satellite images\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F60\u002Fbuilding-segmentation-disaster-resilience\u002F -> building segmentation\n* https:\u002F\u002Fcaptain-whu.github.io\u002FDOTA\u002F -> large dataset for object detection in aerial imagery\n* https:\u002F\u002Fspacenet.ai\u002F -> set of 8 challenges such as road network detection\n* https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fcompetitions\u002FChaBuD-ECML-PKDD2023 -> binary image segmentation task on forest fires monitored over California\n\u003C!-- markdown-link-check-disable -->\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F6269285b14d764000d798fde -> ML for floods\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F60002402f5647f00129f7287 -> lightning and extreme weather\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F6025107d79c197001219c481\u002Ftrue -> ~1TB dataset for precipitation forecasting\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F61c0a1b9ff8868000dfb79e1\u002Ftrue -> Sentinel-2 image super-resolution\n\u003C!-- markdown-link-check-enable --\n","\u003Cdiv align=\"center\">\n  \u003Cp>\n    \u003Ca href=\"https:\u002F\u002Fwww.satellite-image-deep-learning.com\u002F\">\n        \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsatellite-image-deep-learning_datasets_readme_6558fd4f23db.png\" width=\"700\">\n    \u003C\u002Fa>\n\u003C\u002Fp>\n  \u003Ch2>用于卫星与航空影像深度学习的数据集。\u003C\u002Fh2>\n\n# 👉 [satellite-image-deep-learning.com](https:\u002F\u002Fwww.satellite-image-deep-learning.com\u002F) 👈\n\n\u003C\u002Fdiv>\n\n**如何使用本仓库：** 如果您确切知道要查找的内容（例如，论文名称），可以使用 `Control+F` 在此页面中搜索（或直接在原始 Markdown 文件中搜索）。\n\n# 数据集列表\n\u003C!-- markdown-link-check-disable -->\n* [地球观测数据库](https:\u002F\u002Feod-grss-ieee.com\u002F)\n\u003C!-- markdown-link-check-enable -->\n* [awesome-satellite-imagery-datasets](https:\u002F\u002Fgithub.com\u002Fchrieke\u002Fawesome-satellite-imagery-datasets)\n* [Awesome_Satellite_Benchmark_Datasets](https:\u002F\u002Fgithub.com\u002FSeyed-Ali-Ahmadi\u002FAwesome_Satellite_Benchmark_Datasets)\n* [awesome-remote-sensing-change-detection](https:\u002F\u002Fgithub.com\u002Fwenhwu\u002Fawesome-remote-sensing-change-detection) -> 专门针对变化检测\n* [Callisto-Dataset-Collection](https:\u002F\u002Fgithub.com\u002FAgri-Hub\u002FCallisto-Dataset-Collection) -> 使用哥白尼\u002F哨兵数据集的集合\n* [geospatial-data-catalogs](https:\u002F\u002Fgithub.com\u002Fgiswqs\u002Fgeospatial-data-catalogs) -> AWS、Earth Engine、Planetary Computer 和 STAC Index 上可用的开源地理空间数据集列表\n* [BED4RS](https:\u002F\u002Fcaptain-whu.github.io\u002FBED4RS\u002F)\n* [Satellite-Image-Time-Series-Datasets](https:\u002F\u002Fgithub.com\u002Fcorentin-dfg\u002FSatellite-Image-Time-Series-Datasets)\n\n# 遥感数据集中心\n* [Radiant MLHub](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241213015814\u002Fhttps:\u002F\u002Fmlhub.earth\u002F) -> 同时提供数据集和模型\n* [AWS 开放数据注册表](https:\u002F\u002Fregistry.opendata.aws)\n* [微软 Planetary Computer 数据目录](https:\u002F\u002Fplanetarycomputer.microsoft.com\u002Fcatalog)\n* [谷歌 Earth Engine 数据目录](https:\u002F\u002Fdevelopers.google.com\u002Fearth-engine\u002Fdatasets)\n\n## 哨兵系列\n作为 [欧盟哥白尼计划](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCopernicus_Programme) 的一部分，多颗哨兵卫星正在采集影像 -> 参见 [维基百科](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCopernicus_Programme#Sentinel_missions)。\n\n### 哨兵-1（SAR）\n* [用于哥白尼哨兵-1 卫星数据产品的 Xarray 后端](https:\u002F\u002Fgithub.com\u002Fbopen\u002Fxarray-sentinel)\n* [mmflood](https:\u002F\u002Fgithub.com\u002Fedornd\u002Fmmflood) -> 基于哨兵-1 SAR 影像的洪水范围提取，相关论文见 [IEEE Xplore](https:\u002F\u002Fieeexplore.ieee.org\u002Fabstract\u002Fdocument\u002F9882096)\n* [Sentinel-1 for Science Amazonas](https:\u002F\u002Fsen4ama.gisat.cz\u002F) -> 森林损失时间序列数据集\n* [CYCleSS](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41597-025-06528-x) -> 一个综合性的英国作物产量数据集，整合了卫星、气象和土壤类型等信息\n\n### Sentinel-2（光学）\n* [Sentinel-2 云优化 GeoTIFF 文件](https:\u002F\u002Fregistry.opendata.aws\u002Fsentinel-2-l2a-cogs\u002F) 和 [Sentinel-2 L2A 120m 拼接图](https:\u002F\u002Fregistry.opendata.aws\u002Fsentinel-s2-l2a-mosaic-120\u002F)\n* [GCP 上的开放数据](https:\u002F\u002Fconsole.cloud.google.com\u002Fstorage\u002Fbrowser\u002Fgcp-public-data-sentinel-2?prefix=tiles%2F31%2FT%2FCJ%2F)\n* [在笔记本中加载 Sentinel 数据的示例](https:\u002F\u002Fgithub.com\u002Fbinder-examples\u002Fgetting-data\u002Fblob\u002Fmaster\u002FSentinel2.ipynb)\n* [使用 Keras 在 Python 中分析 Sentinel-2 卫星数据](https:\u002F\u002Fgithub.com\u002Fjensleitloff\u002FCNN-Sentinel)\n* [SEN2VENµS](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6514159#.YoRxM5PMK3I) -> 用于训练 Sentinel-2 超分辨率算法的数据集\n* [M3LEO](https:\u002F\u002Fhuggingface.co\u002FM3LEO) -> [Github](https:\u002F\u002Fgithub.com\u002Fspaceml-org\u002FM3LEO)。一个超大规模的地理参考数据集，包含 Sentinel 1\u002F2 影像、干涉 SAR 产品以及土地覆盖、生物量和数字高程模型等辅助数据。\n* [SEN12MS](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSEN12MS) -> 为深度学习和数据融合精心构建的地理参考多光谱 Sentinel-1\u002F2 影像数据集。可查看 [SEN12MS 工具箱](https:\u002F\u002Fgithub.com\u002Fschmitt-muc\u002FSEN12MS)，并在 [paperswithcode.com](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fsen12ms) 上找到大量相关应用。\n* [SEN2NAIP](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Ftacofoundation\u002FSEN2NAIPv2) -> 空间和光谱协调的 Sen-2 + NAIP 数据集，用于实现 4 倍 RGB-NIR 超分辨率。\n* [Sen4AgriNet](https:\u002F\u002Fgithub.com\u002FOrion-AI-Lab\u002FS4A) -> 一个多年跨国家的 Sentinel-2 基准数据集，用于作物分类和分割的深度学习，并提供 [模型](https:\u002F\u002Fgithub.com\u002FOrion-AI-Lab\u002FS4A-Models)。\n* [sentinel2tools](https:\u002F\u002Fgithub.com\u002FQuantuMobileSoftware\u002Fsentinel2tools) -> 下载和基础处理 Sentinel 2 图像的工具库。阅读 [Sentinel2tools：下载 Sentinel-2 卫星图像的简单库](https:\u002F\u002Fmedium.com\u002Fgeekculture\u002Fsentinel2tools-simple-lib-for-downloading-sentinel-2-satellite-images-f8a6be3ee894)。\n* [open-sentinel-map](https:\u002F\u002Fgithub.com\u002FVisionSystemsInc\u002Fopen-sentinel-map) -> OpenSentinelMap 数据集包含 Sentinel-2 影像以及基于 OpenStreetMap 衍生的像素级语义标签掩码。\n* [Canadian-cropland-dataset](https:\u002F\u002Fgithub.com\u002FbioinfoUQAM\u002FCanadian-cropland-dataset) -> 一个新颖的基于补丁的数据集，由从 Sentinel-2 获取的加拿大农业耕地光学卫星影像编译而成。\n* [Sentinel-2 云覆盖分割数据集](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20230606184945\u002Fhttps:\u002F\u002Fmlhub.earth\u002Fdata\u002Fref_cloud_cover_detection_challenge_v1) 在 Radiant mlhub 上。\n* [Azavea 云数据集](https:\u002F\u002Fwww.azavea.com\u002Fblog\u002F2021\u002F08\u002F02\u002Fthe-azavea-cloud-dataset\u002F) 用于训练此 [云模型](https:\u002F\u002Fgithub.com\u002Fazavea\u002Fcloud-model)。\n* [fMoW-Sentinel](https:\u002F\u002Fpurl.stanford.edu\u002Fvg497cb6002) -> 世界功能地图 - Sentinel-2 对应影像（fMoW-Sentinel）数据集由 Sentinel-2 卫星采集的影像时间序列组成，对应于世界功能地图（fMoW）数据集中多个不同地点的不同时间点。用于 [SatMAE](https:\u002F\u002Fgithub.com\u002Fsustainlab-group\u002FSatMAE)。\n* [地球地表水数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F5205674#.Y4iEFezP1hE) -> 用于深度学习地表水特征的 Sentinel-2 卫星影像数据集。参见 [在 torchgeo 中使用该数据集的示例](https:\u002F\u002Ftowardsdatascience.com\u002Fartificial-intelligence-for-geospatial-analysis-with-pytorchs-torchgeo-part-1-52d17e409f09)。\n* [Ship-S2-AIS 数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F7229756#.Y5GsgOzP1hE) -> 从 29 个免费 Sentinel-2 产品中提取的 13,000 个裁剪块。其中 2,000 张图像显示丹麦主权水域中的船只：可以检测到货船、渔船或集装箱船。\n* [亚马逊雨林语义分割数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F3233081#.Y6LPLOzP1hE) -> Sentinel 2 影像。\n* [MATTER](https:\u002F\u002Fgithub.com\u002Fperiakiva\u002FMATTER) -> 用于自监督训练的 Sentinel 2 数据集。\n* [S2GLC](https:\u002F\u002Fs2glc.cbk.waw.pl\u002F) -> 欧洲高分辨率土地覆盖地图。\n* [从多光谱 Sentinel-2 卫星影像生成不透水面地图](https:\u002F\u002Fzenodo.org\u002Frecord\u002F7058860#.ZDrAeuzMLdo)。\n* [Sentinel-2 水体边缘数据集（SWED）](https:\u002F\u002Fopenmldata.ukho.gov.uk\u002F)。\n* [Sentinel2 Munich480](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fartelabsuper\u002Fsentinel2-munich480) -> 利用 Sentinel-2 卫星的时间序列进行作物制图的数据集。\n* [草地 vs 果园](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbaptistel\u002Fmeadows-vs-orchards) -> 一个像素时间序列数据集。\n* [用于作物制图的 Sentinel-2 影像时间序列](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fignazio\u002Fsentinel2-crop-mapping) -> 意大利伦巴第大区的数据。\n* [基于 Sentinel2 数据的乌克兰森林砍伐情况](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fisaienkov\u002Fdeforestation-in-ukraine)。\n* [satellite-change-events](https:\u002F\u002Fwww.cs.cornell.edu\u002Fprojects\u002Fsatellite-change-events\u002F) -> CaiRoad 和 CalFire 的变化检测 Sentinel 2 数据集。\n* [用于船舶检测的 Sentinel-2 数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F3923841)，也被编辑并重新发布为 [VDS2RAW](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7982468#.ZIiLxS8QOo4)。\n* [MineSegSAT](https:\u002F\u002Fgithub.com\u002Fmacdonaldezra\u002FMineSegSAT) -> 论文《利用 Sentinel-2 影像评估采矿扰动区域范围的自动化系统》所用的数据集。\n* [CaBuAr](https:\u002F\u002Fgithub.com\u002FDarthReca\u002FCaBuAr) -> 加州烧毁区域数据集，用于划定边界。\n* [sen12mscr](https:\u002F\u002Fpatricktum.github.io\u002Fcloud_removal\u002Fsen12mscr\u002F) -> 多模态云去除。\n* [Greenearthnet](https:\u002F\u002Fgithub.com\u002Fvitusbenson\u002Fgreenearthnet\u002Ftree\u002Fmain) -> 专为高分辨率植被预测设计的数据集。\n* [Floating-Marine-Debris-Data](https:\u002F\u002Fgithub.com\u002Fmiguelmendesduarte\u002FFloating-Marine-Debris-Data) -> 浮动海洋垃圾，附有六类垃圾的标注，包括塑料、漂流木、海藻、浮石、海涕和海沫。\n* [Sen2Fire](https:\u002F\u002Fzenodo.org\u002Frecords\u002F10881058) -> 使用 Sentinel 数据进行野火检测的具有挑战性的基准数据集。\n* [L1BSR](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7826696) -> 从两份 L1B 产品中提取的 3,740 对重叠图像裁剪。\n* [GloSoFarID](https:\u002F\u002Fgithub.com\u002Fyzyly1992\u002FGloSoFarID) -> 全球多光谱数据集，用于太阳能电站识别。\n* [MARIDA](https:\u002F\u002Fmarine-debris.github.io\u002Findex.html) -> 从 Sentinel-2 检测海洋垃圾。\n* [MADOS](https:\u002F\u002Fgithub.com\u002Fgkakogeorgiou\u002Fmados) -> 从 Sentinel-2 检测海洋垃圾和溢油。\n* [用于船舶检测和特征分析的 Sentinel-2 数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F10418786) -> RGB。\n* [S2-SHIPS](https:\u002F\u002Fgithub.com\u002Falina2204\u002Fcontrastive_SSL_ship_detection) -> 包含全部 12 个波段。\n* [ChatEarthNet](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FChatEarthNet) -> 一个全球规模的图像-文本数据集，赋能视觉-语言地理基础模型，利用 Sentinel-2 数据并结合 ChatGPT 生成的描述。\n* [UKFields](https:\u002F\u002Fgithub.com\u002FSpiruel\u002FUKFields) -> 超过 230 万个自动划定的田块边界，覆盖英格兰、威尔士、苏格兰和北爱尔兰。\n* [ShipWakes](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7947694) -> 基于关键点方法，利用深度学习识别 Sentinel-2 影像中的船迹成分。\n* [TimeSen2Crop](https:\u002F\u002Fzenodo.org\u002Frecords\u002F4715631) -> 用于作物类型分类的百万样本标记的 Sentinel 2 影像时间序列数据集。\n* [AgriSen-COG](https:\u002F\u002Fgithub.com\u002Ftselea\u002Fagrisen-cog) -> 一个多国、多时相的大规模 Sentinel-2 基准数据集，用于作物制图：包含异常检测预处理步骤。\n* [MagicBathyNet](https:\u002F\u002Fwww.magicbathy.eu\u002Fmagicbathynet.html) -> 一个新的多模态基准数据集，由 Sentinel-2、SPOT-6 和航空影像的图像补丁、栅格格式的水深数据以及海底类别标注组成。\n* [MuS2：用于 Sentinel-2 多图像超分辨率的基准](https:\u002F\u002Fdataverse.harvard.edu\u002Fdataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2F1JMRAT)。\n* [Sen4Map](https:\u002F\u002Fdatapub.fz-juelich.de\u002Fsen4map\u002F) -> Sentinel-2 时间序列影像，覆盖欧盟超过 335,125 个带有地理标签的位置。这些地理标签位置关联着详细的土地覆盖和土地利用信息。\n* [CloudSEN12Plus](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fisp-uv-es\u002FCloudSEN12Plus) -> 目前最大的用于 Sentinel-2 的云检测数据集。\n* [mayrajeo S2 船舶检测](https:\u002F\u002Fgithub.com\u002Fmayrajeo\u002Fship-detection) -> 使用 YOLOv8 从 Sentinel-2 影像中检测海上船只的标签。\n* [Fields of The World](https:\u002F\u002Ffieldsofthe.world\u002F) -> 农业田块边界的实例分割。\n* [ai4boundaries](https:\u002F\u002Fgithub.com\u002Fwaldnerf\u002Fai4boundaries) -> 结合 Sentinel-2 和航空摄影进行田块边界检测。\n* [加州野火地理影像数据集 - CWGID](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.16380) -> 开发并应用基于 Sentinel-2 卫星影像的深度学习驱动森林野火检测数据集。\n* [substation-seg](https:\u002F\u002Fgithub.com\u002FLindsay-Lab\u002Fsubstation-seg) -> 变电站分割数据集。\n* [PhilEO-downstream](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FPhilEO-community\u002FPhilEO-downstream) -> 一个 400GB 的 Sentinel-2 数据集，用于建筑密度估计、道路分割和土地覆盖分类。\n* [PhilEO-pretrain](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FPhilEO-community\u002FPhilEO-pretrain) -> 一个 500GB 的全球 Sentinel-2 影像数据集，用于模型预训练。\n* [KappaSet：Sentinel-2 KappaZeta 云和云影掩码](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7100327)。\n* [AllClear](https:\u002F\u002Fallclear.cs.cornell.edu\u002F) 一个全面的卫星影像去云数据集和基准。\n* [通过主动学习方法生成的 Sentinel-2 参考云掩码](https:\u002F\u002Fzenodo.org\u002Frecords\u002F1460961)。\n* [利用深度学习填补云隙以改善草原监测](https:\u002F\u002Fzenodo.org\u002Frecords\u002F11651601)。\n* [遥感船迹数据集](https:\u002F\u002Fgithub.com\u002Fzjze\u002FRSSW_Dateset)。\n* [ERAS-dataset](https:\u002F\u002Fgithub.com\u002Fcscribano\u002FERAS-dataset) -> 艾米利亚-罗马涅地区农业分割（ERAS）田块分割数据集。\n* [瑞士两个地区五年内 92 场景的 Sentinel 2 超分辨率数据立方体](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241003205022\u002Fhttps:\u002F\u002Fieee-dataport.org\u002Fdocuments\u002Fsentinel-2-super-resolved-data-cubes-92-scenes-over-2-regions-switzerland-spanning-5-years)。\n* [SeasoNet](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6979994) -> 一个针对德国卫星影像的季节性场景分类、分割和检索数据集。土地覆盖类别基于 CORINE Land Cover 数据库（CLC）2018 年版本。\n* [EuroCropsML](https:\u002F\u002Fgithub.com\u002Fdida-do\u002Feurocropsml) -> 一个即用型基准数据集，用于使用 Sentinel-2 影像进行少样本作物类型分类。\n* [CanadaFireSat](https:\u002F\u002Fgithub.com\u002Feceo-epfl\u002FCanadaFireSat-Data) -> Sentinel-2 Level-1C 时间序列。\n* [ssl4eco](https:\u002F\u002Fgithub.com\u002FPlekhanovaElena\u002Fssl4eco) -> 一种构建预训练数据集的配方，能够捕捉全球生态系统的地理和物候多样性。\n* [IRRISIGHT](https:\u002F\u002Fgithub.com\u002FNibir088\u002FIRRISIGHT) -> 一个大规模的多模态遥感数据集，用于灌溉分类、土壤-水分映射和农业监测。\n* [SentinelKilnDB](https:\u002F\u002Fsustainability-lab.github.io\u002Fsentinelkilndb\u002F) -> 用于监测南亚砖窑排放的 Sentinel-2 数据集。\n* [MSSWD - 多光谱船迹数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13870226)。\n* [MOSAIC-SEN2-CC](https:\u002F\u002Fgithub.com\u002FChangeCapsInRS\u002FMOSAIC-SEN2-CC) -> 一个用于遥感变化标注的多光谱数据集和适配框架。\n* [PLUTo](https:\u002F\u002Fzenodo.org\u002Frecords\u002F15629667) -> 热带地区森林砍伐后的土地利用情况。\n* [SentinelKilnDB](https:\u002F\u002Fgithub.com\u002Frishabh-mondal\u002FSENTINELKILNDB_NeurIPS_2025) -> 一个大型数据集和基准，用于在南亚利用卫星影像检测定向包围盒（OBB）形式的砖窑。\n* [GSDD](https:\u002F\u002Fzenodo.org\u002Frecords\u002F17161810) -> 全球冰川表面碎屑数据集。\n* [MT4AFE](https:\u002F\u002Fzenodo.org\u002Frecords\u002F15395167) -> 多任务学习用于农田提取。\n* [agripotential](https:\u002F\u002Fgithub.com\u002FMohammadElSakka\u002Fagripotential) -> 一个包含 34 个 Sentinel-2 时间帧的卫星影像时间序列（STIS），涵盖 5 类农业潜力。\n* [YieldSAT](https:\u002F\u002Fyieldsat.github.io\u002F) -> 一个用于高分辨率作物产量预测的多模态基准数据集。\n\n### 综合哨兵\n* [awesome-sentinel](https:\u002F\u002Fgithub.com\u002FFernerkundung\u002Fawesome-sentinel) -> 一个精选的列表，包含与哥白尼哨兵卫星数据相关的优秀工具、教程和API。\n* 通过[sentinel-hub](https:\u002F\u002Fwww.sentinel-hub.com\u002F)和[python-api](https:\u002F\u002Fgithub.com\u002Fsentinel-hub\u002Fsentinelhub-py)可付费访问哨兵和Landsat数据。\n* [用于处理存储在S3上的Sentinel-5P Level 2数据的Jupyter Notebooks](https:\u002F\u002Fgithub.com\u002FSentinel-5P\u002Fdata-on-s3)。数据可在[这里](https:\u002F\u002Fmeeo-s5p.s3.amazonaws.com\u002Findex.html#\u002F?t=catalogs)浏览。\n* [哨兵NetCDF数据](https:\u002F\u002Fgithub.com\u002Facgeospatial\u002FSentinel-5P\u002Fblob\u002Fmaster\u002FSentinel_5P.ipynb)。\n* [earthspy](https:\u002F\u002Fgithub.com\u002FAdrienWehrle\u002Fearthspy) -> 利用Sinergise公司EO研究团队开发的哨兵Hub服务，实时（NRT）监测和研究地球上任何地点。\n* [金矿开采与秘密简易机场数据集](https:\u002F\u002Fgithub.com\u002Fearthrise-media\u002Fmining-detector)。\n* [工业烟 plume](https:\u002F\u002Fzenodo.org\u002Frecord\u002F4250706)。\n* [MARIDA：海洋垃圾档案](https:\u002F\u002Fgithub.com\u002Fmarine-debris\u002Fmarine-debris.github.io)。\n* [OMS2CD](https:\u002F\u002Fgithub.com\u002FDibz15\u002FOpenMineChangeDetection) -> 用于露天矿区变化检测的手工标注图像。\n* [煤电厂排放数据](https:\u002F\u002Ftransitionzero.medium.com\u002Festimating-coal-power-plant-operation-from-satellite-images-with-computer-vision-b966af56919e) -> 包含图像、元数据和标签的煤电厂排放数据集。\n* [RapidAI4EO](https:\u002F\u002Frapidai4eo.radiant.earth\u002F) -> 欧洲范围内50万个地点采样的密集时间序列卫星影像，包括S2和Planet影像，并附有2018年CORINE土地覆盖多类别标签。\n* [MS-HS-BCD-dataset](https:\u002F\u002Fgithub.com\u002Farcgislearner\u002FMS-HS-BCD-dataset) -> 多源变化检测数据集，用于论文《基于深度学习融合多源遥感影像光谱与纹理特征的建筑物变化检测：以GF-1和Sentinel 2B数据为例》。\n* [CropNet：面向气候变化的作物产量预测的开放大型多模态数据集](https:\u002F\u002Fgithub.com\u002Ffudong03\u002FCropNet) -> 大量公开可用的多模态数据集，用于支持气候变化背景下的作物产量预测。\n* [Tiny CropNet数据集](https:\u002F\u002Fgithub.com\u002Ffudong03\u002FMMST-ViT)。\n* [利用卫星影像估算发电厂温室气体排放的多任务学习](https:\u002F\u002Fzenodo.org\u002Frecord\u002F5644746)。\n* [METER-ML：用于自动化甲烷源测绘的多传感器地球观测基准数据集](https:\u002F\u002Fstanfordmlgroup.github.io\u002Fprojects\u002Fmeter-ml\u002F) -> 数据可在[Zenodo](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6911013)上找到。\n* [MultiSenGE](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6375466) -> 大规模多模态、多时相基准数据集。\n* [SEN12MS](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSEN12MS) -> 用于深度学习和数据融合的地理参考多光谱哨兵1\u002F2影像精选数据集。可查看[SEN12MS工具箱](https:\u002F\u002Fgithub.com\u002Fschmitt-muc\u002FSEN12MS)，并在[paperswithcode.com](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fsen12ms)上找到大量引用案例。\n* [Space2Ground](https:\u002F\u002Fgithub.com\u002FAgri-Hub\u002FSpace2Ground) -> 包含太空（哨兵1\u002F2）和地面（街景图像）两部分的数据集，标注了作物类型标签，用于农业监测。\n* [MSCDUnet](https:\u002F\u002Fgithub.com\u002FLihy256\u002FMSCDUnet) -> 包含VHR、多光谱（哨兵2）和SAR（哨兵1）影像的变化检测数据集。\n* [OMBRIA](https:\u002F\u002Fgithub.com\u002Fgeodrak\u002FOMBRIA) -> 用于解决洪水制图问题的哨兵1和2数据集。\n* [卫星烧毁区域数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6597139#.Y9ufiezP1hE) -> 包含与过去森林火灾相关的多颗卫星影像的分割数据集。其中包括来自哨兵2和哨兵1（哥白尼计划）的73幅影像。\n* [SEN12_GUM](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6914898) -> 哨兵12全球城市制图数据集。\n* [哨兵1和2影像对（SAR与光学）](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Frequiemonk\u002Fsentinel12-image-pairs-segregated-by-terrain)。\n* [MSOSCD](https:\u002F\u002Fgithub.com\u002FLihy256\u002FMSCDUnet) -> 包含VHR、多光谱（哨兵2）和SAR（哨兵1）影像的变化检测数据集。\n* [SICKLE](https:\u002F\u002Fgithub.com\u002FDepanshu-Sani\u002FSICKLE) -> 一个标注了多种关键种植参数的多传感器卫星影像数据集。包含来自Landsat-8、哨兵1和哨兵2的多分辨率时序影像。\n* [哨兵1和2船只检测](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fvessel-detection-sentinels)。\n* [TreeSatAI](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6780578) -> 哨兵1、哨兵2。\n* [AI2-S2-NAIP](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fallenai\u002Fs2-naip) -> 对齐后的NAIP、哨兵2、哨兵1和Landsat影像，覆盖整个美国大陆。\n* [POPCORN：基于哨兵1和哨兵2生成的高分辨率人口地图](https:\u002F\u002Fpopcorn-population.github.io\u002F)。\n* [CropClimateX](https:\u002F\u002Fgithub.com\u002Fdrnhhl\u002FCropClimateX) -> 用于极端气候条件下作物监测的大规模多任务、多传感数据集。\n* [SmallMinesDS](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fellaampy\u002FSmallMinesDS) -> 用于绘制手工及小型金矿的地图的多模态数据集。该数据集中的影像也被重新用于[CocoaMiningDS](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fellaampy\u002FCocoaMiningDS)。\n* [Hoss-ReID](https:\u002F\u002Fgithub.com\u002FAlioth2000\u002FHoss-ReID) -> 通过光学和SAR影像进行跨模态船舶再识别。\n* [IDEABench基准数据集](https:\u002F\u002Fgithub.com\u002FIDEAtlas\u002Fai-dua-mapping) -> 针对全球城市样本的城市贫困状况测绘与基准测试。\n* [ImpactMesh](https:\u002F\u002Fgithub.com\u002FIBM\u002FImpactMesh) -> 用于洪水和野火制图的大规模多模态、多时相数据集。\n* [Sen12Landslides](https:\u002F\u002Fgithub.com\u002FPaulH97\u002FSen12Landslides) -> 空间-时间滑坡与异常检测数据集。\n* [Cryo-Bench](https:\u002F\u002Fgithub.com\u002FSk-2103\u002FCryo-Bench) -> 用于评估地理空间基础模型在冰冻圈应用中的基准。\n* [BigEarthNet.txt](https:\u002F\u002Ftxt.bigearth.net\u002F) -> 用于地球观测的大规模多传感器图像-文本数据集及基准。\n* [婆罗洲森林扰动数据集](https:\u002F\u002Fgithub.com\u002FColmKeyes\u002FBorneo_Forest_Disturbance_Dataset) -> 利用哨兵2数据和RADD警报扰动信息的森林扰动数据集。\n\n## Landsat\n美国长期运行的卫星计划 -> 参见 [维基百科](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FLandsat_program)\n* 8个波段，分辨率15至60米，幅宽185公里，时间分辨率16天\n* [Google上的Landsat 4、5、7和8影像](https:\u002F\u002Fcloud.google.com\u002Fstorage\u002Fdocs\u002Fpublic-datasets\u002Flandsat)，详见[GCP存储桶](https:\u002F\u002Fconsole.cloud.google.com\u002Fstorage\u002Fbrowser\u002Fgcp-public-data-landsat\u002F)，其中Landsat 8影像采用COG格式，并在[此笔记本](https:\u002F\u002Fgithub.com\u002Fpangeo-data\u002Fpangeo-example-notebooks\u002Fblob\u002Fmaster\u002Flandsat8-cog-ndvi.ipynb)中进行了NDVI分析\n* [AWS上的Landsat 8影像](https:\u002F\u002Fregistry.opendata.aws\u002Flandsat-8\u002F)，附带大量教程和工具\n* https:\u002F\u002Fgithub.com\u002Fkylebarron\u002Flandsat-mosaic-latest -> 基于AWS SNS通知自动更新的无云Landsat 8镶嵌图\n* [使用Datashader可视化Landsat影像](https:\u002F\u002Fexamples.pyviz.org\u002Flandsat\u002Flandsat.html#landsat-gallery-landsat)\n* [Landsat-mosaic-tiler](https:\u002F\u002Fgithub.com\u002Fkylebarron\u002Flandsat-mosaic-tiler) -> 该仓库托管了landsatlive.live网站及API的所有代码。\n* [LandsatSCD](https:\u002F\u002Fgithub.com\u002FggsDing\u002FSCanNet\u002Ftree\u002Fmain) -> 一种变化检测数据集，包含8468对图像，每对图像的空间分辨率为416 × 416\n* [爱尔兰海岸线分割Landsat数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F8414665)\n* [Wildfire-Spread-Dataset](https:\u002F\u002Fgithub.com\u002FBEEILAB\u002FWildfire-Spread-Dataset) -> ABNextFire：基于多源深度学习的野火蔓延预测数据集\n\n## VENμS\n新型微型卫星上的植被与环境监测（[VENμS](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FVEN%CE%BCS)）\n* [VENUS L2A云优化GeoTIFF文件](https:\u002F\u002Fregistry.opendata.aws\u002Fvenus-l2a-cogs\u002F)\n* [VENuS云掩膜训练数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7040177)\n* [Sen2Venµs](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6514159) -> 用于训练Sentinel-2超分辨率算法的数据集\n* [sen2venus-pytorch-dataset](https:\u002F\u002Fgithub.com\u002Fpiclem\u002Fsen2venus-pytorch-dataset) -> PyTorch数据加载器及其他实用工具\n\n## Vantor\nVantor公司（前身为Maxar & DigitalGlobe）拥有的卫星包括[GeoEye-1](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FGeoEye-1)、[WorldView-2](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWorldView-2)、[3号](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWorldView-3)和[4号](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWorldView-4)\n* [Maxar开放数据计划](https:\u002F\u002Fgithub.com\u002Fopengeos\u002Fmaxar-open-data)提供事件发生前后高分辨率卫星影像，以支持应急规划、响应、损失评估和灾后恢复\n* [WorldView-2欧洲城市影像](https:\u002F\u002Fearth.esa.int\u002Feogateway\u002Fcatalog\u002Fworldview-2-european-cities) -> 覆盖欧洲人口最密集地区的数据集，分辨率为40厘米\n\n## Planet\n另请参阅本页后续的Spacenet-7以及Kaggle上的船舶和飞机分类数据集\n* [Planet提供的全球热带地区高分辨率、可直接分析的镶嵌影像](https:\u002F\u002Fwww.planet.com\u002Fnicfi\u002F)，由挪威国际气候与森林倡议支持。[BBC报道](https:\u002F\u002Fwww.bbc.co.uk\u002Fnews\u002Fscience-environment-54651453)\n* Planet曾通过Kaggle竞赛提供影像数据\n* [阿尔伯塔油井数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13743323) -> 从卫星影像中精确定位油气井\n* [ARGO船舶分类数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6058710) -> 来自PlanetScope四波段卫星的1750张标注图像。创建于[此处](https:\u002F\u002Fgithub.com\u002Felizamanelli\u002Fship_dataset\u002Ftree\u002Fmain)\n* [PlanetScope影像中的海洋垃圾检测数据集](https:\u002F\u002Fcmr.earthdata.nasa.gov\u002Fsearch\u002Fconcepts\u002FC2781412735-MLHUB.html)\n* [LitterLines](https:\u002F\u002Fgithub.com\u002FgeoJoost\u002FLitterLines) -> 用于检测PlanetScope影像中海洋垃圾堆积带的标注数据集\n* [FloodPlanet洪水淹没数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F15238572) -> 多传感器配准数据集，基于3米分辨率的PlanetScope数据进行标注，并与Sentinel-1、Sentinel-2和Landsat-8数据在空间上重叠、时间上接近\n* [Zhijie_FloodPlanet_2023](https:\u002F\u002Fdatacommons.cyverse.org\u002Fbrowse\u002Fiplant\u002Fhome\u002Fshared\u002Fcommons_repo\u002Fcurated\u002FZhijie_FloodPlanet_2023) -> 包含2017年至2020年间发生的19次洪水事件\n\n## UC Merced\n土地利用分类数据集，包含21个类别，每个类别有100张RGB TIFF图像。每张图像尺寸为256×256像素，像素分辨率为1英尺\n* http:\u002F\u002Fweegee.vision.ucmerced.edu\u002Fdatasets\u002Flanduse.html\n* 同时也[以多标签数据集形式提供](https:\u002F\u002Ftowardsdatascience.com\u002Fmulti-label-land-cover-classification-with-deep-learning-d39ce2944a3d)\n* 阅读[用于遥感图像分类的视觉Transformer](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F13\u002F3\u002F516\u002Fhtm)，其中Vision Transformer分类器在Merced数据集上达到了98.49%的分类准确率\n\n## EuroSAT\nSentinel-2卫星影像的土地利用分类数据集，覆盖13个光谱波段，包含10个类别，共27000个已标注且地理参考的样本。提供RGB版本和13波段版本\n* [EuroSAT：使用Sentinel-2进行土地利用与地表覆盖分类](https:\u002F\u002Fgithub.com\u002Fphelber\u002FEuroSAT) -> 一篇发表论文，其中CNN模型实现了98.57%的分类准确率\n* 使用fastai的仓库[这里](https:\u002F\u002Fgithub.com\u002Fshakasom\u002FDeep-Learning-for-Satellite-Imagery)和[这里](https:\u002F\u002Fwww.luigiselmi.eu\u002Feo\u002Flulc-classification-deeplearning.html)\n* [evolved_channel_selection](http:\u002F\u002Fmatpalm.com\u002Fblog\u002Fevolved_channel_selection\u002F) -> 探讨混合分辨率与是否使用某个波段之间的权衡，并附有[仓库](https:\u002F\u002Fgithub.com\u002Fmatpalm\u002Fevolved_channel_selection)\n* RGB版本可在[PyTorch数据集中找到](https:\u002F\u002Fpytorch.org\u002Fvision\u002Fstable\u002Fgenerated\u002Ftorchvision.datasets.EuroSAT.html#torchvision.datasets.EuroSAT)，而13波段版本则可在[torchgeo中找到](https:\u002F\u002Ftorchgeo.readthedocs.io\u002Fen\u002Flatest\u002Fapi\u002Fdatasets.html#eurosat)。请查看关于使用该数据集进行[数据增强的教程](https:\u002F\u002Ftorchgeo.readthedocs.io\u002Fen\u002Flatest\u002Ftutorials\u002Ftransforms.html)\n* [EuroSAT-SAR](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fwangyi111\u002FEuroSAT-SAR) -> 根据地理坐标将EuroSAT中的每张Sentinel-2图像与一张Sentinel-1影像匹配\n\n## PatternNet\n土地利用分类数据集，包含38个类别，每个类别有800张RGB JPG图像\n* https:\u002F\u002Fsites.google.com\u002Fview\u002Fzhouwx\u002Fdataset?authuser=0\n* 发表论文：[PatternNet：用于遥感图像检索性能评估的基准数据集](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.03424)\n\n## Gaofen图像数据集（GID）用于分类\n- https:\u002F\u002Fcaptain-whu.github.io\u002FGID\u002F\n- 一个大规模分类数据集和一个精细地表覆盖分类数据集\n\n## Million-AID\n一个包含百万级样本的大规模基准数据集，用于遥感场景分类，共51个场景类别，按层次化类别组织。\n* https:\u002F\u002Fcaptain-whu.github.io\u002FDiRS\u002F\n* [预训练模型](https:\u002F\u002Fgithub.com\u002FViTAE-Transformer\u002FViTAE-Transformer-Remote-Sensing)\n* 另请参阅 [AID](https:\u002F\u002Fcaptain-whu.github.io\u002FAID\u002F)、[AID多标签数据集](https:\u002F\u002Fgithub.com\u002FHua-YS\u002FAID-Multilabel-Dataset) 和 [DFC15多标签数据集](https:\u002F\u002Fgithub.com\u002FHua-YS\u002FDFC15-Multilabel-Dataset)\n\n## DIOR目标检测数据集\n一个用于光学遥感图像中目标检测的大规模基准数据集，包含23,463张图像和192,518个由水平边界框标注的目标实例。\n* https:\u002F\u002Fgcheng-nwpu.github.io\u002F\n* https:\u002F\u002Farxiv.org\u002Fabs\u002F1909.00133\n* [ors-detection](https:\u002F\u002Fgithub.com\u002FVlad15lav\u002Fors-detection) -> 使用YOLOv3在DIOR数据集上进行目标检测\n* [dior_detect](https:\u002F\u002Fgithub.com\u002Fhm-better\u002Fdior_detect) -> DIOR数据集上的目标检测基准测试\n* [Tools](https:\u002F\u002Fgithub.com\u002FCrazyStoneonRoad\u002FTools) -> 用于处理DIOR数据集的工具\n* [Object_Detection_Satellite_Imagery_Yolov8_DIOR](https:\u002F\u002Fgithub.com\u002FJohnPPinto\u002FObject_Detection_Satellite_Imagery_Yolov8_DIOR)\n\n## Multiscene\nMultiScene数据集旨在解决两个任务：开发多场景识别算法以及在带噪声标签的数据上进行网络学习。\n* https:\u002F\u002Fmultiscene.github.io\u002F 和 https:\u002F\u002Fgithub.com\u002FHua-YS\u002FMulti-Scene-Recognition\n\n## FAIR1M目标检测数据集\n一个用于高分辨率遥感图像中细粒度目标识别的基准数据集。\n* [arXiv论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.05569)\n* 可从gaofen-challenge.com下载\n* [2020Gaofen](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002F2020Gaofen) -> 2020年高分挑战赛的数据、基线和评估指标\n\n## DOTA目标检测数据集\n一个用于航空图像中目标检测的大规模基准及挑战赛。分割标注可在iSAID数据集中找到。\n* https:\u002F\u002Fcaptain-whu.github.io\u002FDOTA\u002Findex.html\n* [DOTA_devkit](https:\u002F\u002Fgithub.com\u002FCAPTAIN-WHU\u002FDOTA_devkit) 用于加载数据集\n* [arXiv论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1711.10398)\n* [mmrotate中的预训练模型](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmrotate)\n* [DOTA2VOCtools](https:\u002F\u002Fgithub.com\u002FComplicateddd\u002FDOTA2VOCtools) -> 数据集拆分并转换为VOC格式\n* [dotatron](https:\u002F\u002Fgithub.com\u002Fnaivelogic\u002Fdotatron) -> 2021年基于DOTA数据集的“学习理解航空图像”挑战赛\n\n## iSAID实例分割数据集\n一个用于航空图像中实例分割的大规模数据集。\n* https:\u002F\u002Fcaptain-whu.github.io\u002FiSAID\u002Fdataset.html\n* 使用了DOTA数据集中的图像。\n\n## HRSC RGB船舶目标检测数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fguofeng\u002Fhrsc2016\n* [mmrotate中的预训练模型](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmrotate)\n* [Rotation-RetinaNet-PyTorch](https:\u002F\u002Fgithub.com\u002FHsLOL\u002FRotation-RetinaNet-PyTorch)\n\n## SAR船舶检测数据集（SSDD）\n* https:\u002F\u002Fgithub.com\u002FTianwenZhang0825\u002FOfficial-SSDD\n* [Rotation-RetinaNet-PyTorch](https:\u002F\u002Fgithub.com\u002FHsLOL\u002FRotation-RetinaNet-PyTorch)\n\n## 高分辨率SAR旋转船舶检测数据集（SRSDD）\n* [GitHub](https:\u002F\u002Fgithub.com\u002FHeuristicLU\u002FSRSDD-V1.0)\n* [一种用于复杂场景SAR图像中船舶检测与识别的轻量级模型](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F14\u002F23\u002F6053)\n\n## LEVIR船舶数据集\n一个用于中等分辨率遥感图像下微小船舶检测的数据集。标注采用边界框格式。\n* [LEVIR-Ship](https:\u002F\u002Fgithub.com\u002FWindVChen\u002FLEVIR-Ship)\n\u003C!-- markdown-link-check-disable -->\n* 托管于[Nucleus](https:\u002F\u002Fdashboard.scale.com\u002Fnucleus\u002Fds_cbsghny30nf00b1x3w7g?utm_source=open_dataset&utm_medium=github&utm_campaign=levir_ships)\n\u003C!-- markdown-link-check-enable -->\n\n## SAR飞机检测数据集\n收集了2966个不重叠的224×224切片，包含7835个飞机目标。\n* https:\u002F\u002Fgithub.com\u002Fhust-rslab\u002FSAR-aircraft-data\n\n## xView1：航拍影像中的上下文对象\n一个细粒度的目标检测数据集，包含60个对象类别，涵盖8种类别的本体结构。超过100万个对象分布在超过1,400平方公里的0.3米分辨率影像中。标注采用边界框格式。\n* [官方网站](http:\u002F\u002Fxviewdataset.org\u002F)\n* [arXiv论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.07856)\n* [paperswithcode](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fxview)\n* [Satellite_Imagery_Detection_YOLOV7](https:\u002F\u002Fgithub.com\u002FRadhika-Keni\u002FSatellite_Imagery_Detection_YOLOV7) -> 将YOLOV7应用于xView1\n\n## xView2：xBD建筑物损毁评估\n一个用于建筑物损毁评估的高分辨率卫星影像标注数据集，提供精确的分割掩码和四级损伤标签，影像分辨率为0.3米。\n* [官方网站](https:\u002F\u002Fxview2.org\u002F)\n* [arXiv论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1911.09296)\n* [paperswithcode](https:\u002F\u002Fpaperswithcode.com\u002Fpaper\u002Fxbd-a-dataset-for-assessing-building-damage)\n* [xView2_baseline](https:\u002F\u002Fgithub.com\u002FDIUx-xView\u002FxView2_baseline) -> TensorFlow中的基线解决方案\n* [metadamagenet](https:\u002F\u002Fgithub.com\u002Fnimaafshar\u002Fmetadamagenet) -> PyTorch解决方案\n* [来自michal2409的U-Net模型](https:\u002F\u002Fgithub.com\u002Fmichal2409\u002FxView2)\n* [DAHiTra](https:\u002F\u002Fgithub.com\u002Fnka77\u002FDAHiTra) -> 2022年[论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2208.02205)的代码：利用新型分层Transformer架构对卫星图像进行大规模建筑物损毁评估。使用xView2 xBD数据集。\n* [使用Amazon SageMaker地理空间功能和自定义SageMaker模型进行损毁评估](https:\u002F\u002Faws.amazon.com\u002Fblogs\u002Fmachine-learning\u002Fdamage-assessment-using-amazon-sagemaker-geospatial-capabilities-and-custom-sagemaker-models\u002F)\n* [Xview2_Strong_Baseline](https:\u002F\u002Fgithub.com\u002FPaulBorneP\u002FXview2_Strong_Baseline) -> 强基线的简单实现\n\n## xView3：SAR影像中的暗色船只检测\n在合成孔径雷达（SAR）影像中检测从事非法、未报告和无管制（IUU）捕捞活动的暗色船只。该多模态数据集包含人类和算法标注的船只及固定设施实例，覆盖43,200,000平方公里的Sentinel-1影像，使算法能够检测并分类暗色船只。\n* [官方网站](https:\u002F\u002Fiuu.xview.us\u002F)\n* [arXiv论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2206.00897)\n* [GitHub](https:\u002F\u002Fgithub.com\u002FDIUx-xView) -> 包含所有参考代码、数据处理工具以及获奖模型的代码和权重\n* [paperswithcode](https:\u002F\u002Fpaperswithcode.com\u002Fdataset\u002Fxview3-sar)\n* [xview3_ship_detection](https:\u002F\u002Fgithub.com\u002Fnaivelogic\u002Fxview3_ship_detection)\n\n## 航空影像中的车辆检测（VEDAI）\n航空影像中的车辆检测。标注采用边界框格式。\n* https:\u002F\u002Fdownloads.greyc.fr\u002Fvedai\u002F\n* [pytorch-vedai](https:\u002F\u002Fgithub.com\u002FMichelHalmes\u002Fpytorch-vedai)\n\n## 俯视视角车辆数据集（COWC）\n包含大量标注的俯视视角车辆图像。为目标检测和计数任务提供了基准数据集。标注采用边界框格式。\n* http:\u002F\u002Fgdo152.ucllnl.org\u002Fcowc\u002F\n* https:\u002F\u002Fgithub.com\u002FLLNL\u002Fcowc\n* [利用航空影像检测车辆以应对北约创新挑战](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241110114250\u002Fhttps:\u002F\u002Farthurdouillard.com\u002Fpost\u002Fnato-challenge\u002F)\n* [LINZ 和 UGRC](https:\u002F\u002Fgithub.com\u002Fhumansensinglab\u002FAGenDA\u002Ftree\u002Fmain\u002FData)\n\n## AI-TOD & AI-TOD-v2 - 微小目标检测\nAI-TOD 数据集中物体的平均尺寸约为 12.8 像素，远小于其他数据集。标注采用边界框格式。v2 是对 v1 数据集进行细致重新标注的结果。\n* https:\u002F\u002Fgithub.com\u002Fjwwangchn\u002FAI-TOD\n* https:\u002F\u002Fchasel-tsui.github.io\u002FAI-TOD-v2\u002F\n* [NWD](https:\u002F\u002Fgithub.com\u002Fjwwangchn\u002FNWD) -> 用于 2021 年 [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2110.13389) 的代码：一种用于微小目标检测的归一化高斯 Wasserstein 距离。使用 AI-TOD 数据集。\n* [ORFENet](https:\u002F\u002Fgithub.com\u002Fdyl96\u002FORFENet) -> 基于目标重建与多感受野自适应特征增强的遥感图像微小目标检测。使用 LEVIR-ship 和 AI-TOD-v2。\n\n## RarePlanes\n* [RarePlanes](https:\u002F\u002Fregistry.opendata.aws\u002Frareplanes\u002F) -> 包含真实和合成生成的卫星影像，其中包括飞机。阅读 [arXiv 论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2006.02963) 并查看 [此仓库](https:\u002F\u002Fgithub.com\u002Fjdc08161063\u002FRarePlanes)。请注意，该数据集可通过 AWS 开放数据计划免费下载。\n* [理解 RarePlanes 数据集并构建飞机检测模型](https:\u002F\u002Fencord.com\u002Fblog\u002Frareplane-dataset-aircraft-detection-model\u002F) -> 博客文章。\n* 阅读 [NVIDIA 的这篇文章](https:\u002F\u002Fdeveloper.nvidia.com\u002Fblog\u002Fpreparing-models-for-object-detection-with-real-and-synthetic-data-and-tao-toolkit\u002F)，其中讨论了如何用 10% 的真实数据对基于合成数据（Rareplanes）预训练的模型进行微调，然后通过剪枝减少模型大小，最后量化模型以提高推理速度。\n* [yoltv4](https:\u002F\u002Fgithub.com\u002Favanetten\u002Fyoltv4) 包含使用 [RarePlanes 数据集](https:\u002F\u002Fregistry.opendata.aws\u002Frareplanes\u002F) 的示例。\n* [rareplanes-yolov5](https:\u002F\u002Fgithub.com\u002Fjeffaudi\u002Frareplanes-yolov5) -> 使用 YOLOv5 和 RarePlanes 数据集来检测并分类飞机的子特征，并附有 [文章](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Fdetecting-aircrafts-on-airbus-pleiades-imagery-with-yolov5-5f3d464b75ad)。\n\n## Counting from Sky\n用于遥感目标计数的大规模数据集及基准方法\n* https:\u002F\u002Fgithub.com\u002Fgaoguangshuai\u002FCounting-from-Sky-A-Large-scale-Dataset-for-Remote-Sensing-Object-Counting-and-A-Benchmark-Method\n\n## AIRS（用于屋顶分割的航空影像）\n公开数据集，用于从超高分辨率航空影像（7.5cm）中进行屋顶分割。覆盖新西兰南岛最大城市克赖斯特彻奇的几乎全部区域。\n* [在 Kaggle 上](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fatilol\u002Faerialimageryforroofsegmentation)\n* [Rooftop-Instance-Segmentation](https:\u002F\u002Fgithub.com\u002FMasterSkepticista\u002FRooftop-Instance-Segmentation) -> 使用 VGG-16 进行实例分割，数据来源为 Airs 数据集。\n\n## Inria 建筑物\u002F非建筑物分割数据集\n空间分辨率为 0.3 m 的 RGB GeoTIFF 文件。数据涵盖奥斯汀、芝加哥、基茨普县、西蒂罗尔和东蒂罗尔、因斯布鲁克、旧金山和维也纳。\n* https:\u002F\u002Fproject.inria.fr\u002Faerialimagelabeling\u002Fcontest\u002F\n* [SemSegBuildings](https:\u002F\u002Fgithub.com\u002FSharpestProjects\u002FSemSegBuildings) -> 使用 fast.ai 框架对 Inria 建筑物分割数据集进行语义分割的项目。\n* [UNet_keras_for_RSimage](https:\u002F\u002Fgithub.com\u002Floveswine\u002FUNet_keras_for_RSimage) -> 用于二值语义分割的 Keras 代码。\n\n## AICrowd Mapping Challenge：建筑物分割数据集\n300×300 像素的 RGB 图像，标注采用 COCO 格式。影像似乎覆盖全球，但北美地区的比例较大。\n* 作为 [mapping-challenge](https:\u002F\u002Fwww.aicrowd.com\u002Fchallenges\u002Fmapping-challenge) 的一部分发布的数据集。\n* neptune.ai 发布的获胜方案 [这里](https:\u002F\u002Fgithub.com\u002Fneptune-ai\u002Fopen-solution-mapping-challenge)，使用带有 Resnet 的 Unet 实现了 0.943 的精确率和 0.954 的召回率。\n* [mappingchallenge](https:\u002F\u002Fgithub.com\u002Fkrishanr\u002Fmappingchallenge) -> 将 YOLOv5 应用于 AICrowd Mapping Challenge 数据集。\n\n## BONAI - 建筑物轮廓数据集\nBONAI（斜视角航空影像中的建筑物）是一个用于从斜视角航空影像中提取建筑物轮廓（BFE）的数据集。\n* https:\u002F\u002Fgithub.com\u002Fjwwangchn\u002FBONAI\n\n## LEVIR-CD 建筑物变化检测数据集\n* https:\u002F\u002Fjustchenhao.github.io\u002FLEVIR\u002F\n* [FCCDN_pytorch](https:\u002F\u002Fgithub.com\u002Fchenpan0615\u002FFCCDN_pytorch) -> 使用 PyTorch 实现 FCCDN 算法进行变化检测任务。\n* [RSICC](https:\u002F\u002Fgithub.com\u002FChen-Yang-Liu\u002FRSICC) -> 遥感图像变化描述数据集使用 LEVIR-CD 影像。\n\n## Onera（OSCD）Sentinel-2 变化检测数据集\n该数据集由 2015 年至 2018 年间从 Sentinel-2 卫星拍摄的 24 对多光谱图像组成。\n* [Onera 卫星变化检测数据集](https:\u002F\u002Fieee-dataport.org\u002Fopen-access\u002Foscd-onera-satellite-change-detection) 包括 2015 年至 2018 年间从 Sentinel-2 卫星拍摄的 24 对多光谱图像。\n* [网站](https:\u002F\u002Frcdaudt.github.io\u002Foscd\u002F)\n* [change_detection_onera_baselines](https:\u002F\u002Fgithub.com\u002Fprevitus\u002Fchange_detection_onera_baselines) -> Siamese 版本的 U-Net 基准模型。\n* [使用卷积神经网络进行多光谱地球观测的城市变化检测](https:\u002F\u002Fgithub.com\u002Frcdaudt\u002Fpatch_based_change_detection) -> 附有 [论文](https:\u002F\u002Fieeexplore.ieee.org\u002Fabstract\u002Fdocument\u002F8518015)。\n* [DS_UNet](https:\u002F\u002Fgithub.com\u002FSebastianHafner\u002FDS_UNet) -> 2021 年论文的代码：利用双流 U-Net 进行 Sentinel-1 和 Sentinel-2 数据融合以实现城市变化检测，使用 Onera 卫星变化检测数据集。\n* [ChangeDetection_wOnera](https:\u002F\u002Fgithub.com\u002Ftonydp03\u002FChangeDetection_wOnera)。\n* [OSCD + 额外日期](https:\u002F\u002Fgithub.com\u002Fgranularai\u002Ffabric) -> 扩展为包含三个不同日期的数据集。\n* [MSOSCD](https:\u002F\u002Fgithub.com\u002FLihy256\u002FMSCDUnet) -> 包含 VHR、多光谱（Sentinel-2）和 SAR（Sentinel-1）的变化检测数据集。\n\n## SECOND - 语义变化检测\n* https:\u002F\u002Fcaptain-whu.github.io\u002FSCD\u002F\n* 在像素级别进行变化检测。\n\n## 亚马逊雨林和大西洋森林数据集\n用于使用 Sentinel 2 进行语义分割。\n* [亚马逊雨林和大西洋森林语义分割影像数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F4498086#.Y6LPLuzP1hE)\n* [attention-mechanism-unet](https:\u002F\u002Fgithub.com\u002Fdavej23\u002Fattention-mechanism-unet) -> 基于注意力机制的 U-Net，用于检测卫星传感器影像中的森林砍伐。\n* [TransUNetplus2](https:\u002F\u002Fgithub.com\u002Faj1365\u002FTransUNetplus2) -> 重新思考带注意力门控的 TransU-Net，用于森林砍伐地图绘制。\n\n## 世界功能地图（fMoW）\n* https:\u002F\u002Fgithub.com\u002FfMoW\u002Fdataset\n* RGB与多光谱变体\n* 高分辨率、芯片分类数据集\n* 目的：基于卫星影像的时间序列及丰富的元数据特征，预测建筑物的功能用途和土地利用类型\n\n## HRSCD变化检测\n* https:\u002F\u002Frcdaudt.github.io\u002Fhrscd\u002F\n* 291对高分辨率RGB航空影像的配准图像对\n* 提供像素级的变化和地表覆盖标注\n\n## MiniFrance-DFC22 - 半监督语义分割\n* [MiniFrance-DFC22（MF-DFC22）数据集](https:\u002F\u002Fieee-dataport.org\u002Fcompetitions\u002Fdata-fusion-contest-2022-dfc2022)扩展并修改了[MiniFrance数据集](https:\u002F\u002Fieee-dataport.org\u002Fopen-access\u002Fminifrance)，用于训练半监督语义分割模型，以进行土地利用\u002F地表覆盖制图。\n* [dfc2022-baseline](https:\u002F\u002Fgithub.com\u002Fisaaccorley\u002Fdfc2022-baseline) -> 使用TorchGeo、PyTorch Lightning和Segmentation Models PyTorch，基于ResNet-18骨干网络，结合Focal + Dice损失函数，训练U-Net模型，完成DFC2022数据集上的语义分割任务，作为2022年IEEE GRSS数据融合竞赛（DFC2022）的基准解决方案。\n* https:\u002F\u002Fgithub.com\u002Fmveo\u002Fmveo-challenge\n\n## FLAIR\n由法国国家地理与森林信息研究所（IGN）提出的语义分割与领域适应挑战赛。该数据集包含超过7万张带有像素级标注的航空影像块，以及5万景Sentinel-2卫星影像。\n* [Codalab上的挑战赛](https:\u002F\u002Fcodalab.lisn.upsaclay.fr\u002Fcompetitions\u002F13447)\n* [FLAIR-2 GitHub仓库](https:\u002F\u002Fgithub.com\u002FIGNF\u002FFLAIR-2)\n* [flair-2第八名解决方案](https:\u002F\u002Fgithub.com\u002Fassociation-rosia\u002Fflair-2)\n* [IGNF HuggingFace页面](https:\u002F\u002Fhuggingface.co\u002FIGNF)\n\n## ISPRS\n语义分割数据集。38个6000×6000像素的影像块，每个块由从更大范围的正射影像拼接图中裁剪出的真实正射影像（TOP）和数字表面模型（DSM）组成。分辨率为5厘米。\n* https:\u002F\u002Fwww.isprs.org\u002Fresources\u002Fdatasets\u002Fbenchmarks\u002FUrbanSemLab\u002F2d-sem-label-potsdam.aspx\n\n## SpaceNet\nSpaceNet是一系列比赛的总称，提供数据集和相关工具。涵盖的挑战包括：(1 & 2) 建筑物分割，(3) 道路分割，(4) 斜视角建筑物，(5) 道路网络提取，(6) 多传感器测绘，(7) 多时相城市变化，(8) 基于多类别分割的洪水检测挑战。\n* [spacenet.ai](https:\u002F\u002Fspacenet.ai\u002F) 是一个在线平台，提供数据、挑战、算法和工具。\n* [SpaceNet 7多时相城市发展挑战：数据集发布](https:\u002F\u002Fmedium.com\u002Fthe-downlinq\u002Fthe-spacenet-7-multi-temporal-urban-development-challenge-dataset-release-9e6e5f65c8d5)\n* [spacenet-three-topcoder](https:\u002F\u002Fgithub.com\u002Fsnakers4\u002Fspacenet-three-topcoder) 解决方案\n* [官方工具库](https:\u002F\u002Fgithub.com\u002FSpaceNetChallenge\u002Futilities) -> 旨在帮助将SpaceNet卫星影像数据预处理为机器学习算法可使用的格式。\n* [andraugust spacenet-utils](https:\u002F\u002Fgithub.com\u002Fandraugust\u002Fspacenet-utils) -> 可显示带有建筑物多边形叠加的geotiff图像，并利用kNN算法根据像素光谱对建筑物进行标注。\n* [Spacenet-Building-Detection](https:\u002F\u002Fgithub.com\u002FIdanC1s2\u002FSpacenet-Building-Detection) -> 使用Keras和[SpaceNet 1数据集](https:\u002F\u002Fspacenet.ai\u002Fspacenet-buildings-dataset-v1\u002F)。\n* [SpaceNet 8获奖者博客文章](https:\u002F\u002Fmedium.com\u002F@SpaceNet_Project\u002Fspacenet-8-a-closer-look-at-the-winning-approaches-75ff4033bf53)\n\n## WorldStrat数据集\n近1万平方公里的免费高分辨率卫星影像，覆盖全球独特的地理位置，确保对各类土地利用类型的分层代表性：从农业到冰盖，从森林到不同密度的城市化区域。\n* https:\u002F\u002Fgithub.com\u002Fworldstrat\u002Fworldstrat\n* [WorldStrat数据集快速导览](https:\u002F\u002Fmedium.com\u002F@robmarkcole\u002Fquick-tour-of-the-worldstrat-dataset-b2d1c2d435db)\n* 每张高分辨率影像（1.5米\u002F像素）都配有来自免费开放的低分辨率Sentinel-2卫星的多时相低分辨率影像（10米\u002F像素）。\n* 已有多个超分辨率基准模型在此数据集上进行训练。\n\n## Satlas Pretrain\nSatlasPretrain是一个大规模的预训练数据集，用于涉及理解卫星影像的任务。通过Sentinel-2和NAIP等来源，地球大部分地区的定期更新卫星数据公开可用，可用于支持多种应用，从打击非法砍伐到监测海洋基础设施。\n* [网站](https:\u002F\u002Fsatlas-pretrain.allen.ai\u002F)\n* [代码](https:\u002F\u002Fgithub.com\u002Fallenai\u002Fsatlas)\n\n## FLAIR 1 & 2 分割数据集\n* https:\u002F\u002Fignf.github.io\u002FFLAIR\u002F\n* FLAIR #1语义分割数据集包含77,412个高分辨率影像块（512×512，空间分辨率为0.2米），涵盖19种语义类别。\n* FLAIR #2则包含了扩展的Sentinel-2时间序列数据集，用于多模态语义分割。\n\n## 五十亿像素分割数据集\n* https:\u002F\u002Fx-ytong.github.io\u002Fproject\u002FFive-Billion-Pixels.html\n* 中国地区的4m分辨率高分二号影像\n* 24种地表覆盖类别\n* 论文和代码展示了如何将模型迁移到Sentinel-2和Planetscope影像上。\n* 扩展了[GID15大规模语义分割数据集](https:\u002F\u002Fcaptain-whu.github.io\u002FGID15\u002F)。\n* [GID](https:\u002F\u002Fx-ytong.github.io\u002Fproject\u002FGID.html) -> 高分影像数据集是一个基于高分二号（GF-2）卫星影像的大规模地表覆盖数据集。\n* [MM-5B数据集](https:\u002F\u002Fgithub.com\u002FAI-Tianlong\u002FHieraRS) -> 多模态五十亿像素数据集是一个大规模、多模态、层次化的地表覆盖与土地利用（LCLU）数据集，建立在五十亿像素数据集的基础上。\n\n## RF100目标检测基准\nRF100由100个跨越多个领域的现实世界数据集组成。其目的是通过对该数据集的性能评估，能够更细致地指导模型在不同领域中的表现。包含1万张航空影像。\n* https:\u002F\u002Fwww.rf100.org\u002F\n* https:\u002F\u002Fgithub.com\u002Froboflow-ai\u002Froboflow-100-benchmark\n\n## SATIN（SATellite ImageNet）\nSATIN是一个多任务遥感分类元数据集，由27个数据集组成，分为6个任务。影像分辨率跨度达5个数量级，涵盖超过250个不同的类别标签，以及多种视场大小。SATIN整体基准及其27个子数据集均通过HuggingFace发布。同时提供公开排行榜，用于指导和跟踪视觉-语言模型在SATIN上的进展。\n* [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.11619)\n* [网站](https:\u002F\u002Fsatinbenchmark.github.io\u002F)\n* [数据](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fjonathan-roberts1\u002FSATIN)\n\n## SODA-A旋转边界框\n* https:\u002F\u002Fshaunyuan22.github.io\u002FSODA\u002F\n* SODA-A包含2513张高分辨率航空场景图像，共标注了872,069个实例，采用定向矩形框标注，涵盖9个类别。\n* https:\u002F\u002Fgithub.com\u002Fshaunyuan22\u002FCFINet\n\n## Satellogic 的 EarthView 数据集\n* https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fsatellogic\u002FEarthView\n* 用于基础模型的数据集，包含 Sentinel 1 和 2 卫星数据以及 1 米分辨率的 RGB 影像。\n\n## 微软的数据集\n* [美国建筑物轮廓](https:\u002F\u002Fgithub.com\u002FMicrosoft\u002FUSBuildingFootprints) -> 美国50个州的建筑物轮廓，GeoJSON格式，通过语义分割生成。此外还有[澳大利亚](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FAustraliaBuildingFootprints)、[加拿大](https:\u002F\u002Fgithub.com\u002FMicrosoft\u002FCanadianBuildingFootprints)、[乌干达-坦桑尼亚](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FUganda-Tanzania-Building-Footprints)、[肯尼亚-尼日利亚](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FKenyaNigeriaBuildingFootprints)以及[全球建筑物轮廓](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FGlobalMLBuildingFootprints)等版本。可以使用[RasterizingBuildingFootprints](https:\u002F\u002Fgithub.com\u002Fmehdiheris\u002FRasterizingBuildingFootprints)将矢量形状文件转换为栅格图层。\n* [微软行星计算机](https:\u002F\u002Fplanetarycomputer.microsoft.com\u002F) 是一个基于 Dask-Gateway 的 JupyterHub 部署，专注于支持可扩展的地理空间分析，其源代码仓库为[这里](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fplanetary-computer-hub)。\n* [landcover-orinoquia](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Flandcover-orinoquia) -> 与哥伦比亚野生动物保护协会合作，对哥伦比亚奥里诺基亚地区进行土地覆盖分类。这是一个 #AIforEarth 项目。\n* [微软道路检测数据集](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRoadDetections)\n\n## 谷歌的数据集\n* [open-buildings](https:\u002F\u002Fsites.research.google\u002Fopen-buildings\u002F) -> 一个用于支持公益应用的建筑物轮廓数据集，覆盖了非洲大陆的64%。阅读文章[利用卫星影像绘制非洲建筑地图](https:\u002F\u002Fai.googleblog.com\u002F2021\u002F07\u002Fmapping-africas-buildings-with.html)。\n\n## Google Earth Engine (GEE)\n由于 GEE 拥有一个庞大的社区，这里不再赘述，仅列出一些精选资源。可以从 https:\u002F\u002Fdevelopers.google.com\u002Fearth-engine\u002F 开始学习。\n* 提供多种遥感影像和气候数据集，包括 Landsat 和 Sentinel 卫星影像。\n* 支持使用传统算法进行大规模处理，例如用于土地利用分类的聚类分析。对于深度学习，可以将 GEE 中的数据导出为 tfrecords 格式，在您选择的 GPU 平台上训练模型，然后将推理结果上传回 GEE。\n* [awesome-google-earth-engine](https:\u002F\u002Fgithub.com\u002Fgee-community\u002Fawesome-google-earth-engine)\n* [Awesome-GEE](https:\u002F\u002Fgithub.com\u002Fgiswqs\u002FAwesome-GEE)\n* [awesome-earth-engine-apps](https:\u002F\u002Fgithub.com\u002Fphilippgaertner\u002Fawesome-earth-engine-apps)\n* [如何使用 Google Earth Engine 和 Python API 将图像导出到 Roboflow](https:\u002F\u002Fblog.roboflow.com\u002Fhow-to-use-google-earth-engine-with-roboflow\u002F) -> 用于获取训练数据。\n* [ee-fastapi](https:\u002F\u002Fgithub.com\u002Fcsaybar\u002Fee-fastapi) 是一个简单的 FastAPI Web 应用程序，后端使用 Google Earth Engine 进行洪水检测。\n* [如何下载地球上任何地点的高分辨率卫星数据](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241003151941\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fhow-to-download-high-resolution-satellite-data-for-anywhere-on-earth-5e6dddee2803)\n* [wxee](https:\u002F\u002Fgithub.com\u002Faazuspan\u002Fwxee) -> 使用 wxee 将 GEE 中的数据导出为 xarray 格式，然后使用 PyTorch 或 TensorFlow 模型进行训练。这很有用，因为 GEE 本身只支持导出 tfrecord 格式。\n\n## 图像描述数据集\n* [RSICD](https:\u002F\u002Fgithub.com\u002F201528014227051\u002FRSICD_optimal) -> 包含10921张图片，每张图片配有五句描述。该数据集被用于[使用遥感（卫星）图像和描述微调 CLIP 模型](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Ffine-tune-clip-rsicd)，相关模型可在[这个仓库](https:\u002F\u002Fgithub.com\u002Farampacha\u002FCLIP-rsicd)中找到。\n* [RSICC](https:\u002F\u002Fgithub.com\u002FChen-Yang-Liu\u002FRSICC) -> 遥感图像变化描述数据集包含10077对不同时期的遥感图像，以及50385句描述图像之间差异的文字。使用 LEVIR-CD 影像。\n* [ChatEarthNet](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FChatEarthNet) -> 全球规模的图文数据集，用于赋能视觉-语言地理基础模型，采用 Sentinel-2 数据，并由 ChatGPT 生成描述文字。\n\n## 天气数据集\n* NASA（需提交请求，准备就绪后会通过邮件发送）-> https:\u002F\u002Fsearch.earthdata.nasa.gov\n* NOAA（需要 BigQuery）-> https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fnoaa\u002Fgoes16\u002Fhome\n* 美国多个城市的气象时间序列数据 -> https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fselfishgene\u002Fhistorical-hourly-weather-data\n* [DeepWeather](https:\u002F\u002Fgithub.com\u002Fadamhazimeh\u002FDeepWeather) -> 通过分析卫星图像来提高天气预报的准确性。\n\n## 云数据集\n* [Planet-CR](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FPlanet-CR) -> 用于高分辨率光学遥感影像去云处理的多模态、多分辨率数据集，分辨率为3米，并有相关论文[在这里](https:\u002F\u002Farxiv.org\u002Fabs\u002F2301.03432)。\n* [Azavea 云数据集](https:\u002F\u002Fwww.azavea.com\u002Fblog\u002F2021\u002F08\u002F02\u002Fthe-azavea-cloud-dataset\u002F)，用于训练[这款云检测模型](https:\u002F\u002Fgithub.com\u002Fazavea\u002Fcloud-model)。\n* [Sentinel-2 云覆盖分割数据集](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20230606184945\u002Fhttps:\u002F\u002Fmlhub.earth\u002Fdata\u002Fref_cloud_cover_detection_challenge_v1) 在 Radiant mlhub 上提供。\n* [cloudsen12](https:\u002F\u002Fcloudsen12.github.io\u002F) -> 参见[视频](https:\u002F\u002Fyoutu.be\u002FGhQwnVhJ1wo)。\n* [WHUS2-CD+](https:\u002F\u002Fzenodo.org\u002Frecord\u002F5511793) -> 包含36幅人工标注的10米分辨率云掩膜及其对应的 Sentinel-2 影像，均匀分布在中国大陆各地，用于训练 CD-FM3SF [云检测模型](https:\u002F\u002Fgithub.com\u002FNeooolee\u002FWHUS2-CD)。\n* [HRC_WHU](https:\u002F\u002Fgithub.com\u002Fdr-lizhiwei\u002FHRC_WHU) -> 高分辨率云检测数据集，包含150张RGB影像，分辨率在不同全球区域介于0.5至15米之间。\n* [AIR-CD](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FAIR-CD) -> 一个具有挑战性的云检测数据集，具有更高的空间分辨率和更具代表性的地表类型。\n* [Landsat 8 云覆盖评估验证数据](https:\u002F\u002Flandsat.usgs.gov\u002Flandsat-8-cloud-cover-assessment-validation-data)\n\n## 森林数据集\n* [OpenForest](https:\u002F\u002Fgithub.com\u002FRolnickLab\u002FOpenForest) -> 一个开放获取的森林数据集目录\n* [awesome-forests](https:\u002F\u002Fgithub.com\u002Fblutjens\u002Fawesome-forests) -> 面向机器学习和林业社区的精选地面真值森林数据集列表\n* [ReforesTree](https:\u002F\u002Fgithub.com\u002Fgyrrei\u002FReforesTree) -> 基于无人机和地面数据估算热带森林生物量的数据集\n* [yosemite-tree-dataset](https:\u002F\u002Fgithub.com\u002Fnightonion\u002Fyosemite-tree-dataset) -> 用于航拍图像中树木计数的基准数据集\n* [亚马逊雨林语义分割数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F3233081#.Y6LPLOzP1hE) -> Sentinel 2 影像。用于论文《基于注意力机制的 U-Net 用于检测卫星传感器影像中的森林砍伐》\n* [亚马逊和大西洋森林语义分割影像数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F4498086#.Y6LPLuzP1hE) -> Sentinel 2 影像。用于论文《基于注意力机制的 U-Net 用于检测卫星传感器影像中的森林砍伐》\n* [TreeSatAI](https:\u002F\u002Fzenodo.org\u002Frecords\u002F6780578) -> Sentinel-1、Sentinel-2\n* [PureForest](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FIGNF\u002FPureForest) -> VHR RGB + 近红外与激光雷达，每个图块代表一片单一树种的森林\n\n## 地理空间数据集\n* [Resource Watch](https:\u002F\u002Fresourcewatch.org\u002Fdata\u002Fexplore) 提供广泛的地理空间数据集及可视化界面\n\n## 时间序列与变化检测数据集\n* [BreizhCrops](https:\u002F\u002Fgithub.com\u002Fdl4sits\u002FBreizhCrops) -> 用于作物类型制图的时间序列数据集\n* SeCo 数据集包含来自 Sentinel-2 图幅的图像块，在每个地理位置的不同时间点采集。[在此下载 SeCo](https:\u002F\u002Fgithub.com\u002FElementAI\u002Fseasonal-contrast)\n* [SYSU-CD](https:\u002F\u002Fgithub.com\u002Fliumency\u002FSYSU-CD) -> 该数据集包含 20000 对 0.5 米分辨率的航拍图像，尺寸为 256×256，拍摄于 2007 年至 2014 年期间的香港地区\n\n### DEM（数字高程地图）\n* 航天飞机雷达地形测绘任务，可在 usgs.gov 在线查询\n* Copernicus 数字高程模型 (DEM) 存储在 S3 上，表示地球表面，包括建筑物、基础设施和植被。数据以云优化 GeoTIFF 格式提供。[链接](https:\u002F\u002Fregistry.opendata.aws\u002Fcopernicus-dem\u002F)\n* [Awesome-DEM](https:\u002F\u002Fgithub.com\u002FDahnJ\u002FAwesome-DEM)\n\n## 无人机数据集\n* 许多数据集可在 https:\u002F\u002Fwww.visualdata.io 上找到\n* [AU-AIR 数据集](https:\u002F\u002Farxiv.org\u002Fabs\u002F2104.06781) -> 用于目标检测的多模态无人机数据集\n* [ERA](https:\u002F\u002Flcmou.github.io\u002FERA_Dataset\u002F) -> 用于航拍视频中事件识别的数据集和深度学习基准\n* [航拍海事无人机数据集](https:\u002F\u002Fpublic.roboflow.ai\u002Fobject-detection\u002Faerial-maritime) -> 边界框标注\n* [RetinaNet 用于行人检测](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241002023333\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fpedestrian-detection-in-aerial-images-using-retinanet-9053e8a72c6) -> 边界框标注\n* [BIRDSAI：用于航拍热红外视频中检测与跟踪的数据集](https:\u002F\u002Fgithub.com\u002Fexb7900\u002FBIRDSAI) -> 人类和动物的热红外视频\n* [ERA：用于航拍视频中事件识别的数据集和深度学习基准](https:\u002F\u002Flcmou.github.io\u002FERA_Dataset\u002F)\n* [DroneVehicle](https:\u002F\u002Fgithub.com\u002FVisDrone\u002FDroneVehicle) -> 基于无人机的 RGB-红外跨模态车辆检测，采用不确定性感知学习。标注为旋转边界框。配套 GitHub 仓库：[UA-CMDet](https:\u002F\u002Fgithub.com\u002FSunYM2020\u002FUA-CMDet)\n* [UAVOD10](https:\u002F\u002Fgithub.com\u002Fweihancug\u002F10-category-UAV-small-weak-object-detection-dataset-UAVOD10) -> 10 类物体，分辨率为 15 厘米。类别包括：建筑物、船只、车辆、预制房屋、水井、电缆塔、游泳池、滑坡区域、养殖网箱和采石场。标注为边界框\n* [繁忙停车场数据集——无人机视频中的车辆检测](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FBusy-parking-lot-dataset---vehicle-detection-in-UAV-video) -> 车辆实例分割。标注格式尚不明确，可能是 MATLAB 特有的格式\n* [dd-ml-segmentation-benchmark](https:\u002F\u002Fgithub.com\u002Fdronedeploy\u002Fdd-ml-segmentation-benchmark) -> DroneDeploy 机器学习分割基准\n* [SeaDronesSee](https:\u002F\u002Fgithub.com\u002FBen93kie\u002FSeaDronesSee) -> 海上搜救视觉基准。标注包括边界框目标检测、单目标跟踪和多目标跟踪\n* [aeroscapes](https:\u002F\u002Fgithub.com\u002Fishann\u002Faeroscapes) -> 语义分割基准，由商用无人机从 5 至 50 米高度拍摄的图像组成\n* [ALTO](https:\u002F\u002Fgithub.com\u002FMetaSLAM\u002FALTO) -> 航拍视角下的大规模地形导向数据集。用于基于深度学习的无人机视觉定位与场景识别任务\n* [HIT-UAV-Infrared-Thermal-Dataset](https:\u002F\u002Fgithub.com\u002Fsuojiashun\u002FHIT-UAV-Infrared-Thermal-Dataset) -> 用于无人机的高空红外热成像目标检测数据集\n* [caltech-aerial-rgbt-dataset](https:\u002F\u002Fgithub.com\u002Faerorobotics\u002Fcaltech-aerial-rgbt-dataset) -> 同步的 RGB、热成像、GPS 和 IMU 数据\n* [叶状大戟数据集](https:\u002F\u002Fleafy-spurge-dataset.github.io\u002F) -> 航拍无人机影像中的真实世界杂草分类\n* [Agriculture-Vision 2021 数据集](https:\u002F\u002Fwww.agriculture-vision.com\u002Fagriculture-vision-2021\u002Fdataset-2021)\n* [UAV-HSI-Crop-Dataset](https:\u002F\u002Fgithub.com\u002FMrSuperNiu\u002FUAV-HSI-Crop-Dataset) -> 用于“HSI-TransUNet：基于 Transformer 的语义分割模型，用于从无人机高光谱影像中进行作物制图”的数据集\n* [UAVVaste](https:\u002F\u002Fgithub.com\u002FPUTvision\u002FUAVVaste) -> 类似 COCO 的数据集，用于航拍图像中的有效垃圾检测\n* [BSB-Aerial-Dataset](https:\u002F\u002Fgithub.com\u002Fosmarluiz\u002FBSB-Aerial-Dataset) -> 巴西巴西利亚航拍影像的全景分割数据集\n\n## 其他数据集\n\n### 目标检测与分类\n* [RSOD-Dataset](https:\u002F\u002Fgithub.com\u002FRSIA-LIESMARS-WHU\u002FRSOD-Dataset-) -> 用于目标检测的数据集，采用PASCAL VOC格式。包含飞机、游乐场、立交桥和油罐等类别。\n* [VHR-10_dataset_coco](https:\u002F\u002Fgithub.com\u002Fchaozhong2010\u002FVHR-10_dataset_coco) -> 基于NWPU VHR-10数据集的目标检测与实例分割数据集。包含RGB和SAR两种模态。\n* [MAR20](https:\u002F\u002Fgcheng-nwpu.github.io\u002F) -> 军用飞机识别数据集。\n* [RSAPS-ASD](https:\u002F\u002Fgithub.com\u002FSKLSEIIT\u002FRSAPS-ASD) -> 遥感机场全景分割与飞机状态数据集，构建于“从单时相高分辨率遥感图像中进行飞机状态判别”研究中。\n* [Sewage-Treatment-Plant-Dataset](https:\u002F\u002Fgithub.com\u002Fpeijinwang\u002FSewage-Treatment-Plant-Dataset) -> 目标检测数据集。\n* [TGRS-HRRSD-Dataset](https:\u002F\u002Fgithub.com\u002FCrazyStoneonRoad\u002FTGRS-HRRSD-Dataset) -> 高分辨率遥感目标检测（HRRSD）数据集。\n* [OGST](https:\u002F\u002Fdata.mendeley.com\u002Fdatasets\u002Fbkxj8z84m9\u002F3) -> 石油天然气储罐数据集。\n* [SearchAndRescueNet](https:\u002F\u002Fgithub.com\u002Fmichaelthoreau\u002FSearchAndRescueNet) -> 用于搜救任务的卫星影像数据集，并附有Faster R-CNN模型示例。\n* [UBC-dataset](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FUBC-dataset) -> 用于建筑物检测与分类的数据集，基于超高分辨率卫星影像，重点在于对单个建筑物的对象级解读。\n* [Building_Dataset](https:\u002F\u002Fgithub.com\u002FQiaoWenfan\u002FBuilding_Dataset) -> 高速铁路沿线建筑物展示数据集。\n* [RID](https:\u002F\u002Fgithub.com\u002FTUMFTM\u002FRID) -> 用于基于计算机视觉的光伏潜力评估的屋顶信息数据集。相关论文：[链接](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F14\u002F10\u002F2299)。\n* [APKLOT](https:\u002F\u002Fgithub.com\u002Flangheran\u002FAPKLOT) -> 用于航空影像中停车位分割的数据集。\n* [SAR-ACD](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FSAR-ACD) -> SAR-ACD包含4322段飞机视频片段，涵盖6类民用飞机和14类其他飞机。\n* [SODA](https:\u002F\u002Fshaunyuan22.github.io\u002FSODA\u002F) -> 大规模小目标检测数据集。SODA-A包含2510张高分辨率航拍图像，标注了9个类别的800,203个实例，使用定向矩形框标注。\n* [urban-tree-detection-data](https:\u002F\u002Fgithub.com\u002Fjonathanventura\u002Furban-tree-detection-data) -> 用于训练和评估城市环境中树木检测器的航拍影像数据集。\n* [包含船舶的卫星影像数据集](https:\u002F\u002Fgithub.com\u002FNaLiu613\u002FSatellite-Imagery-Datasets-Containing-Ships) -> 列举了一系列用于船舶检测、分类、语义分割和实例分割任务的雷达及光学卫星数据集。\n* [Roofline-Extraction](https:\u002F\u002Fgithub.com\u002Floosgagnet\u002FRoofline-Extraction) -> 用于论文《基于知识的单张航拍图像三维建筑物重建（3DBR）及卷积神经网络（CNN）应用》的数据集。\n* [Building-detection-and-roof-type-recognition](https:\u002F\u002Fgithub.com\u002Floosgagnet\u002FBuilding-detection-and-roof-type-recognition) -> 用于论文《基于CNN的单张航拍图像自动建筑物检测与屋顶类型识别方法》的数据集。\n* [OnlyPlanes](https:\u002F\u002Fgithub.com\u002Fnaivelogic\u002FOnlyPlanes) -> 用于Detectron2的合成数据集及预训练模型。\n* [SV248S](https:\u002F\u002Fgithub.com\u002Fxdai-dlgvv\u002FSV248S) -> 单目标跟踪数据集，用于跟踪车辆、大型车辆、船舶和飞机。\n* [NWPU-MOC](https:\u002F\u002Fgithub.com\u002Flyongo\u002FNWPU-MOC) -> 航空影像中细粒度多类别物体计数基准数据集。\n* [卫星遥感车辆感知](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.00703) -> 用于卫星交通监测的大规模基准数据集。\n* [SARDet-100K](https:\u002F\u002Fgithub.com\u002Fzcablii\u002FSARDet_100K) -> 大规模合成孔径雷达（SAR）目标检测数据集。\n* [城市车辆分割数据集（UV6K）](https:\u002F\u002Fzenodo.org\u002Frecords\u002F8404754)\n* [ShipRSImageNet](https:\u002F\u002Fgithub.com\u002Fzzndream\u002FShipRSImageNet) -> 用于高分辨率光学遥感影像中船舶检测的大规模细粒度数据集。\n* [VME：中东及其他地区车辆检测的卫星影像数据集与基准](https:\u002F\u002Fgithub.com\u002Fnalemadi\u002FVME_CDSI_dataset_benchmark)\n* [VHRV：超高分辨率船舶检测基准数据集](https:\u002F\u002Fgithub.com\u002Fbuyukkanber\u002Fvhrv)\n\n### 土地利用与土地覆被\n* [land-use-land-cover-datasets](https:\u002F\u002Fgithub.com\u002Fr-wenger\u002Fland-use-land-cover-datasets)\n* [RSD46-WHU](https:\u002F\u002Fgithub.com\u002FRSIA-LIESMARS-WHU\u002FRSD46-WHU) -> 用于图像分类的46个场景类别，免费供教育、科研和商业用途使用\n* [RSSCN7](https:\u002F\u002Fgithub.com\u002Fpalewithout\u002FRSSCN7) -> 文章“基于深度学习的遥感场景分类特征选择”中的数据集\n* [geonrw](https:\u002F\u002Fieee-dataport.org\u002Fopen-access\u002Fgeonrw) -> 正射校正的航空照片、由LiDAR生成的数字高程模型以及包含10个类别的分割地图。配套[仓库](https:\u002F\u002Fgithub.com\u002Fgbaier\u002Fgeonrw)\n* [Attribute-Cooperated-Classification-Datasets](https:\u002F\u002Fgithub.com\u002FCrazyStoneonRoad\u002FAttribute-Cooperated-Classification-Datasets) -> 基于AID、UCM和Sydney的三个数据集。每张图像都配有场景分类标签和属性项标签向量。\n* [open_earth_map](https:\u002F\u002Fgithub.com\u002Fbao18\u002Fopen_earth_map) -> 全球高分辨率土地覆被制图的基准数据集\n* [孟买语义分割数据集](https:\u002F\u002Fgithub.com\u002FGeoAI-Research-Lab\u002FMumbai-Semantic-Segmentation-Dataset)\n* [GAMUS](https:\u002F\u002Fgithub.com\u002FEarthNets\u002FRSI-MMSegmentation) -> 面向遥感数据的几何感知多模态语义分割基准数据集\n* [openWUSU](https:\u002F\u002Fgithub.com\u002FAngieNikki\u002FopenWUSU) -> WUSU是一个专注于武汉城市结构及城市化进程的语义理解数据集\n* [RSE_Cross-city](https:\u002F\u002Fgithub.com\u002Fdanfenghong\u002FRSE_Cross-city) -> 跨城事务：基于高分辨率领域适应网络的跨城语义分割多模态遥感基准数据集\n* [AErial Lane](https:\u002F\u002Fgithub.com\u002FAidasDir\u002FAerialLaneNet) -> AErial Lane (AEL) 数据集是首个为车道检测构建的大规模航空影像数据集，在约80公里道路的高分辨率图像上提供了高质量的折线型车道标注\n* [切萨皮克道路空间上下文（RSC）](https:\u002F\u002Fgithub.com\u002Fisaaccorley\u002FChesapeakeRSC)\n* [So2Sat-POP-DL](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSo2Sat-POP-DL) -> 数据集发现：覆盖98个欧盟城市的So2Sat人口数据集\n* [HouseTS](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fshengkunwang\u002Fhousets-dataset) -> 涵盖美国30个大都市区的长期多模态住房数据集。使用NAIP数据。[附论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.00765)\n* [印度全国1万块农作物田块边界](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7315090) -> 使用Airbus SPOT卫星数据\n* [OpenEarthMap-SAR](https:\u002F\u002Fgithub.com\u002Fcliffbb\u002FOpenEarthMap-SAR) -> 用于2025年IEEE GRSS数据融合竞赛赛道1：全天候土地覆被制图的VHR SAR数据。使用Umbra和Capella Space的数据\n* [东京土地利用土地覆被数据集](https:\u002F\u002Fgithub.com\u002FTusaifei\u002FTokyo_dataset) -> 0.5米分辨率影像、两种10米分辨率LCP以及两种30米分辨率LCP\n\n### 变化检测\n* [S2Looking](https:\u002F\u002Fgithub.com\u002FS2Looking\u002FDataset) -> 用于建筑物变化检测的卫星侧视数据集，[论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2107.09244)\n* [Haiming-Z\u002FMtS-WH-reference-map](https:\u002F\u002Fgithub.com\u002FHaiming-Z\u002FMtS-WH-reference-map) -> 基于MtS-WH的变化检测参考地图\n* [MtS-WH-Dataset](https:\u002F\u002Fgithub.com\u002Frulixiang\u002FMtS-WH-Dataset) -> 多时相武汉场景（MtS-WH）数据集\n* [SZTAKI](http:\u002F\u002Fweb.eee.sztaki.hu\u002Fremotesensing\u002Fairchange_benchmark.html) -> 用于光学航空影像中变化检测的真实标签集合，这些影像拍摄时间间隔数年\n* [DSIFN](https:\u002F\u002Fgithub.com\u002FGeoZcx\u002FA-deeply-supervised-image-fusion-network-for-change-detection-in-remote-sensing-images\u002Ftree\u002Fmaster\u002Fdataset) -> 变化检测数据集，由六幅大型双时相高分辨率影像组成，覆盖中国六个城市\n* [道路变化检测数据集](https:\u002F\u002Fgithub.com\u002FfightingMinty\u002FRoad-Change-Detection-Dataset)\n* [3DCD](https:\u002F\u002Fsites.google.com\u002Funiroma1.it\u002F3dchangedetection\u002Fhome-page) -> 仅使用遥感光学双时相影像作为输入，无需数字高程模型（DEM），即可推断出3D变化检测地图\n* [TUE-CD](https:\u002F\u002Fgithub.com\u002FRSMagneto\u002FMSI-Net) -> 用于地震后建筑物损毁评估的变化检测方法\n* [Hi-UCD](https:\u002F\u002Fgithub.com\u002FDaisy-7\u002FHi-UCD-S) -> 超高分辨率城市变化检测，用于城市语义变化检测\n* [LEVIR-CC-Dataset](https:\u002F\u002Fgithub.com\u002FChen-Yang-Liu\u002FLEVIR-CC-Dataset) -> 用于遥感影像变化描述的大规模数据集\n* [GDCLD](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13612636) -> 通过多源高分辨率遥感影像进行同震滑坡测绘的全球分布数据集\n* [BANet变化检测数据集 - 遥感影像到地籍图](https:\u002F\u002Fgithub.com\u002Flqycrystal\u002FBANet)\n* [印度城市变化检测（ICCD）数据集](https:\u002F\u002Fieee-dataport.org\u002Fdocuments\u002Findian-cities-change-detection-iccd-dataset)\n\n### SAR专用数据集\n* [HRSID](https:\u002F\u002Fgithub.com\u002Fchaozhong2010\u002FHRSID) -> 用于船舶检测、语义分割和实例分割任务的高分辨率SAR影像数据集\n* [LS-SSDD-v1.0-OPEN](https:\u002F\u002Fgithub.com\u002FTianwenZhang0825\u002FLS-SSDD-v1.0-OPEN) -> 大规模SAR船舶检测数据集\n* [WHU-SEN-City](https:\u002F\u002Fgithub.com\u002Fwhu-csl\u002FWHU-SEN-City) -> 一个涵盖中国34个大城市的SAR与光学影像配对翻译数据集\n* [SAR_vehicle_detection_dataset](https:\u002F\u002Fgithub.com\u002Fwhu-csl\u002FSAR_vehicle_detection_dataset) -> 104张用于车辆检测的SAR影像，来源于Sandia MiniSAR\u002FFARAD SAR影像和MSTAR影像\n* [AIR-PolSAR-Seg](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FAIR-PolSAR-Seg) -> 一个具有挑战性的PolSAR地形分割数据集\n* [QXS-SAROPT](https:\u002F\u002Fgithub.com\u002Fyaoxu008\u002FQXS-SAROPT) -> 来自[论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.08259)的光学与SAR配对数据集：用于SAR-光学数据融合深度学习的QXS-SAROPT数据集\n* [SynthWakeSAR](https:\u002F\u002Fdata.bris.ac.uk\u002Fdata\u002Fdataset\u002F30kvuvmatwzij2mz1573zqumfx) -> 用于海上船舶深度学习分类的合成SAR数据集，[附论文](https:\u002F\u002Fwww.mdpi.com\u002F2072-4292\u002F14\u002F16\u002F3999)\n* [SAR2Opt-Heterogeneous-Dataset](https:\u002F\u002Fgithub.com\u002FMarsZhaoYT\u002FSAR2Opt-Heterogeneous-Dataset) -> 用于遥感影像变化检测和图像转换基准的SAR-光学影像\n* [OpenSARWake](https:\u002F\u002Fgithub.com\u002Flibzzluo\u002FOpenSARWake) -> 用于SAR船舶航迹旋转检测的基准数据集。\n\n### 专用应用\n* [MUSIC4HA](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FMUSIC4HA) -> 多波段卫星影像用于目标分类（MUSIC），以检测热点区域\n* [MUSIC4GC](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FMUSIC4GC) -> 多波段卫星影像用于目标分类（MUSIC），以检测高尔夫球场\n* [MUSIC4P3](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FMUSIC4P3) -> 多波段卫星影像用于目标分类（MUSIC），以检测光伏电站（太阳能电池板）\n* [ABCDdataset](https:\u002F\u002Fgithub.com\u002Fgistairc\u002FABCDdataset) -> 损害检测数据集，用于识别建筑物是否被海啸冲毁\n* [火力发电厂数据集](https:\u002F\u002Fgithub.com\u002FwenxinYin\u002FAIR-TPPDD)\n* [SolarDK](https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.01260) -> 一个高分辨率的城市太阳能电池板图像分类与定位数据集\n* [油气基础设施测绘（OGIM）数据库](https:\u002F\u002Fzenodo.org\u002Frecord\u002F7922117) -> 包含重要的甲烷排放源——各类油气基础设施的位置及设施属性\n* [架空风力涡轮机数据集 - NAIP](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7385227#.Y419qezMLdr)\n* [CloudTracks: 用于在云层卫星图像中定位船舶航迹的数据集](https:\u002F\u002Fzenodo.org\u002Frecords\u002F10042922) -> 该数据集包含1,780张MODIS卫星图像，人工标注了超过12,000条船舶航迹。\n* [数字台风数据集](https:\u002F\u002Fgithub.com\u002Fkitamoto-lab\u002Fdigital-typhoon\u002F) -> 旨在为长期时空数据的机器学习模型提供基准测试\n* [BirdSAT](https:\u002F\u002Fgithub.com\u002Fmvrl\u002FBirdSAT) -> 跨视角 iNaturalist 鸟类 2021：该跨视角鸟类物种数据集由地面拍摄的鸟类图像和卫星图像配对组成，并附有与 iNaturalist-2021 数据集相关的元信息。\n* [RSHaze+](https:\u002F\u002Fzenodo.org\u002Frecords\u002F13837162) -> PhDnet 中的遥感去雾数据集：一种针对遥感图像的新型物理感知去雾网络\n* [GMSEUS](https:\u002F\u002Fgithub.com\u002Fstidjaco\u002FGMSEUS) -> 美国全面的地面安装式太阳能能源数据集，包含子阵列设计元数据\n* [MultiviewRS](https:\u002F\u002Fgithub.com\u002Ffmenat\u002FmultiviewRS-datasets) -> 用于探索多视角学习的遥感（RS）多视角数据集列表\n* [SatDepth](https:\u002F\u002Fsatdepth.pythonanywhere.com\u002F) -> 一个用于卫星图像匹配和深度估计的新颖数据集\n* [OpenSatMap](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fz-hb\u002FOpenSatMap) -> 用于大规模地图构建及自动驾驶等下游任务\n\n### 农业与环境\n* [高光谱变化检测数据集——灌溉农业区](https:\u002F\u002Fgithub.com\u002FSicongLiuRS\u002FHyperspectral-Change-Detection-Dataset-Irrigated-Agricultural-Area)\n* [CNN-RNN-产量预测](https:\u002F\u002Fgithub.com\u002Fsaeedkhaki92\u002FCNN-RNN-Yield-Prediction) -> 大豆数据集\n* [FireRisk](https:\u002F\u002Fgithub.com\u002FCharmonyShen\u002FFireRisk) -> 用于火灾风险评估的遥感数据集，并提供了基于监督学习和自监督学习的基准测试\n* [TimeMatch](https:\u002F\u002Fzenodo.org\u002Frecords\u002F5636422) -> 用于作物识别的跨区域适应数据集，来自欧洲四个不同地区的SITS数据\n* [Landsat 8 云覆盖评估验证数据](https:\u002F\u002Flandsat.usgs.gov\u002Flandsat-8-cloud-cover-assessment-validation-data)\n* [用于超分辨率的遥感卫星视频数据集](https:\u002F\u002Fzenodo.org\u002Frecord\u002F6969604#.ZCBd-OzMJhE)\n* [SpatioTemporalYield](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fellaampy\u002FSpatioTemporalYield) -> 涵盖美国玉米产量前五的州：爱荷华州、伊利诺伊州、印第安纳州、内布拉斯加州和明尼苏达州。\n* [棕榈树数据集](https:\u002F\u002Fgithub.com\u002FNourO93\u002FPalm-Tree-Dataset\u002Ftree\u002Fmain)\n* [ts-satfire](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fz789456sx\u002Fts-satfire) -> 一个多任务卫星图像时间序列数据集，用于野火检测与预测\n* [GTPBD](https:\u002F\u002Fgithub.com\u002FZ-ZW-WXQ\u002FGTPBD\u002F) -> 全球梯田地块与边界数据集\n\n### 高光谱与多模态\n* [AeroRIT](https:\u002F\u002Fgithub.com\u002Faneesh3108\u002FAeroRIT) -> 高光谱图像分析的新场景\n* [Data-CSHSI](https:\u002F\u002Fgithub.com\u002FYuxiangZhang-BIT\u002FData-CSHSI) -> 用于跨场景高光谱图像分类的开源数据集，包括休斯敦、帕维亚和HyRank数据集\n* [HySpecNet-11k](https:\u002F\u002Fhyspecnet.rsim.berlin\u002F) -> 一个大规模的高光谱基准数据集\n* [STARCOP 数据集：利用高光谱机器学习模型进行甲烷羽流语义分割](https:\u002F\u002Fzenodo.org\u002Frecords\u002F7863343)\n* [图卢兹高光谱数据集](https:\u002F\u002Fwww.toulouse-hyperspectral-data-set.com\u002F)\n* [图卢兹高光谱数据集](https:\u002F\u002Fgithub.com\u002FRomain3Ch216\u002FTlseHypDataSet)\n* [多模态图像匹配](https:\u002F\u002Fgithub.com\u002FStaRainJ\u002FMulti-modality-image-matching-database-metrics-methods) -> 包括多种遥感模态的图像匹配数据集\n* [PanCollection](https:\u002F\u002Fgithub.com\u002Fliangjiandeng\u002FPanCollection) -> 来自WorldView 2、WorldView 3、QuickBird、高分二号传感器的全色锐化数据集\n\n### 基准与基础模型\n* [EORSSD-dataset](https:\u002F\u002Fgithub.com\u002Frmcong\u002FEORSSD-dataset) -> 扩展光学遥感显著性检测（EORSSD）数据集\n* [ERA-DATASET](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FERA-DATASET) -> 用于航空视频中事件识别的数据集和深度学习基准\n* [SSL4EO-S12](https:\u002F\u002Fgithub.com\u002Fzhu-xlab\u002FSSL4EO-S12) -> 用于地球观测自监督学习的大规模数据集\n* [AIR-CD](https:\u002F\u002Fgithub.com\u002FAICyberTeam\u002FAIR-CD) -> 一个具有更高空间分辨率和更具代表性的地表覆盖类型的挑战性云检测数据集，名为AIR-CD\n* [HRC_WHU](https:\u002F\u002Fgithub.com\u002Fdr-lizhiwei\u002FHRC_WHU) -> 高分辨率云检测数据集，包含150张RGB图像，其分辨率在全球不同地区介于0.5至15米之间\n* [University1652-Baseline](https:\u002F\u002Fgithub.com\u002Flayumi\u002FUniversity1652-Baseline) -> 用于无人机地理定位的多视角、多源基准\n* [benchmark_ISPRS2021](https:\u002F\u002Fgithub.com\u002Fwhuwuteng\u002Fbenchmark_ISPRS2021) -> 一个新的用于深度学习的立体密集匹配基准数据集\n* [WHU-Stereo](https:\u002F\u002Fgithub.com\u002FSheng029\u002FWHU-Stereo) -> 用于高分辨率卫星图像立体匹配的挑战性基准\n* [GeoPile预训练数据集](https:\u002F\u002Fgithub.com\u002Fmmendiet\u002FGFM) -> 汇编了来自其他数据集的影像，包括RSD46-WHU、MLRSNet和RESISC45，用于基础模型的预训练\n* [pangaea-bench](https:\u002F\u002Fgithub.com\u002Fyurujaja\u002Fpangaea-bench) -> 用于地理空间基础模型的全球性和包容性基准\n* [VRSBench: 用于遥感图像理解的多功能视觉-语言基准数据集](https:\u002F\u002Fvrsbench.github.io\u002F)\n* [SeeFar](https:\u002F\u002Fcoastalcarbon.ai\u002Fseefar) -> 适用于地理空间基础模型的卫星无关多分辨率数据集\n* [dynnet](https:\u002F\u002Fgithub.com\u002Faysim\u002Fdynnet) -> DynamicEarthNet：用于语义变化分割的日度多光谱卫星数据集\n* [Awesome-Remote-Sensing-Relative-Radiometric-Normalization-Datasets](https:\u002F\u002Fgithub.com\u002FArminMoghimi\u002FAwesome-Remote-Sensing-Relative-Radiometric-Normalization-Datasets)\n* [AISD](https:\u002F\u002Fgithub.com\u002FRSrscoder\u002FAISD) -> 用于阴影检测的航空影像数据集## Kaggle\nKaggle 上托管了超过 200 个卫星图像数据集，[搜索结果在此](https:\u002F\u002Fwww.kaggle.com\u002Fsearch?q=satellite+image+in%3Adatasets)。\n[Kaggle 博客](http:\u002F\u002Fblog.kaggle.com) 是一篇有趣的读物。\n\n### Kaggle - 从太空看亚马逊 - 分类挑战\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fplanet-understanding-the-amazon-from-space\u002Fdata\n* 来自行星鸽群卫星星座的3-5米分辨率GeoTIFF图像\n* 12个类别，包括 - **云层、原始森林 + 水道** 等\n* [第一名获奖者访谈 - 使用了11个自定义CNN](http:\u002F\u002Fblog.kaggle.com\u002F2017\u002F10\u002F17\u002Fplanet-understanding-the-amazon-from-space-1st-place-winners-interview\u002F)\n* [FastAI 多标签图像分类](https:\u002F\u002Ftowardsdatascience.com\u002Ffastai-multi-label-image-classification-8034be646e95)\n* [亚马逊雨林卫星照片的多标签分类](https:\u002F\u002Fmachinelearningmastery.com\u002Fhow-to-develop-a-convolutional-neural-network-to-classify-satellite-photos-of-the-amazon-rainforest\u002F)\n* [通过多标签分类 + VGG-19、Inceptionv3、AlexNet 和迁移学习来理解亚马逊雨林](https:\u002F\u002Ftowardsdatascience.com\u002Funderstanding-the-amazon-rainforest-with-multi-label-classification-vgg-19-inceptionv3-5084544fb655)\n* [amazon-classifier](https:\u002F\u002Fgithub.com\u002Fmikeskaug\u002Famazon-classifier) -> 比较随机森林与CNN\n* [multilabel-classification](https:\u002F\u002Fgithub.com\u002Fmuneeb706\u002Fmultilabel-classification) -> 比较各种CNN架构\n* [Planet-Amazon-Kaggle](https:\u002F\u002Fgithub.com\u002FSkumarr53\u002FPlanet-Amazon-Kaggle) -> 使用fast.ai\n* [deforestation_deep_learning](https:\u002F\u002Fgithub.com\u002Fschumanzhang\u002Fdeforestation_deep_learning)\n* [Track-Human-Footprint-in-Amazon-using-Deep-Learning](https:\u002F\u002Fgithub.com\u002Fsahanasub\u002FTrack-Human-Footprint-in-Amazon-using-Deep-Learning)\n* [Amazon-Rainforest-CNN](https:\u002F\u002Fgithub.com\u002Fcldowdy\u002FAmazon-Rainforest-CNN) -> 在Tensorflow中使用了一个3层CNN\n* [rainforest-tagging](https:\u002F\u002Fgithub.com\u002Fminggli\u002Frainforest-tagging) -> 在Tensorflow中使用卷积神经网络和循环神经网络进行卫星图像的多标签分类\n* [satellite-deforestation](https:\u002F\u002Fgithub.com\u002Fdrewhibbard\u002Fsatellite-deforestation) -> 利用卫星图像识别森林砍伐的先行指标，应用于“从太空看亚马逊”Kaggle挑战赛\n\n### Kaggle - DSTL 分割挑战\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fdstl-satellite-imagery-feature-detection\n* 难度等级：中等，有许多优秀的示例（请参阅讨论区和内核），但由于该比赛是在几年前举行的，许多示例仍使用Python 2\n* WorldView 3 - 45张卫星图像，覆盖1公里×1公里区域，既有3波段（即RGB）图像，也有16波段（400nm - SWIR）图像\n* 10个标注类别包括 - **建筑物、道路、树木、农作物、水道、车辆**\n* [对使用分割网络的第一名获奖者的访谈](http:\u002F\u002Fblog.kaggle.com\u002F2017\u002F04\u002F26\u002Fdstl-satellite-imagery-competition-1st-place-winners-interview-kyle-lee\u002F) - 使用了40多种模型，每种都针对特定目标进行了调整（例如道路、树木）\n* [ZF_UNET_224_Pretrained_Model 第二名解决方案](https:\u002F\u002Fgithub.com\u002FZFTurbo\u002FZF_UNET_224_Pretrained_Model) ->\n* [第三名解决方案](https:\u002F\u002Fgithub.com\u002Fosin-vladimir\u002Fkaggle-satellite-imagery-feature-detection) -> 探索了全色锐化和反射率指数的计算，并发表了[arxiv论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.06169)\n* [Deepsense 第四名解决方案](https:\u002F\u002Fdeepsense.ai\u002Fdeep-learning-for-satellite-imagery-via-image-segmentation\u002F)\n* [lopuhin 的参赛作品](https:\u002F\u002Fgithub.com\u002Flopuhin\u002Fkaggle-dstl) 使用带有批量归一化的UNet\n* [使用U-Net对卫星图像进行多类语义分割](https:\u002F\u002Fgithub.com\u002Frogerxujiang\u002Fdstl_unet) 使用DSTL数据集，TensorFlow 1和Python 2.7。同时附有[文章](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20240930001745\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fdstl-satellite-imagery-contest-on-kaggle-2f3ef7b8ac40)\n* [Deep-Satellite-Image-Segmentation](https:\u002F\u002Fgithub.com\u002Fantoine-spahr\u002FDeep-Satellite-Image-Segmentation)\n* [Dstl-Satellite-Imagery-Feature-Detection-Improved](https:\u002F\u002Fgithub.com\u002Fcuulee\u002FDstl-Satellite-Imagery-Feature-Detection-Improved)\n* [Satellite-imagery-feature-detection](https:\u002F\u002Fgithub.com\u002FArangurenAndres\u002FSatellite-imagery-feature-detection)\n* [Satellite_Image_Classification](https:\u002F\u002Fgithub.com\u002Faditya-sawadh\u002FSatellite_Image_Classification) -> 使用XGBoost和集成分类方法\n* [Unet-for-Satellite](https:\u002F\u002Fgithub.com\u002Fjustinishikawa\u002FUnet-for-Satellite)\n* [building-segmentation](https:\u002F\u002Fgithub.com\u002Fjimpala\u002Fbuilding-segmentation) -> TensorFlow U-Net实现，用于在卫星图像中分割建筑物\n\n### Kaggle - DeepSat 土地覆被分类\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat4 和 https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat6\n* [DeepSat-Kaggle](https:\u002F\u002Fgithub.com\u002Fathulsudheesh\u002FDeepSat-Kaggle) -> 使用 Julia 语言\n* [deepsat-aws-emr-pyspark](https:\u002F\u002Fgithub.com\u002Fhellosaumil\u002Fdeepsat-aws-emr-pyspark) -> 利用 PySpark 对农业用地的卫星影像进行图像分类\n\n### Kaggle - 空中客车船舶检测挑战赛\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fairbus-ship-detection\u002Foverview\n* 难度：中等，大多数解决方案采用深度学习，有许多内核，[一个不错的示例内核](https:\u002F\u002Fwww.kaggle.com\u002Fkmader\u002Fbaseline-u-net-model-part-1)\n* [在卫星影像中检测船舶：五年之后…](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Fdetecting-ships-in-satellite-imagery-five-years-later-28df2e83f987)\n* 我认为这个数据集存在一些问题，导致许多参赛者抱怨比赛被破坏了。\n* [从 Kaggle 的空中客车挑战赛中学到的经验](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20241002082610\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Flessons-learned-from-kaggles-airbus-challenge-252e25c5efac)\n* [Airbus-Ship-Detection](https:\u002F\u002Fgithub.com\u002Fkheyer\u002FAirbus-Ship-Detection) -> 该方案在比赛中获得了 884 个参赛者中的第 139 名，结合了基于 ResNeXt50 的分类器和 U-net 分割模型。\n* [Ship-Detection-Project](https:\u002F\u002Fgithub.com\u002FZTong1201\u002FShip-Detection-Project) -> 使用 Mask R-CNN 和 UNet 模型。\n* [Airbus_SDC](https:\u002F\u002Fgithub.com\u002FWillieMaddox\u002FAirbus_SDC)\n* [Airbus_SDC_dup](https:\u002F\u002Fgithub.com\u002FWillieMaddox\u002FAirbus_SDC_dup) -> 该项目专注于检测重叠卫星影像中的重复区域。应用于空中客车船舶检测数据集。\n* [airbus-ship-detection](https:\u002F\u002Fgithub.com\u002Fjancervenka\u002Fairbus-ship-detection) -> 带有 REST API 的 CNN。\n* [使用 YOLOV4 从卫星图像中检测船舶](https:\u002F\u002Fgithub.com\u002Fdebasis-dotcom\u002FShip-Detection-from-Satellite-Images-using-YOLOV4) -> 使用 Kaggle 空中客车船舶检测数据集。\n* [图像分割：Kaggle 经验](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20240929194243\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Fimage-segmentation-kaggle-experience-9a41cb8924f0) -> 由金牌得主 Vlad Shmyhlo 撰写的 Medium 文章。\n\n### Kaggle - 谷歌地球中的船只\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ftomluther\u002Fships-in-google-earth\n* 794 张 JPEG 图像，展示了卫星影像中各种大小的船只，标注格式为 Pascal VOC，适用于目标检测模型。\n* [\u002Fkaggle-ships-in-satellite-imagery-with-YOLOv8](https:\u002F\u002Fgithub.com\u002Frobmarkcole\u002Fkaggle-ships-in-satellite-imagery-with-YOLOv8)\n\n### Kaggle - 使用 Planet 卫星影像对旧金山湾的船只进行分类\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Frhammell\u002Fships-in-satellite-imagery\n* 4000 张 80x80 的 RGB 图像，标记为“船”或“非船”，像素尺寸为 3 米。\n* [shipsnet-detector](https:\u002F\u002Fgithub.com\u002Frhammell\u002Fshipsnet-detector) -> 使用机器学习检测 Planet 影像中的集装箱船。\n* [DeepLearningShipDetection](https:\u002F\u002Fgithub.com\u002FPenguinDan\u002FDeepLearningShipDetection)\n* [Ship-Detection-Using-Satellite-Imagery](https:\u002F\u002Fgithub.com\u002FDhruvisha29\u002FShip-Detection-Using-Satellite-Imagery)\n\n### Kaggle - Planesnet 分类数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Frhammell\u002Fplanesnet -> 检测 Planet 卫星影像芯片中的飞机。\n* 20x20 的 RGB 图像，“飞机”类别包含 8000 张图像，“非飞机”类别包含 24000 张图像。\n* [数据集仓库](https:\u002F\u002Fgithub.com\u002Frhammell\u002Fplanesnet) 和 [planesnet-detector](https:\u002F\u002Fgithub.com\u002Frhammell\u002Fplanesnet-detector) 展示了在此数据集上训练的小型 CNN 分类器。\n* [ergo-planes-detector](https:\u002F\u002Fgithub.com\u002Fevilsocket\u002Fergo-planes-detector) -> 一个基于 ergo 的项目，依赖于卷积神经网络来检测卫星影像中的飞机，使用 PlanesNet 数据集。\n* [使用 AWS SageMaker\u002FPlanesNet 处理卫星影像](https:\u002F\u002Fgithub.com\u002Fkskalvar\u002Faws-sagemaker-planesnet-imagery)\n* [Airplane-in-Planet-Image](https:\u002F\u002Fgithub.com\u002FMaxLenormand\u002FAirplane-in-Planet-Image) -> PyTorch 模型。\n\n### Kaggle - 带边界框的 CGI 卫星影像中的飞机\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Faceofspades914\u002Fcgi-planes-in-satellite-imagery-w-bboxes\n* 500 张计算机生成的飞机卫星图像。\n* [使用 Faster RCNN 检测飞机](https:\u002F\u002Fgithub.com\u002FShubhankarRawat\u002FAirplane-Detection-for-Satellites)\n* [aircraft-detection-from-satellite-images-yolov3](https:\u002F\u002Fgithub.com\u002Femrekrtorun\u002Faircraft-detection-from-satellite-images-yolov3)\n\n### Kaggle - 使用卫星影像检测游泳池和汽车\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fkbhartiya83\u002Fswimming-pool-and-car-detection\n* 3750 张住宅区的卫星影像，附带游泳池和汽车的标注数据。\n* [使用 RetinaNet 在卫星影像上进行目标检测](https:\u002F\u002Fmedium.com\u002F@ije_good\u002Fobject-detection-on-satellite-imagery-using-retinanet-part-1-training-e589975afbd5)\n\n### Kaggle - Draper 挑战赛：按时间顺序排列图像\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fdraper-satellite-image-chronology\u002Fdata\n* 难度：困难。有用的内核不多。\n* 图像被分成每组 5 张，每组都有相同的 setId。同一组中的每张图像都是在不同日期拍摄的（但不一定每天同一时间）。每组图像覆盖的区域大致相同，但并不完全对齐。\n* Kaggle 对使用 [XGBOOST](http:\u002F\u002Fblog.kaggle.com\u002F2016\u002F09\u002F15\u002Fdraper-satellite-image-chronology-machine-learning-solution-vicens-gaitan\u002F) 和 [人类\u002FML 混合方法](http:\u002F\u002Fblog.kaggle.com\u002F2016\u002F09\u002F08\u002Fdraper-satellite-image-chronology-damien-soukhavong\u002F) 的参赛者进行了采访。\n* [deep-cnn-sat-image-time-series](https:\u002F\u002Fgithub.com\u002FMickyDowns\u002Fdeep-cnn-sat-image-time-series) -> 使用 LSTM。\n\n### Kaggle - 迪拜语义分割\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fhumansintheloop\u002Fsemantic-segmentation-of-aerial-imagery\n* 72 张阿联酋迪拜的卫星影像，被分割成 6 个类别。\n* [dubai-satellite-imagery-segmentation](https:\u002F\u002Fgithub.com\u002Fayushdabra\u002Fdubai-satellite-imagery-segmentation) -> 由于数据集较小，使用了图像增强技术。\n* [使用 U-Net 对不平衡航空影像进行语义分割](https:\u002F\u002Ftowardsdatascience.com\u002Fu-net-for-semantic-segmentation-on-unbalanced-aerial-imagery-3474fa1d3e56) -> 使用迪拜数据集。\n* [Semantic-Segmentation-using-U-Net](https:\u002F\u002Fgithub.com\u002FAnay21110\u002FSemantic-Segmentation-using-U-Net) -> 使用 Keras。\n* [unet_satelite_image_segmentation](https:\u002F\u002Fgithub.com\u002Fnassimaliou\u002Funet_satelite_image_segmentation)\n\n### Kaggle - 马萨诸塞州道路与建筑数据集 - 分割\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fmassachusetts-roads-dataset\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fmassachusetts-buildings-dataset\n* [官方发布的数据集](https:\u002F\u002Fwww.cs.toronto.edu\u002F~vmnih\u002Fdata\u002F)\n* [Road_seg_dataset](https:\u002F\u002Fgithub.com\u002Fparth1620\u002FRoad_seg_dataset) -> 道路数据集的一个子集，仅包含200张图像和对应的掩码\n* [卫星影像中的道路和建筑语义分割](https:\u002F\u002Fgithub.com\u002FPaulymorphous\u002FRoad-Segmentation) 使用U-Net模型和Keras框架处理马萨诸塞州道路数据集\n* [fuweifu-vtoo的语义分割仓库](https:\u002F\u002Fgithub.com\u002Ffuweifu-vtoo\u002FSemantic-segmentation) -> 使用PyTorch框架，并基于[马萨诸塞州建筑与道路数据集](https:\u002F\u002Fwww.cs.toronto.edu\u002F~vmnih\u002Fdata\u002F)\n* [ssai-cnn](https:\u002F\u002Fgithub.com\u002Fmitmul\u002Fssai-cnn) -> 这是Volodymyr Mnih在其博士论文中提出的方法在马萨诸塞州道路与建筑数据集上的实现\n* [building-footprint-segmentation](https:\u002F\u002Fgithub.com\u002Ffuzailpalnak\u002Fbuilding-footprint-segmentation) -> 一个可通过pip安装的库，用于训练卫星和航空影像中的建筑物轮廓分割模型，应用于马萨诸塞州建筑数据集和Inria航空影像标注数据集\n* [使用语义分割和Albumentations数据增强进行道路检测](https:\u002F\u002Fweb.archive.org\u002Fweb\u002F20240929191243\u002Fhttps:\u002F\u002Ftowardsdatascience.com\u002Froad-detection-using-segmentation-models-and-albumentations-libraries-on-keras-d5434eaf73a8)，使用马萨诸塞州道路数据集、U-Net和Keras\n* [Image-Segmentation)](https:\u002F\u002Fgithub.com\u002Fmschulz\u002FImage-Segmentation) -> 使用马萨诸塞州道路数据集和fast.ai框架\n\n### Kaggle - Deepsat分类挑战赛\n这不是卫星影像，而是航空影像。每个样本图像为28x28像素，包含红、绿、蓝和近红外四个波段。训练和测试标签为1x6的独热编码向量。每张图像块都被归一化为28x28像素大小。数据以`.mat`格式存储，可能是JPEG格式？\n* [Sat4](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat4) 包含50万张图像块，覆盖四大类地表覆盖类型——**荒地、树木、草地以及其他所有不属于前三种的地表覆盖类型**\n* [Sat6](https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcrawford\u002Fdeepsat-sat6) 包含40.5万张28x28像素大小的图像块，覆盖六种地表覆盖类型——**荒地、树木、草地、道路、建筑物和水体**\n\n### Kaggle - 高分辨率舰船数据集2016 (HRSC2016)\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fguofeng\u002Fhrsc2016\n* 从Google Earth上采集的舰船图像\n* [HRSC2016_SOTA](https:\u002F\u002Fgithub.com\u002Fming71\u002FHRSC2016_SOTA) -> 对HRSC2016数据集上不同算法的公平比较\n\n### Kaggle - SWIM-船舶尾迹影像马萨诸塞州\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Flilitopia\u002Fswimship-wake-imagery-mass\n* 一个专为深度学习构建的光学船舶尾迹检测基准数据集\n* [WakeNet](https:\u002F\u002Fgithub.com\u002FLilytopia\u002FWakeNet) -> 基于CNN的光学图像船舶尾迹检测器，代码对应2021年的论文：重新思考自动船舶尾迹检测：基于光学图像的最新CNN尾迹检测技术\n\n### Kaggle - 从卫星图像理解云层\n在这个挑战赛中，你需要构建一个模型来对卫星图像中的云层组织模式进行分类。\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Funderstanding_cloud_organization\u002F\n* [naivelamb在Github上提交的第三名解决方案](https:\u002F\u002Fgithub.com\u002Fnaivelamb\u002Fkaggle-cloud-organization)\n* [Soongja在Github上提交的第15名解决方案](https:\u002F\u002Fgithub.com\u002FSoongja\u002Fkaggle-clouds)\n* [yukkyo在Github上提交的第69名解决方案](https:\u002F\u002Fgithub.com\u002Fyukkyo\u002FKaggle-Understanding-Clouds-69th-solution)\n* [michal-nahlik在Github上提交的第161名解决方案](https:\u002F\u002Fgithub.com\u002Fmichal-nahlik\u002Fkaggle-clouds-2019)\n* [yurayli的解决方案](https:\u002F\u002Fgithub.com\u002Fyurayli\u002Fsatellite-cloud-segmentation)\n* [HazelMartindale的解决方案](https:\u002F\u002Fgithub.com\u002FHazelMartindale\u002Fkaggle_understanding_clouds_learning_project) 使用了三种不同版本的U-Net架构\n* [khornlund的解决方案](https:\u002F\u002Fgithub.com\u002Fkhornlund\u002Funderstanding-cloud-organization)\n* [Diyago的解决方案](https:\u002F\u002Fgithub.com\u002FDiyago\u002FUnderstanding-Clouds-from-Satellite-Images)\n* [tanishqgautam的解决方案](https:\u002F\u002Fgithub.com\u002Ftanishqgautam\u002FMulti-Label-Segmentation-With-FastAI)\n\n### Kaggle - 38-Cloud云分割\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsorour\u002F38cloud-cloud-segmentation-in-satellite-images\n* 包含38张Landsat 8影像以及手动提取的像素级真值\n* [38-Cloud GitHub仓库](https:\u002F\u002Fgithub.com\u002FSorourMo\u002F38-Cloud-A-Cloud-Segmentation-Dataset) 及其后续的[95-Cloud](https:\u002F\u002Fgithub.com\u002FSorourMo\u002F95-Cloud-An-Extension-to-38-Cloud-Dataset)数据集\n* [如何从零开始在PyTorch中为Kaggle上的多波段卫星图像数据集创建自定义数据集\u002F加载器](https:\u002F\u002Fmedium.com\u002Fanalytics-vidhya\u002Fhow-to-create-a-custom-dataset-loader-in-pytorch-from-scratch-for-multi-band-satellite-images-c5924e908edf)\n* [Cloud-Net: 一种用于云检测的语义分割CNN](https:\u002F\u002Fgithub.com\u002FSorourMo\u002FCloud-Net-A-semantic-segmentation-CNN-for-cloud-detection) -> 一种端到端的云检测算法，适用于Landsat 8影像，并在38-Cloud训练集上进行训练\n* [利用深度学习对卫星图像中的云层进行分割](https:\u002F\u002Fmedium.com\u002Fswlh\u002Fsegmentation-of-clouds-in-satellite-images-using-deep-learning-a9f56e0aa83d) -> 使用Unet模型对Kaggle 38-Cloud数据集进行语义分割\n\n### Kaggle - 空客飞机检测数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fairbusgeo\u002Fairbus-aircrafts-sample-dataset\n* 包括一百个民用机场和超过3000架已标注的商用飞机\n* [使用YOLOv5在空客Pleiades影像上检测飞机](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Fdetecting-aircrafts-on-airbus-pleiades-imagery-with-yolov5-5f3d464b75ad)\n* [pytorch-remote-sensing](https:\u002F\u002Fgithub.com\u002Fmiko7879\u002Fpytorch-remote-sensing) -> 使用“空客飞机检测”数据集和PyTorch中的Faster-RCNN模型（骨干网络为ResNet-50）进行飞机检测\n\n### Kaggle - 空客石油储存检测数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fairbusgeo\u002Fairbus-oil-storage-detection-dataset\n* [使用Mask R-CNN进行油罐实例分割](https:\u002F\u002Fgithub.com\u002Fgeorgiosouzounis\u002Finstance-segmentation-mask-rcnn\u002Fblob\u002Fmain\u002Fmask_rcnn_oiltanks_gpu.ipynb)，并配有相关文章[使用Mask R-CNN进行油罐实例分割](https:\u002F\u002Fmedium.com\u002F@georgios.ouzounis\u002Foil-storage-tank-instance-segmentation-with-mask-r-cnn-77c94433045f)\n* [使用YOLOX在空客影像上检测石油储存设施](https:\u002F\u002Fmedium.com\u002Fartificialis\u002Foil-storage-detection-on-airbus-imagery-with-yolox-9e38eb6f7e62) -> 使用Kaggle空客石油储存检测数据集\n* [油罐数据准备-YOLO格式](https:\u002F\u002Fgithub.com\u002Fshah0nawaz\u002FOil-Storage-Tanks-Data-Preparation-YOLO-Format)\n\n### Kaggle - 飓风灾害卫星影像\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fkmader\u002Fsatellite-images-of-hurricane-damage\n* https:\u002F\u002Fgithub.com\u002Fdbuscombe-usgs\u002FHurricaneHarvey_buildingdamage\n\n### Kaggle - 奥斯汀分区卫星图像\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ffranchenstein\u002Faustin-zoning-satellite-images\n* 将奥斯汀的图像分类到其各个区域，例如住宅区、工业区等。共有3667张卫星图像。\n\n### Kaggle - Statoil\u002FC-CORE 冰山分类挑战赛\n将SAR图像中的目标分类为船只或冰山。该竞赛的数据集包含5000张从Sentinel-1卫星采集的多通道SAR数据中提取的图像。排名靠前的参赛者通过集成方法将预测准确率从约92%提升至97%。\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fstatoil-iceberg-classifier-challenge\u002Fdata\n* [对大卫·奥斯汀的采访：第一名获奖者](https:\u002F\u002Fpyimagesearch.com\u002F2018\u002F03\u002F26\u002Finterview-david-austin-1st-place-25000-kaggles-popular-competition\u002F)\n* [radar-image-recognition](https:\u002F\u002Fgithub.com\u002Fsiarez\u002Fradar-image-recognition)\n* [Iceberg-Classification-Using-Deep-Learning](https:\u002F\u002Fgithub.com\u002Fmankadronit\u002FIceberg-Classification-Using-Deep-Learning) -> 使用Keras\n* [Deep-Learning-Project](https:\u002F\u002Fgithub.com\u002Fsingh-shakti94\u002FDeep-Learning-Project) -> 使用Keras\n* [shehabsunny的冰山分类挑战赛解决方案](https:\u002F\u002Fgithub.com\u002FShehabSunny\u002Ficeberg-classifier-challenge) -> 使用Keras\n* [利用深度学习分析卫星雷达影像](https:\u002F\u002Fuk.mathworks.com\u002Fcompany\u002Fnewsletters\u002Farticles\u002Fanalyzing-satellite-radar-imagery-with-deep-learning.html) -> 由Matlab实现，使用带有贪心搜索的集成方法\n* [第16名的解决方案](https:\u002F\u002Fgithub.com\u002Fsergeyshilin\u002Fkaggle-statoil-iceberg-classifier-challenge)\n* [fastai解决方案](https:\u002F\u002Fgithub.com\u002Fsmarkochev\u002Fds_notebooks\u002Fblob\u002Fmaster\u002FStatoil_Kaggle_competition_google_colab_notebook.ipynb)\n\n### Kaggle - DeepGlobe挑战赛的土地覆盖分类数据集 - 分割\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fdeepglobe-land-cover-classification-dataset\n* [使用CNN进行卫星影像语义分割](https:\u002F\u002Fjoshting.medium.com\u002Fsatellite-imagery-segmentation-with-convolutional-neural-networks-f9254de3b907) -> 7种不同的分割类别，基于DeepGlobe土地覆盖分类挑战赛的数据集，并附有[代码库](https:\u002F\u002Fgithub.com\u002Fjustjoshtings\u002Fsatellite_image_segmentation)\n* [使用U-Net进行土地覆盖分类](https:\u002F\u002Fbaratam-tarunkumar.medium.com\u002Fland-cover-classification-with-u-net-aa618ea64a1b) -> 使用PyTorch实现的U-Net进行卫星影像多类语义分割任务，采用DeepGlobe土地覆盖分割数据集，并提供[代码](https:\u002F\u002Fgithub.com\u002FTarunKumar1995-glitch\u002Fland_cover_classification_unet)\n* [DeepGlobe土地覆盖分类挑战赛的解决方案](https:\u002F\u002Fgithub.com\u002FGeneralLi95\u002Fdeepglobe_land_cover_classification_with_deeplabv3plus)\n\n### Kaggle - 次日野火蔓延\n一个基于遥感数据预测野火蔓延的数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ffantineh\u002Fnext-day-wildfire-spread\n* https:\u002F\u002Farxiv.org\u002Fabs\u002F2112.02447\n\n### Kaggle - 卫星次日野火蔓延\n受上述数据集启发，使用了不同的数据源\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsatellitevu\u002Fsatellite-next-day-wildfire-spread\n* https:\u002F\u002Fgithub.com\u002FSatelliteVu\u002FSatelliteVu-AWS-Disaster-Response-Hackathon\n\n## Kaggle - Spacenet 7 多时相城市变化检测\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Famerii\u002Fspacenet-7-multitemporal-urban-development\n* [SatFootprint](https:\u002F\u002Fgithub.com\u002FPriyanK7n\u002FSatFootprint) -> 在Spacenet 7数据集上进行建筑物分割\n\n## Kaggle - 卫星图像用于预测非洲贫困\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsandeshbhat\u002Fsatellite-images-to-predict-povertyafrica\n* 利用卫星影像和夜间灯光数据来预测当地贫困水平\n* [Predicting-Poverty](https:\u002F\u002Fgithub.com\u002Fjmather625\u002Fpredicting-poverty-replication) -> 结合卫星影像和机器学习以预测贫困，使用PyTorch实现\n\n## Kaggle - NOAA渔业斯特勒海狮种群计数\n* https:\u002F\u002Fwww.kaggle.com\u002Fcompetitions\u002Fnoaa-fisheries-steller-sea-lion-population-count -> 从航拍图像中统计海狮数量\n* [Sealion-counting](https:\u002F\u002Fgithub.com\u002Fbabyformula\u002FSealion-counting)\n* [Sealion_Detection_Classification](https:\u002F\u002Fgithub.com\u002Fyyc9268\u002FSealion_Detection_Classification)\n\n## Kaggle - 北极海冰图像掩膜\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Falexandersylvester\u002Farctic-sea-ice-image-masking\n* [sea_ice_remote_sensing](https:\u002F\u002Fgithub.com\u002Fsum1lim\u002Fsea_ice_remote_sensing)\n\n## Kaggle - Overhead-MNIST\n* 作为MNIST的替代基准卫星数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fdatamunge\u002Foverheadmnist -> Kaggle\n* https:\u002F\u002Farxiv.org\u002Fabs\u002F2102.04266 -> 论文\n* https:\u002F\u002Fgithub.com\u002Freveondivad\u002Fov-mnist -> GitHub\n\n## Kaggle - 卫星图像分类\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fmahmoudreda55\u002Fsatellite-image-classification\n* [satellite-image-classification-pytorch](https:\u002F\u002Fgithub.com\u002Fdilaraozdemir\u002Fsatellite-image-classification-pytorch)\n\n## Kaggle - EuroSAT - Sentinel-2 数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fraoofnaushad\u002Feurosat-sentinel2-dataset\n* 使用Sentinel-2卫星进行RGB土地覆盖与土地利用分类\n* 用于论文[卫星图像的数据增强](https:\u002F\u002Farxiv.org\u002Fabs\u002F2207.14580)\n\n## Kaggle - 水体卫星图像\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ffranciscoescobar\u002Fsatellite-images-of-water-bodies\n* [pytorch-waterbody-segmentation](https:\u002F\u002Fgithub.com\u002Fgauthamk02\u002Fpytorch-waterbody-segmentation) -> 基于Kaggle水体卫星图像数据集训练的UNET模型。该模型已部署在Hugging Face Spaces上。\n\n## Kaggle - NOAA海狮计数\n* https:\u002F\u002Fwww.kaggle.com\u002Fc\u002Fnoaa-fisheries-steller-sea-lion-population-count\n* [noaa](https:\u002F\u002Fgithub.com\u002Fdarraghdog\u002Fnoaa) -> 使用UNET、目标检测及图像级回归方法\n\n### Kaggle - 杂项\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Freubencpereira\u002Fspatial-data-repo -> 卫星影像 + 贷款数据\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ftowardsentropy\u002Foil-storage-tanks -> 工业油罐的图像数据，附有边界框标注，可通过阴影估算油罐的填充百分比\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fairbusgeo\u002Fairbus-wind-turbines-patches -> 空客SPOT卫星拍摄的风力涡轮机区域图像，用于分类任务\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Faceofspades914\u002Fcgi-planes-in-satellite-imagery-w-bboxes -> CGI飞机目标检测数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fatilol\u002Faerialimageryforroofsegmentation -> 用于屋顶分割的航拍影像\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fandrewmvd\u002Fship-detection -> 621张船只和舰艇的图片\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Falpereniek\u002Fvehicle-detection-from-satellite-images-data-set\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsergiishchus\u002Fmaxar-satellite-data -> Maxar公司提供的示例数据，分辨率为15厘米\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fcici118\u002Fswimming-pool-detection-algarves-landscape\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fdonkroco\u002Fsolar-panel-module -> 太阳能电池板的目标检测\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fbalraj98\u002Fdeepglobe-road-extraction-dataset -> 道路分割\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Ftowardsentropy\u002Foil-storage-tanks -> 工业储油罐的图像数据，附有边界框标注\n* https:\u002F\u002Fwww.kaggle.com\u002Fcompetitions\u002Fwidsdatathon2019\u002F -> 棕榈油种植园\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fsiddharthkumarsah\u002Fships-in-aerial-images -> 航拍图像中的船只\u002F船舶\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fjangsienicajzkowy\u002Fafo-aerial-dataset-of-floating-objects -> 用于海上搜救应用的航拍数据集\n* https:\u002F\u002Fwww.kaggle.com\u002Fdatasets\u002Fyaroslavnaychuk\u002Fsatelliteimagesegmentation -> 对高分卫星影像进行分割，数据来自GID-15数据集\n\n# 竞赛\n竞赛是获取干净、可直接使用的卫星数据集及模型基准测试的绝佳途径。\n\n* https:\u002F\u002Fcodalab.lisn.upsaclay.fr\u002Fcompetitions\u002F9603 -> 多样化卫星影像的目标检测\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F143\u002Ftick-tick-bloom\u002F -> 检测并分类藻华\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F81\u002Fdetect-flood-water\u002F -> 根据雷达影像绘制洪水范围图\n* https:\u002F\u002Fplatform.ai4eo.eu\u002Fenhanced-sentinel2-agriculture -> 利用Sentinel影像绘制耕地分布图\n* https:\u002F\u002Fwww.diu.mil\u002Fai-xview-challenge -> 多个挑战，涵盖从渔船检测到建筑物损毁评估等多个方向\n* https:\u002F\u002Fcompetitions.codalab.org\u002Fcompetitions\u002F30440 -> 洪水检测\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F83\u002Fcloud-cover\u002F -> 云层覆盖检测\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F78\u002Foverhead-geopose-challenge\u002Fpage\u002F372\u002F -> 基于单视角斜视卫星影像预测地心姿态\n* https:\u002F\u002Fwww.drivendata.org\u002Fcompetitions\u002F60\u002Fbuilding-segmentation-disaster-resilience\u002F -> 建筑物分割\n* https:\u002F\u002Fcaptain-whu.github.io\u002FDOTA\u002F -> 用于航拍影像中目标检测的大型数据集\n* https:\u002F\u002Fspacenet.ai\u002F -> 包含道路网络检测等8个挑战的数据集\n* https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fcompetitions\u002FChaBuD-ECML-PKDD2023 -> 针对加利福尼亚州森林火灾监测的二值图像分割任务\n\u003C!-- markdown-link-check-disable -->\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F6269285b14d764000d798fde -> 用于洪水相关的机器学习\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F60002402f5647f00129f7287 -> 闪电与极端天气\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F6025107d79c197001219c481\u002Ftrue -> 约1TB的降水预报数据集\n* https:\u002F\u002Fspaceml.org\u002Frepo\u002Fproject\u002F61c0a1b9ff8868000dfb79e1\u002Ftrue -> Sentinel-2影像超分辨率\n\u003C!-- markdown-link-check-enable --","# Datasets 快速上手指南\n\n本指南旨在帮助开发者快速了解并利用该仓库中整理的卫星与航空影像深度学习数据集资源。请注意，**`datasets` 本身是一个 curated list（精选列表）而非一个可直接安装的 Python 库**。以下指南将指导你如何获取环境、查找数据以及使用相关工具加载数据。\n\n## 环境准备\n\n在开始使用前，请确保你的开发环境满足以下要求：\n\n*   **操作系统**：Linux (推荐), macOS, 或 Windows (WSL2 推荐)。\n*   **Python 版本**：建议 Python 3.8 及以上版本。\n*   **前置依赖**：\n    *   **Git**：用于克隆仓库或下载脚本。\n    *   **深度学习框架**：根据你计划使用的具体数据集模型，通常需要 `PyTorch` 或 `TensorFlow`。\n    *   **地理空间处理库**：处理卫星影像通常需要的库，如 `rasterio`, `geopandas`, `xarray`, `torchgeo`。\n    *   **云存储工具**：许多数据集托管在 AWS S3 或 Google Cloud Storage，建议安装 `awscli` 或配置好 `gcloud` SDK。\n\n**推荐安装基础地理空间与深度学习栈：**\n\n```bash\npip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\npip install rasterio geopandas xarray torchgeo huggingface_hub\n```\n\n> **提示**：国内开发者可使用清华源或阿里源加速 Python 包安装：\n> `pip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple \u003Cpackage_name>`\n\n## 安装步骤\n\n由于本项目是数据集索引列表，无需执行传统的 `pip install datasets`。你需要做的是获取列表并定位到你需要的数据源。\n\n### 1. 克隆或浏览仓库\n你可以直接在线浏览 [satellite-image-deep-learning.com](https:\u002F\u002Fwww.satellite-image-deep-learning.com\u002F) 或使用 `Control+F` 在页面中搜索关键词（如 \"Sentinel-2\", \"Ship detection\", \"Cloud removal\"）。\n\n若需离线查看或贡献，可克隆仓库：\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fchrieke\u002Fawesome-satellite-imagery-datasets.git\n# 注意：实际仓库地址请以项目最新为准，此处为示例逻辑\n```\n\n### 2. 获取特定数据集\n找到目标数据集后，根据其提供的链接进行“安装”（通常是下载数据或配置访问权限）。常见方式包括：\n\n*   **Hugging Face Datasets**: 许多现代数据集托管在 HF 上。\n    ```bash\n    pip install datasets\n    ```\n*   **AWS Open Data \u002F Google Cloud**: 使用 CLI 工具同步数据。\n    ```bash\n    aws s3 sync s3:\u002F\u002Fsentinel-cogs\u002Fsentinel-s2-l2a-cogs\u002F .\u002Flocal_data_path --no-sign-request\n    ```\n*   **Zenodo \u002F Kaggle**: 通过网页下载或使用其官方 API 客户端。\n\n## 基本使用\n\n以下展示如何使用 Python 加载一个典型的卫星影像数据集（以 Hugging Face 托管的 Sentinel 相关数据集为例，这是目前最通用的方式）。\n\n### 示例：加载 Sentinel-2 数据集\n\n假设你找到了一个托管在 Hugging Face 上的数据集（例如 `tacofoundation\u002FSEN2NAIPv2` 或其他类似数据集）：\n\n```python\nfrom datasets import load_dataset\n\n# 1. 加载数据集\n# 替换为你在列表中找到的具体 dataset_id\ndataset = load_dataset(\"tacofoundation\u002FSEN2NAIPv2\", split=\"train\")\n\n# 2. 查看数据结构\nprint(dataset.features)\nprint(dataset[0])\n\n# 3. 简单的数据迭代\nfor item in dataset:\n    # item 通常包含图像张量、标签、元数据等\n    image = item['image'] \n    label = item['label']\n    \n    # 在此处添加你的 PyTorch\u002FTensorFlow 预处理逻辑\n    break\n```\n\n### 示例：使用 TorchGeo 加载本地\u002F云端数据\n\n如果你下载了原始 GeoTIFF 文件（如来自 AWS Open Data 的 Sentinel-2 L2A），推荐使用 `torchgeo` 库进行标准化加载：\n\n```python\nfrom torchgeo.datasets import Sentinel2\nfrom torchgeo.samplers import RandomGeoSampler\nfrom torch.utils.data import DataLoader\n\n# 定义数据根目录和波段\nroot = \".\u002Fsentinel_data\"\nbands = [\"B04\", \"B03\", \"B02\"] # RGB\n\n# 初始化数据集\ndataset = Sentinel2(root=root, bands=bands, transforms=None)\n\n# 创建采样器 (需要 geo 边界信息)\nsampler = RandomGeoSampler(dataset, size=256, length=100)\n\n# 构建 DataLoader\ndataloader = DataLoader(dataset, batch_size=4, sampler=sampler, collate_fn=lambda x: x)\n\n# 开始训练循环\nfor batch in dataloader:\n    # 处理批次数据\n    pass\n```\n\n### 关键资源索引\n在使用时，请回到原列表查找以下特定领域的专用数据集链接：\n*   **变化检测 (Change Detection)**: 搜索 `awesome-remote-sensing-change-detection`\n*   **云层检测 (Cloud Detection)**: 参考 `CloudSEN12Plus` 或 `Azavea Cloud Dataset`\n*   **船舶检测 (Ship Detection)**: 参考 `S2-SHIPS` 或 `Ship-S2-AIS`\n*   **农业分类 (Crop Classification)**: 参考 `Sen4AgriNet` 或 `TimeSen2Crop`\n\n通过上述步骤，你可以快速定位并集成适合你任务的卫星影像数据到深度学习工作流中。","某农业科技公司数据团队正致力于开发基于卫星影像的作物产量预测模型，急需整合多源遥感数据以训练高精度深度学习算法。\n\n### 没有 datasets 时\n- 数据搜集极其耗时，工程师需手动在 AWS、Google Earth Engine 及各类论文附录中分散查找 Sentinel-1\u002F2 数据，往往数周无法凑齐实验所需样本。\n- 数据格式混乱且不统一，不同来源的影像分辨率、坐标系和预处理标准各异，导致大量时间浪费在清洗和对齐数据上，而非模型优化。\n- 缺乏权威基准测试集，团队难以验证新算法的有效性，无法与业界最新成果（如 SEN12MS 或 M3LEO）进行公平对比，研发方向容易偏离。\n- 特定任务数据稀缺，针对洪水监测或超分辨率等细分场景，很难找到带有高质量标注的现成数据集，迫使团队从零开始标注，成本高昂。\n\n### 使用 datasets 后\n- 一键获取丰富资源，通过索引直接定位到 mmflood 洪水数据集或 CYCleSS 作物产量数据，将数据准备周期从数周缩短至数小时。\n- 标准化数据流无缝接入，直接调用已预处理的 Sentinel-2 COGs 或 SEN12MS 融合数据集，确保输入数据格式统一，让团队能立即投入模型训练。\n- 依托权威基准快速迭代，利用 Radiant MLHub 或 Awesome_Satellite_Benchmark_Datasets 中的标准集进行评测，迅速明确模型性能差距并调整策略。\n- 细分场景即时可用，针对变化检测或森林覆盖分析，直接复用专门的开源集合，大幅降低标注成本并加速原型验证。\n\ndatasets 通过聚合全球优质遥感数据资源，彻底消除了数据获取与预处理的壁垒，让开发者能专注于核心算法创新与应用落地。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsatellite-image-deep-learning_datasets_a7e5a8ea.png","satellite-image-deep-learning","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fsatellite-image-deep-learning_9c7a8bfc.png","",null,"robmarkcole@gmail.com","https:\u002F\u002Fwww.satellite-image-deep-learning.com\u002F","https:\u002F\u002Fgithub.com\u002Fsatellite-image-deep-learning",1125,127,"2026-04-15T07:41:39",1,"未说明",{"notes":86,"python":84,"dependencies":87},"该 README 内容并非针对名为 'datasets' 的可执行软件工具，而是一个卫星与航空影像深度学习数据集的汇总列表（Awesome List）。它提供了指向各种数据集、论文、代码库和数据枢纽（如 AWS, Google Earth Engine）的链接，因此不包含具体的操作系统、GPU、内存、Python 版本或依赖库等运行环境需求。用户需根据列表中具体选定的某个数据集或其关联的代码库去查询相应的环境要求。",[],[15,16],[65,90,91,92,93,94],"remote-sensing","earth-observation","satellite-data","satellite-imagery","sentinel","2026-03-27T02:49:30.150509","2026-04-18T14:24:33.908982",[98,103],{"id":99,"question_zh":100,"answer_zh":101,"source_url":102},40095,"遥感图像中为什么会出现阴影？这是否意味着它们不是真正的遥感图像？","遥感图像中出现阴影是正常的。图像可能以一定角度拍摄，或者拍摄时太阳处于不同位置，因此会产生阴影。此外，“遥感”不仅包含卫星图像，也包括从飞机等航空器上拍摄的光学图像。","https:\u002F\u002Fgithub.com\u002Fsatellite-image-deep-learning\u002Fdatasets\u002Fissues\u002F8",{"id":104,"question_zh":105,"answer_zh":106,"source_url":107},40096,"如何添加新的数据集（例如 OpenSatMap）到项目中？","维护者通常通过提交代码 commit 的方式添加新数据集。例如，OpenSatMap 数据集已通过以下提交记录被加入：https:\u002F\u002Fgithub.com\u002Fsatellite-image-deep-learning\u002Fdatasets\u002Fcommit\u002F06513c649f18a05ee58fbc585cf06b3217aee5d0。用户可以参考该提交了解数据集的集成方式或直接访问 Hugging Face 链接获取数据。","https:\u002F\u002Fgithub.com\u002Fsatellite-image-deep-learning\u002Fdatasets\u002Fissues\u002F12",[]]