Datasets for deep learning applied to satellite and aerial imagery.

👉 satellite-image-deep-learning.com 👈

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).

Lists of datasets

• Earth Observation Database

• awesome-satellite-imagery-datasets
• Awesome_Satellite_Benchmark_Datasets
• awesome-remote-sensing-change-detection -> dedicated to change detection
• Callisto-Dataset-Collection -> datasets that use Copernicus/sentinel data
• geospatial-data-catalogs -> A list of open geospatial datasets available on AWS, Earth Engine, Planetary Computer, and STAC Index
• BED4RS
• Satellite-Image-Time-Series-Datasets

Remote sensing dataset hubs

• Radiant MLHub -> both datasets and models
• Registry of Open Data on AWS
• Microsoft Planetary Computer data catalog
• Google Earth Engine Data Catalog

Sentinel

As part of the EU Copernicus program, multiple Sentinel satellites are capturing imagery -> see wikipedia

Sentinel-1 (SAR)

• Xarray backend to Copernicus Sentinel-1 satellite data products
• mmflood -> Flood delineation from Sentinel-1 SAR imagery
• Sentinel-1 for Science Amazonas -> forest lost time series dataset
• CYCleSS -> A comprehensive UK crop yield dataset incorporating satellite, weather, and soil type informations
• DeepOWT -> dataset of offshore wind energy infrastructure locations and their temporal deployment dynamics on a global scale.
• Sentinel-1 SAR Oil Spill Image Dataset, Part I -> training and validation data for oil-spill detection and segmentation, comprising 2048×2048×2 Sentinel-1 Sigma0 images in decibels and 2048×2048 ground-truth masks in TIFF format
• OPD v1.0.1 -> Sentinel-1-derived Offshore Platform Dataset providing geospatial locations, lifespans, and spatial attributes of offshore oil and gas platforms
• OpenSARShip Ship Detection Dataset (OSSDD) -> 15,197 Sentinel-1 GRD amplitude patches in VV and VH polarization with binary ship masks and axis-aligned and rotated bounding boxes for 55,759 ships. Paper: OSSDD - a New Open Dataset for Sentinel-1 Ship Detection

Sentinel-2 (Optical)

• Sentinel-2 Cloud-Optimized GeoTIFFs and Sentinel-2 L2A 120m Mosaic

• Open access data on GCP

• Example loading sentinel data in a notebook

• Analyzing Sentinel-2 satellite data in Python with Keras

• SEN2VENµS -> a dataset for the training of Sentinel-2 super-resolution algorithms

• SEN2NEON -> a multispectral Sentinel-2 super-resolution benchmark with 2,269 spatially aligned tiles derived from AVIRIS-NG hyperspectral imagery

• M3LEO -> Github. A very large scale georeferenced dataset of Sentinel 1/2 imagery plus interferometric SAR products and auxiliary datasets such as Land cover, Biomass and Digital Elevation Models.

• SEN12MS -> A Curated Dataset of Georeferenced Multi-spectral Sentinel-1/2 Imagery for Deep Learning and Data Fusion. Checkout SEN12MS toolbox and many referenced uses on paperswithcode.com

• SEN2NAIP -> Spatially and spectrally harmonized Sen-2 + NAIP dataset for 4x RGB-NIR super-resolution.

• SEN2NEON -> Spatially and spectrally harmonized Sen-2 + Multispectral NEON for super-resolution of all Sen2 bands at 2.5m, including HR reference

• LR-HR-dataset -> paired Low Resolution Sentinel-2 and High Resolution PlanetScope patches (333x333x3) over the YarHussain region, Swabi District, Pakistan, suitable for super-resolution, land cover classification, and geospatial analyses

• Sen4AgriNet -> A Sentinel-2 multi-year, multi-country benchmark dataset for crop classification and segmentation with deep learning, with and models

• sentinel2tools -> downloading & basic processing of Sentinel 2 imagesry. Read Sentinel2tools: simple lib for downloading Sentinel-2 satellite images

• open-sentinel-map -> The OpenSentinelMap dataset contains Sentinel-2 imagery and per-pixel semantic label masks derived from OpenStreetMap

• Canadian-cropland-dataset -> a novel patch-based dataset compiled using optical satellite images of Canadian agricultural croplands retrieved from Sentinel-2

• Sentinel-2 Cloud Cover Segmentation Dataset on Radiant mlhub

• The Azavea Cloud Dataset which is used to train this cloud-model

• fMoW-Sentinel -> 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

• Earth Surface Water Dataset -> a dataset for deep learning of surface water features on Sentinel-2 satellite images. See this ref using it in torchgeo

• Ship-S2-AIS dataset -> 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

• Amazon Rainforest dataset for semantic segmentation -> Sentinel 2 images

• MATTER -> a Sentinel 2 dataset for Self-Supervised Training

• S2GLC -> High resolution Land Cover Map of Europe

• Generating Imperviousness Maps from Multispectral Sentinel-2 Satellite Imagery

• Sentinel-2 Water Edges Dataset (SWED)

• Sentinel2 Munich480 -> dataset for crop mapping by exploiting the time series of Sentinel-2 satellite

• Meadows vs Orchards -> a pixel time series dataset

• Sentinel-2 Image Time Series for Crop Mapping -> data for the Lombardy region in Italy

• Deforestation in Ukraine from Sentinel2 data

• satellite-change-events -> CaiRoad & CalFire change detection Sentinel 2 datasets

• Sentinel-2 dataset for ship detection, also edited and redistributed as VDS2RAW

• MineSegSAT -> dataset for paper: AN AUTOMATED SYSTEM TO EVALUATE MINING DISTURBED AREA EXTENTS FROM SENTINEL-2 IMAGERY

• CaBuAr -> California Burned Areas dataset for delineation

• sen12mscr -> Multimodal Cloud Removal

• Greenearthnet -> dataset specifically designed for high-resolution vegetation forecasting

• Floating-Marine-Debris-Data -> floating marine debris, with annotations for six debris classes, including plastic, driftwood, seaweed, pumice, sea snot, and sea foam.

• Sen2Fire -> A Challenging Benchmark Dataset for Wildfire Detection using Sentinel Data

• L1BSR -> 3740 pairs of overlapping image crops extracted from two L1B products

• GloSoFarID -> Global multispectral dataset for Solar Farm IDentification

• MARIDA -> Marine Debris detection from Sentinel-2

• MADOS -> Marine Debris and Oil Spill from Sentinel-2

• Sentinel-2 dataset for ship detection and characterization -> RGB

• S2-SHIPS -> all 12 channels

• ChatEarthNet -> A Global-Scale Image-Text Dataset Empowering Vision-Language Geo-Foundation Models, utilizes Sentinel-2 data with captions generated by ChatGPT

• UKFields -> over 2.3 million automatically delineated field boundaries spanning England, Wales, Scotland, and Northern Ireland

• ShipWakes -> Keypoints Method for Recognition of Ship Wake Components in Sentinel-2 Images by Deep Learning

• TimeSen2Crop -> a Million Labeled Samples Dataset of Sentinel 2 Image Time Series for Crop Type Classification

• AgriSen-COG -> a Multicountry, Multitemporal Large-Scale Sentinel-2 Benchmark Dataset for Crop Mapping: includes an anomaly detection preprocessing step

• MagicBathyNet -> 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

• MuS2: A Benchmark for Sentinel-2 Multi-Image Super-Resolution

• Sen4Map -> 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

• CloudSEN12Plus -> the largest cloud detection dataset to date for Sentinel-2

• mayrajeo S2 ship detection -> labels for Detecting marine vessels from Sentinel-2 imagery with YOLOv8

• Fields of The World -> instance segmentation of agricultural field boundaries

• ai4boundaries -> field boundaries with Sentinel-2 and aerial photography

• California Wildfire GeoImaging Dataset - CWGID -> Development and Application of a Sentinel-2 Satellite Imagery Dataset for Deep-Learning Driven Forest Wildfire Detection

• California Sentinel-2 Active-Fire Segmentation Dataset -> 2,148 Sentinel-2 false-colour image-mask chips (512x512 pixels at 20 m resolution) with per-pixel active-flame labels across 25 California wildfires from 2020–2026

• substation-seg -> segmenting substations dataset

• PhilEO-downstream -> a 400GB Sentinel-2 dataset for building density estimation, road segmentation, and land cover classification.

• PhilEO-pretrain -> a 500GB global dataset of Sentinel-2 images for model pre-training.

• KappaSet: Sentinel-2 KappaZeta Cloud and Cloud Shadow Masks

• AllClear A Comprehensive Dataset and Benchmark for Cloud Removal in Satellite Imagery

• Sentinel-2 reference cloud masks generated by an active learning method

• Cloud gap-filling with deep learning for improved grassland monitoring

• Remote Sensing Ship Wake Dataset

• ERAS-dataset -> Emilia-Romagna Agri Seg (ERAS) field segmentation dataset

• Sentinel 2 super-resolved data cubes - 92 scenes over 2 regions in Switzerland spanning 5 years

• SeasoNet -> 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

• EuroCropsML -> a ready-to-use benchmark dataset for few-shot crop type classification using Sentinel-2 imagery

• CanadaFireSat -> Sentinel-2 Level-1C time series

• ssl4eco -> a recipe for building pretraining sets that capture the geographical and phenological diversity of ecosystems across the globe

• IRRISIGHT -> a large-scale, multimodal remote sensing dataset for irrigation classification, soil-water mapping, and agricultural monitoring.

• SentinelKilnDB -> Sentinel-2 dataset for monitoring brick kiln emissions in South Asia

• MSSWD - Multi-Spectral Ship Wake Dataset

• MOSAIC-SEN2-CC -> A Multispectral Dataset and Adaptation Framework for Remote Sensing Change Captioning

• PLUTo -> post-deforestation land uses across the tropics

• SentinelKilnDB -> A Large-Scale Dataset and Benchmark for Oriented Bounding Box (OBB) Brick Kiln Detection in South Asia Using Satellite Imagery

• GSDD -> Global Supraglacial Debris Dataset

• MT4AFE -> Multi-Task Learning for Agricultural Field Extraction

• agripotential -> a Satellite Image Time Series (STIS) of 34 Sentinel-2 timeframes with 5 classes of agriculutural potential

• YieldSAT -> A Multimodal Benchmark Dataset for High-Resolution Crop Yield Prediction

• MOSAIC-SEN2-CC -> A Multispectral Change Captioning Dataset.

• Ghana Sentinel-2 Forest Dataset

• GeoSR-Bench -> directly connects super-resolution (SR) with downstream Earth monitoring tasks

• MANGO -> a global single-date paired dataset for mangrove segmentation.

• STTORM-CD-Floods -> a Sentinel-2 time-series change detection dataset capturing flood events, with 12 training and 4 test scenes provided as RGB composites and 14-channel NumPy arrays (all 13 L1C bands plus a cloud mask)

• S2GAIA -> a seasonally aware Sentinel-2 dataset for land-use and land-cover mapping in Greece, containing 34,030 annotated patches from 2017–2024 across 22 classes. Paper: S2GAIA: A seasonally aware multi-year dataset for deep learning-based land cover mapping using Sentinel-2 imagery in Greece

• SwissCrop25 -> a national-scale crop mapping benchmark for Switzerland spanning seven growing seasons (2019–2025), combining Sentinel-2 time series, daily temperature observations, 73 crop categories, and five non-crop land-cover classes. Paper: SwissCrop25: A National Multi-Year Benchmark for Operational Crop Mapping

• CASID -> 980 full-resolution satellite image and 5-class land-cover label pairs from 30 Asian regions across four climate zones, designed for domain-adaptive semantic segmentation. Paper: CASID: A Large-scale Climate-Aware Satellite Image Dataset for Domain Adaptive Land-Cover Semantic Segmentation

• CropGlobe -> more than 300,000 crop-type samples from eight countries for cross-region Sentinel-2 time-series classification. Paper: Invariant Features for Global Crop Type Classification

• DISFOR -> 3,823 densely labelled Sentinel-2 pixel and chip time series with forest-disturbance timing and causal-agent labels.

• Amfitrite Inland Waters HAB -> 4,698 multispectral Sentinel-2 tiles of inland waters with low, moderate, or high harmful-algal-bloom labels, water masks, and abundance metadata.

• S2LCD -> 1,533 Sentinel-2 image patches with five land-cover captions each, using natural language and the EAGLE land-monitoring lexicon.

• Middle East AOD Dataset -> 6,183 paired 13-band Sentinel-2 L1C scenes and AERONET aerosol-optical-depth observations from 20 Middle East and eastern Mediterranean sites, 2015–2025.

• SSL4EO-EU-Forest -> seasonal 12-band Sentinel-2 L2A patches over European forests with broadleaf and conifer forest masks derived from the Copernicus HRL 2018 product.

• OlmoEarth Land Cover Change -> point annotations for detecting and characterising land-cover change from 20-image Sentinel-2 time series, with source and destination classes and change dates.

• Crop Burn Detection — Labeled Sentinel-2 -> labelled Sentinel-2 imagery and metadata for crop-residue burning detection in India during 2025.

• EcoFireBias -> geographically balanced pre/post-fire Sentinel-2 RGB chips with dNBR severity labels across six continents and five ecosystem groups for evaluating VLM bias.

• Archaeological Site Dataset -> 11-channel 1 km Sentinel-2, FABDEM, index, and terrain samples with positive, negative, and unlabelled archaeological-site examples.

• EO Crop Type Belgium -> 5,990 nine-band Sentinel-2 patches covering most of Belgium with 10 m WorldCereal crop-type segmentation labels.

• scburning -> 1,054 Sentinel-2 image samples for sugar-cane burning segmentation on the north and central coast of Peru; GitHub.

• Flood Detection Pair Colombia -> 110 paired pre-event and current Sentinel-2 RGB/SWIR examples from nine Colombian floods, labelled for vision-language flood assessment.

• MegaEarth -> Sentinel-2 L2A surface-reflectance windows at native 10 m and 20 m resolution paired with OpenStreetMap area labels for evaluating transferable semantic representations.

• AIS-Associated Sentinel-2 Image Patch Dataset for Ship Type Classification -> Sentinel-2 B2, B3, B4 and B8 vessel-centred patches matched to AIS positions, with vessel-type labels and a South Florida external-region replication set.

• Sentinel-2 Multispectral Vessel Dataset with Bounding Box Labels -> 128x128 multispectral patches from the United States and Denmark with 1,093 AIS-derived vessel boxes, non-vessel examples, and cloud and land metadata.

• KappaSet_AI4QC -> Sentinel-2 cloud benchmark derived from KappaSet, with 512x512 masks for clear sky, thick cloud, thin cloud and cloud shadow and all 13 spectral bands at their native resolutions.

• Global ML-ready dataset for mining areas -> 1,210 globally distributed 2048x2048 Sentinel-2 tiles with manually verified mine masks, bounding boxes, mine types and train, validation and test splits.

• NortheastChinaSoybeanYield20m -> annual 20 m soybean-yield maps for Northeast China from 2019 to 2023, produced with a WOFOST-GRU framework using Sentinel-2-derived leaf-area index. Paper: NortheastChinaSoybeanYield20m: an annual soybean yield dataset at 20 m in Northeast China from 2019 to 2023

• Sentinel-2 vessels and side-by-side vessel pairs -> hand-reviewed 64x64 Sentinel-2 L2A crops with no_ship, ship and ship_pair classification labels, plus reconstruction and validation code; dataset download pending publication. Paper: A hand-reviewed Sentinel-2 dataset of vessels and side-by-side vessel pairs

Sentinel-3 (Optical)

• TMB-S3 -> GitHub. A burned-area segmentation dataset of Sentinel-3 OLCI time series for 246 European CEMS wildfire activations (2016–2025), split into 770 bounding boxes of about 25 × 25 km, with CEMS burned-area and analysed-area labels and ESA WorldCover land cover. Paper: Temporal Modelling for Burn Scars on Sentinel-3

Combined Sentinel

• awesome-sentinel -> a curated list of awesome tools, tutorials and APIs related to data from the Copernicus Sentinel Satellites.

• Paid access to Sentinel & Landsat data via sentinel-hub and python-api

• Jupyter Notebooks for working with Sentinel-5P Level 2 data stored on S3. The data can be browsed here

• Sentinel NetCDF data

• earthspy -> 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

• Gold Mining and clandestine airstrips datasets

• Industrial Smoke Plumes

• MARIDA: Marine Debris Archive

• OMS2CD -> hand-labelled images for change-detection in open-pit mining areas

• coal power plants' emissions -> a dataset of coal power plants' emissions, including images, metadata and labels.

• RapidAI4EO -> dense time series satellite imagery sampled at 500,000 locations across Europe, comprising S2 & Planet imagery, with CORINE Land Cover multiclass labels for 2018

• MS-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

• CropNet: An Open Large-Scale Dataset with Multiple Modalities for Climate Change-aware Crop Yield Predictions -> terabyte-sized, publicly available, and multi-modal dataset for climate change-aware crop yield predictions

• Tiny CropNet dataset

• Multitask Learning for Estimating Power Plant Greenhouse Gas Emissions from Satellite Imagery

• METER-ML: A Multi-sensor Earth Observation Benchmark for Automated Methane Source Mapping -> data on Zenodo

• MultiSenGE -> large-scale multimodal and multitemporal benchmark dataset

• SEN12MS -> A Curated Dataset of Georeferenced Multi-spectral Sentinel-1/2 Imagery for Deep Learning and Data Fusion. Checkout SEN12MS toolbox and many referenced uses on paperswithcode.com

• Space2Ground -> dataset with Space (Sentinel-1/2) and Ground (street-level images) components, annotated with crop-type labels for agriculture monitoring.

• Space2Ground 2.0 -> GitHub. A parcel-level agricultural monitoring dataset for Cyprus in 2022 combining Sentinel-1/2 time series, official crop labels, and 46,050 street-level images associated with 8,581 parcels. Paper: Space2Ground 2.0: A Multi-Source Dataset and Framework for Agricultural Monitoring through Fusion of Street-Level and Satellite Imagery

• MSCDUnet -> change detection datasets containing VHR, multispectral (Sentinel-2) and SAR (Sentinel-1)

• OMBRIA -> Sentinel-1 & 2 dataset for adressing the flood mapping problem

• GEOID-Flood -> a large-scale multimodal flood segmentation benchmark with co-registered Sentinel-1, Sentinel-2, and elevation data covering 219 flood events across 65 countries. Paper: GEOID-Flood: A Large-Scale Multi-Modal Benchmark Dataset for Flood Segmentation

• Satellite Burned Area Dataset -> segmentation dataset containing several satellite acquisitions related to past forest wildfires. It contains 73 acquisitions from Sentinel-2 and Sentinel-1 (Copernicus).

• SEN12_GUM -> SEN12 Global Urban Mapping Dataset

• Sentinel-1&2 Image Pairs (SAR & Optical)

• MSOSCD -> change detection datasets containing VHR, multispectral (Sentinel-2) and SAR (Sentinel-1)

• SICKLE -> 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

• Sentinel-1 and Sentinel-2 Vessel Detection

• TreeSatAI -> Sentinel-1, Sentinel-2

• AI2-S2-NAIP -> aligned NAIP, Sentinel-2, Sentinel-1, and Landsat images spanning the entire continental US

• POPCORN: High-resolution Population Maps Derived from Sentinel-1 and Sentinel-2

• CropClimateX -> A large-scale Multitask, Multisensory Dataset for Crop Monitoring under Climate Extremes

• CropMapIL Orchard Dataset -> labels, dataset splits, and multi-temporal Sentinel-1/2 features for mapping 12 orchard types in Israel

• SmallMinesDS -> A Multimodal Dataset for Mapping Artisanal and Small-Scale Gold Mines. Imagery reused in CocoaMiningDS

• Hoss-ReID -> Cross-modal Ship Re-Identification via Optical and SAR Imagery

• IDEABench Benchmark Dataset -> Mapping and Benchmarking Urban Deprivation for a Global Sample of Cities

• ImpactMesh -> a large-scale multimodal, multitemporal dataset for flood and wildfire mapping

• Sen12Landslides -> Spatio-Temporal Landslide & Anomaly Detection Dataset

• Cryo-Bench -> A Benchmark for Evaluating Geospatial Foundation Models on Cryosphere Applications

• BigEarthNet.txt -> A Large-Scale Multi-Sensor Image-Text Dataset and Benchmark for Earth Observation

• Borneo_Forest_Disturbance_Dataset -> Forest Disturbance Dataset utilising Sentinel-2 data and RADD alert disturbances.

• GeoMeld -> a large-scale multimodal remote sensing dataset for semantically grounded foundation modeling. It contains approximately 2.5 million spatially aligned samples spanning optical imagery, SAR, elevation, canopy height, land-cover products, and geographic metadata, paired with semantically grounded captions generated through an agentic captioning framework.

• Forest Loss Dataset Northeast India

• Sentinel2Cap -> contains captions for remote sensing images, both manually annotated and automatically generated with Qwen3-VL-8B-Instruct.

• 50Cities -> a global-scale benchmark for multi-modal land-cover segmentation across 50 urban regions on six continents, with 33 labeled and 17 unlabeled cities, using co-registered Sentinel-1 and Sentinel-2 image pairs

• Infra-Bench CLS -> 600 m Sentinel-1 RTC and seven-band Sentinel-2 L2A tiles centred on critical-infrastructure assets across seven continental regions and four infrastructure sectors.

• TreeUQ -> 45,419 labelled 10 m Bavaria patches with seasonal Sentinel-1/2 composites, tree species, height, count and density rasters, plus optional 20 cm orthophotos.

• WaterBench -> a 175 GB multimodal coastal and marine benchmark spanning six regression, classification, and segmentation tasks with Sentinel-1/2 and contextual Sentinel-3 imagery.

• GeoFlood-275 -> 275 flood event–AOI pairs and 133,555 patches combining pre-event Sentinel-2, post-event Sentinel-1, terrain, land cover, and flood masks; GitHub.

• BFC-Bench -> 9,888 global crop-damage chips from 2,760 flood and wildfire events, with pre/post Sentinel-1/2 imagery, damage masks, phenology, and quality-control metadata.

• Sentinel Patch Imagery Dataset for China's Cropland -> four-season Sentinel-1 and Sentinel-2 time-series patches over cropland-dominant areas across China in 2021.

• CryoSentinel GLOF v3 -> 12-band Sentinel-2, dual-polarisation Sentinel-1, Copernicus DEM, and glacial-lake masks for Central Asian mountain ranges; GitHub.

• GeoPulse-Bench -> event-split 10 m Sentinel-1/2 and DEM benchmarks for floods, burned areas, and stand-replacing forest loss.

• Mangrove Biomass and Carbon Stocks in Côte d’Ivoire -> 300 field plots with above- and below-ground biomass, soil carbon, and coincident Sentinel-1/2 measurements from Sassandra and Fresco.

• CEMS-HLS -> 439 eight-band 512×512 Harmonized Landsat Sentinel-2 scenes with CEMS burn-scar masks from roughly 200 wildfire activations during 2020–2024.

• HLS Hail and Wind Damage Events -> event-split HLS GeoTIFF chips and full scenes paired with hail and wind damage masks.

• HWDS-Multimodal -> 7,665 paired 30 m HLS and Sentinel-1 RTC chips with rasterised hail and wind damage masks.

• Hong Kong HLS Phenology -> 2015–2025 six-band HLS vegetation time series over Hong Kong with 3-day composites, training pixels, a full vegetation cube, and phenology metadata.

• PLANTEUR -> crop-type semantic and panoptic segmentation benchmark for French overseas territories, combining Sentinel-1, Sentinel-2 and Landsat-8 time series in a PASTIS-compatible format.

• UEIKAP -> multimodal maritime ship-wake dataset with labelled Sentinel-2 optical chips and Sentinel-1 SAR chips, AIS records, annotations and GeoTIFF imagery.

• Vessel Detection Labeled Patches -> validated Sentinel-1 and Sentinel-2 vessel and hard-negative patches with object annotations and YOLO labels.

• MAMMS -> multi-altitude maritime-surveillance dataset spanning ground RGB, SWIR, UV and thermal sensors, UAV imagery and Sentinel-2 RGB observations; download requires a user agreement.

• PGMpk-dl dataset -> regional Sentinel-2 imagery, pseudo-labels and manually annotated ground truth for plastic-greenhouse mapping, with RGB composites and GeoTIFF labels. Paper: A novel approach: Coupling prior knowledge and deep learning methods for large-scale plastic greenhouse extraction using Sentinel-1/2 data

• SSL4EO-S12 v1.1 -> updated multimodal Sentinel-1/2 self-supervised pretraining dataset with corrected temporal alignment and Zarr and WebDataset distributions.

• HLS Multi-Temporal Crop Classification -> generation pipeline and open dataset of 3,854 three-date HLS chips across CONUS with USDA Cropland Data Layer labels for 13 crop and land-cover classes.

• Sentinel-2 and Sentinel-3 Spatiotemporal Fusion Benchmark -> more than 689 paired Sentinel-2 MSI and Sentinel-3 OLCI observations across 15 diverse sites for evaluating spatiotemporal fusion. Paper: A new benchmark for spatiotemporal fusion of Sentinel-2 and Sentinel-3 OLCI images

• Multi agro-climatic irrigation reference dataset -> field-level irrigated and non-irrigated labels from France and Kazakhstan with crop types, irrigation practices, Sentinel-1 backscatter and Sentinel-2 vegetation-index time series.

• Inner Mongolia sugar-beet maps -> annual 10 m sugar-beet extent maps for 2020–2024, generated from Sentinel-2 phenology and MODIS land-surface temperature.

Landsat

Long running US program -> see Wikipedia

• 8 bands, 15 to 60 meters, 185km swath, the temporal resolution is 16 days

• Landsat 4, 5, 7, and 8 imagery on Google, see the GCP bucket here, with Landsat 8 imagery in COG format analysed in this notebook

• Landsat 8 imagery on AWS, with many tutorials and tools listed

• https://github.com/kylebarron/landsat-mosaic-latest -> Auto-updating cloudless Landsat 8 mosaic from AWS SNS notifications

• Visualise landsat imagery using Datashader

• Landsat-mosaic-tiler -> This repo hosts all the code for landsatlive.live website and APIs.

• LandsatSCD -> a change detection dataset, it consists of 8468 pairs of images, each having the spatial resolution of 416 × 416

• The Landsat Irish Coastal Segmentation Dataset

• Wildfire-Spread-Dataset -> ABNextFire: A Multi-Source Deep Learning Based Dataset for Wildfire Spread Prediction

• LandsatQuake -> 1,042 RGB Landsat patches with object-detection and instance-segmentation annotations drawn from 31 landslide inventories in 21 earthquake-prone regions. Paper: LandsatQuake: A Large-Scale Dataset For Practical Landslide Detection

• Landsat30-AU -> 196,262 image-caption pairs and 17,725 human-verified VQA samples from Landsat 5, 7, 8, and 9 imagery across Australia over more than 36 years. Paper: Landsat30-AU: A Vision-Language Dataset for Australian Landsat Imagery

VENμS

Vegetation and Environment monitoring on a New Micro-Satellite (VENμS)

• VENUS L2A Cloud-Optimized GeoTIFFs
• VENuS cloud mask training dataset
• Sen2Venµs -> a dataset for the training of Sentinel-2 super-resolution algorithms
• sen2venus-pytorch-dataset -> torch dataloader and other utilities
• VENµS raw images for vessel detection -> 284 twelve-band VENµS scenes over Ashdod acquired from 2018–2020, with vessel bounding boxes and AIS-derived identification metadata.
• Raw/L0 Multispectral Ship Classification Datasets -> 64×64 raw multispectral vessel crops from VENµS and Sentinel-2 for ship-type classification.

Vantor

Satellites owned by Vantor (formerly Maxar & DigitalGlobe) include GeoEye-1, WorldView-2, 3 & 4

• Maxar Open Data Program provides pre and post-event high-resolution satellite imagery in support of emergency planning, response, damage assessment, and recovery
• WorldView-2 European Cities -> dataset covering the most populated areas in Europe at 40 cm resolution

Planet

Also see Spacenet-7 and the Kaggle ship and plane classifications datasets later in this page

• Planet’s high-resolution, analysis-ready mosaics of the world’s tropics, supported through Norway’s International Climate & Forests Initiative. BBC coverage
• Fields of the Planet -> a 3 m PlanetScope companion to Fields of The World for field-boundary mapping, pairing 66,584 patches from 24 countries with two seasonal image windows, raster labels, and field polygons
• PKLandSeg -> 3,330 samples for landslide segmentation in Pakistan's Gilgit-Baltistan region, with 3 m PlanetScope RGB imagery, elevation, NDVI, slope, and binary masks at 512×512 pixels; includes fixed training, validation, and test splits
• Planet have made imagery available via kaggle competitions
• Alberta Wells Dataset -> Pinpointing Oil and Gas Wells from Satellite Imagery
• ARGO ship classification dataset -> 1750 labelled images from PlanetScope-4-Band satelites. Created here
• Marine Debris Dataset for Object Detection in Planetscope Imagery
• LitterLines -> An Annotated Dataset for Detection of Marine Litter Windrows in PlanetScope Imagery
• FloodPlanet Inundation Dataset -> multi-sensor co-registered dataset labeled based on 3m PlanetScope data and spatially overlapping, temporally near Sentinel-1, Sentinel-2, and Landsat-8 data
• Zhijie_FloodPlanet_2023 -> 19 flood events that occurred between 2017 and 2020
• LacunaLabels -> A region-wide, multi-year set of crop field boundary labels for Africa

UC Merced

Land 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

• http://weegee.vision.ucmerced.edu/datasets/landuse.html
• Also available as a multi-label dataset
• Read Vision Transformers for Remote Sensing Image Classification where a Vision Transformer classifier achieves 98.49% classification accuracy on Merced

EuroSAT

Land 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

• EuroSAT: Land Use and Land Cover Classification with Sentinel-2 -> publication where a CNN achieves a classification accuracy 98.57%
• Repos using fastai here and here
• evolved_channel_selection -> explores the trade off between mixed resolutions and whether to use a channel at all, with repo
• RGB version available as dataset in pytorch with the 13 band version in torchgeo. Checkout the tutorial on data augmentation with this dataset
• EuroSAT-SAR -> matched each Sentinel-2 image in EuroSAT with one Sentinel-1 patch according to the geospatial coordinates
• Similar but Different -> a 30,927-patch Sentinel-2 land-cover benchmark designed to test whether models use multispectral information when RGB alone cannot easily separate classes; includes 12-band GeoTIFFs and fixed train, validation, and test splits

PatternNet

Land use classification dataset with 38 classes and 800 RGB JPG images for each class

• https://sites.google.com/view/zhouwx/dataset?authuser=0
• Publication: PatternNet: A Benchmark Dataset for Performance Evaluation of Remote Sensing Image Retrieval

Gaofen Image Dataset (GID) for classification

• https://captain-whu.github.io/GID/
• a large-scale classification set and a fine land-cover classification set

Million-AID

A large-scale benchmark dataset containing million instances for RS scene classification, 51 scene categories organized by the hierarchical category

• https://captain-whu.github.io/DiRS/
• Pretrained models
• Also see AID, AID-Multilabel-Dataset & DFC15-multilabel-dataset

DIOR object detection dataset

A 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

• https://gcheng-nwpu.github.io/
• https://arxiv.org/abs/1909.00133
• ors-detection -> Object Detection on the DIOR dataset using YOLOv3
• dior_detect -> benchmarks for object detection on DIOR dataset
• Tools -> for dealing with the DIOR
• Object_Detection_Satellite_Imagery_Yolov8_DIOR
• DIOR on Hugging Face -> repackaged DIOR optical remote-sensing images with 20 object classes and COCO-format bounding boxes for the Hugging Face datasets library. Paper: Object detection in optical remote sensing images: A survey and a new benchmark

LEVIRDet object detection dataset

LEVIRDet is a million-scale, 159-category dataset and foundation model for universal remote sensing object detection.

Multiscene

MultiScene dataset aims at two tasks: Developing algorithms for multi-scene recognition & Network learning with noisy labels

• https://multiscene.github.io/ & https://github.com/Hua-YS/Multi-Scene-Recognition

FAIR1M object detection dataset

A Benchmark Dataset for Fine-grained Object Recognition in High-Resolution Remote Sensing Imagery

• arxiv papr
• Download at gaofen-challenge.com
• 2020Gaofen -> 2020 Gaofen Challenge data, baselines, and metrics

DOTA object detection dataset

A Large-Scale Benchmark and Challenges for Object Detection in Aerial Images. Segmentation annotations available in iSAID dataset

• https://captain-whu.github.io/DOTA/index.html
• DOTA_devkit for loading dataset
• Arxiv paper
• Pretrained models in mmrotate
• DOTA2VOCtools -> dataset split and transform to voc format
• dotatron -> 2021 Learning to Understand Aerial Images Challenge on DOTA dataset
• DOTA on Hugging Face -> DOTA aerial imagery split into 512x512 patches with 16 object classes and COCO-format bounding boxes for the Hugging Face datasets library. Paper: DOTA: A Large-Scale Dataset for Object Detection in Aerial Images

iSAID instance segmentation dataset

A Large-scale Dataset for Instance Segmentation in Aerial Images

• https://captain-whu.github.io/iSAID/dataset.html
• Uses images from the DOTA dataset

HRSC RGB ship object detection dataset

• https://www.kaggle.com/datasets/guofeng/hrsc2016
• Pretrained models in mmrotate
• Rotation-RetinaNet-PyTorch

SAR Ship Detection Dataset (SSDD)

• https://github.com/TianwenZhang0825/Official-SSDD
• Rotation-RetinaNet-PyTorch

High-Resolution SAR Rotation Ship Detection Dataset (SRSDD)

• Github
• A Lightweight Model for Ship Detection and Recognition in Complex-Scene SAR Images

LEVIR ship dataset

A dataset for tiny ship detection under medium-resolution remote sensing images. Annotations in bounding box format

• LEVIR-Ship

• Hosted on Nucleus

SAR Aircraft Detection Dataset

2966 non-overlapped 224×224 slices are collected with 7835 aircraft targets

• https://github.com/hust-rslab/SAR-aircraft-data

xView1: Objects in context for overhead imagery

A 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

• Official website
• arXiv paper.
• paperswithcode
• Satellite_Imagery_Detection_YOLOV7 -> YOLOV7 applied to xView1

xView2: xBD building damage assessment

Annotated high-resolution satellite imagery for building damage assessment, precise segmentation masks and damage labels on a four-level spectrum, 0.3m resolution imagery

• Official website
• arXiv paper
• paperswithcode
• xView2_baseline -> baseline solution in tensorflow
• metadamagenet -> pytorch solution
• U-Net models from michal2409
• DAHiTra -> code for 2022 paper: Large-scale Building Damage Assessment using a Novel Hierarchical Transformer Architecture on Satellite Images. Uses xView2 xBD dataset
• Damage assessment using Amazon SageMaker geospatial capabilities and custom SageMaker models
• Xview2_Strong_Baseline -> a simple implementation of a strong baseline

xView3: Detecting dark vessels in SAR

Detecting 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

• Official website
• arXiv paper
• Github -> all reference code, dataset processing utilities, and winning model codes + weights
• paperswithcode
• xview3_ship_detection

Vehicle Detection in Aerial Imagery (VEDAI)

Vehicle Detection in Aerial Imagery. Bounding box annotations

• https://downloads.greyc.fr/vedai/
• pytorch-vedai

Cars Overhead With Context (COWC)

Large set of annotated cars from overhead. Established baseline for object detection and counting tasks. Annotations in bounding box format

• http://gdo152.ucllnl.org/cowc/
• https://github.com/LLNL/cowc
• Detecting cars from aerial imagery for the NATO Innovation Challenge
• LINZ and UGRC

AI-TOD & AI-TOD-v2 - tiny object detection

The 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

• https://github.com/jwwangchn/AI-TOD
• https://chasel-tsui.github.io/AI-TOD-v2/
• NWD -> code for 2021 paper: A Normalized Gaussian Wasserstein Distance for Tiny Object Detection. Uses AI-TOD dataset
• ORFENet -> Tiny Object Detection in Remote Sensing Images Based on Object Reconstruction and Multiple Receptive Field Adaptive Feature Enhancement. Uses LEVIR-ship & AI-TOD-v2

RarePlanes

• RarePlanes -> incorporates both real and synthetically generated satellite imagery including aircraft. Read the arxiv paper and checkout this repo. Note the dataset is available through the AWS Open-Data Program for free download
• Understanding the RarePlanes Dataset and Building an Aircraft Detection Model -> blog post
• Read this article from NVIDIA 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
• yoltv4 includes examples on the RarePlanes dataset
• rareplanes-yolov5 -> using YOLOv5 and the RarePlanes dataset to detect and classify sub-characteristics of aircraft, with article

Counting from Sky

A Large-scale Dataset for Remote Sensing Object Counting and A Benchmark Method

• https://github.com/gaoguangshuai/Counting-from-Sky-A-Large-scale-Dataset-for-Remote-Sensing-Object-Counting-and-A-Benchmark-Method

AIRS (Aerial Imagery for Roof Segmentation)

Public 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.

• On Kaggle
• Rooftop-Instance-Segmentation -> VGG-16, Instance Segmentation, uses the Airs dataset

Inria building/not building segmentation dataset

RGB GeoTIFF at spatial resolution of 0.3 m. Data covering Austin, Chicago, Kitsap County, Western & Easter Tyrol, Innsbruck, San Francisco & Vienna

• https://project.inria.fr/aerialimagelabeling/contest/
• SemSegBuildings -> Project using fast.ai framework for semantic segmentation on Inria building segmentation dataset
• UNet_keras_for_RSimage -> keras code for binary semantic segmentation

AICrowd Mapping Challenge: building segmentation dataset

300x300 pixel RGB images with annotations in COCO format. Imagery appears to be global but with significant fraction from North America

• Dataset release as part of the mapping-challenge
• Winning solution published by neptune.ai here, achieved precision 0.943 and recall 0.954 using Unet with Resnet.
• mappingchallenge -> YOLOv5 applied to the AICrowd Mapping Challenge dataset

BONAI - building footprint dataset

BONAI (Buildings in Off-Nadir Aerial Images) is a dataset for building footprint extraction (BFE) in off-nadir aerial images

• https://github.com/jwwangchn/BONAI

LEVIR-CD building change detection dataset

• https://justchenhao.github.io/LEVIR/
• FCCDN_pytorch -> pytorch implemention of FCCDN for change detection task
• RSICC -> the Remote Sensing Image Change Captioning dataset uses LEVIR-CD imagery

Onera (OSCD) Sentinel-2 change detection dataset

It comprises 24 pairs of multispectral images taken from the Sentinel-2 satellites between 2015 and 2018.

• Onera Satellite Change Detection Dataset comprises 24 pairs of multispectral images taken from the Sentinel-2 satellites between 2015 and 2018
• Website
• change_detection_onera_baselines -> Siamese version of U-Net baseline model
• Urban Change Detection for Multispectral Earth Observation Using Convolutional Neural Networks
• DS_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
• ChangeDetection_wOnera
• OSCD + additional Dates -> extended with three different dates
• MSOSCD -> change detection datasets containing VHR, multispectral (Sentinel-2) and SAR (Sentinel-1)

SECOND - semantic change detection

• https://captain-whu.github.io/SCD/
• Change detection at the pixel level

Amazon and Atlantic Forest dataset

For semantic segmentation with Sentinel 2

• Amazon and Atlantic Forest image datasets for semantic segmentation
• attention-mechanism-unet -> An attention-based U-Net for detecting deforestation within satellite sensor imagery
• TransUNetplus2 -> Rethinking attention gated TransU-Net for deforestation mapping

Functional Map of the World ( fMoW)

• https://github.com/fMoW/dataset
• RGB & multispectral variants
• High resolution, chip classification dataset
• Purpose: predicting the functional purpose of buildings and land use from temporal sequences of satellite images and a rich set of metadata features

HRSCD change detection

• https://rcdaudt.github.io/hrscd/
• 291 coregistered image pairs of high resolution RGB aerial images
• Pixel-level change and land cover annotations are provided

MiniFrance-DFC22 - semi-supervised semantic segmentation

• The MiniFrance-DFC22 (MF-DFC22) dataset extends and modifies the MiniFrance dataset for training semi-supervised semantic segmentation models for land use/land cover mapping
• dfc2022-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
• https://github.com/mveo/mveo-challenge

FLAIR

Semantic 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.

• Challenge on codalab
• FLAIR-2 github
• flair-2 8th place solution
• IGNF HuggingFace
• FLAIR-HUB -> UperFuse code for the FLAIR-HUB dataset

ISPRS

Semantic 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

• https://www.isprs.org/resources/datasets/benchmarks/UrbanSemLab/2d-sem-label-potsdam.aspx

SpaceNet

SpaceNet 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

• spacenet.ai is an online hub for data, challenges, algorithms, and tools
• The SpaceNet 7 Multi-Temporal Urban Development Challenge: Dataset Release
• spacenet-three-topcoder solution
• official utilities -> Packages intended to assist in the preprocessing of SpaceNet satellite imagery dataset to a format that is consumable by machine learning algorithms
• andraugust spacenet-utils -> Display geotiff image with building-polygon overlay & label buildings using kNN on the pixel spectra
• Spacenet-Building-Detection -> uses keras and Spacenet 1 dataset
• Spacenet 8 winners blog post

WorldStrat Dataset

Nearly 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.

• https://github.com/worldstrat/worldstrat
• Quick tour of the WorldStrat Dataset
• Each high-resolution image (1.5 m/pixel) comes with multiple temporally-matched low-resolution images from the freely accessible lower-resolution Sentinel-2 satellites (10 m/pixel)
• Several super-resolution benchmark models trained on it

Satlas Pretrain

SatlasPretrain 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.

• Website
• Code

FLAIR 1 & 2 Segmentation datasets

• https://ignf.github.io/FLAIR/
• The FLAIR #1 semantic segmentation dataset consists of 77,412 high resolution patches (512x512 at 0.2 m spatial resolution) with 19 semantic classes
• FLAIR #2 includes an expanded dataset of Sentinel-2 time series for multi-modal semantic segmentation

Five Billion Pixels segmentation dataset

• https://x-ytong.github.io/project/Five-Billion-Pixels.html
• 4m Gaofen-2 imagery over China
• 24 land cover classes
• Paper and code demonstrating domain adaptation to Sentinel-2 and Planetscope imagery
• Extends the GID15 large scale semantic segmentation dataset
• GID -> the Gaofen Image Dataset is a large-scale land-cover dataset with Gaofen-2 (GF-2) satellite images
• MM-5B Dataset -> 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.

RF100 object detection benchmark

RF100 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

• https://www.rf100.org/
• https://github.com/roboflow-ai/roboflow-100-benchmark

SATIN (SATellite ImageNet)

SATIN 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.

• Paper
• Website
• Data

SODA-A rotated bounding boxes

• https://shaunyuan22.github.io/SODA/
• SODA-A comprises 2513 high-resolution images of aerial scenes, which has 872069 instances annotated with oriented rectangle box annotations over 9 classes
• https://github.com/shaunyuan22/CFINet

EarthView from Satellogic

• https://huggingface.co/datasets/satellogic/EarthView
• Dataset for foundational models, with Sentinel 1 & 2 and 1m RGB

Microsoft datasets

• US Building Footprints -> building footprints in all 50 US states, GeoJSON format, generated using semantic segmentation. Also Australia, Canadian, Uganda-Tanzania, Kenya-Nigeria and GlobalMLBuildingFootprints are available. Checkout RasterizingBuildingFootprints to convert vector shapefiles to raster layers
• Microsoft Planetary Computer is a Dask-Gateway enabled JupyterHub deployment focused on supporting scalable geospatial analysis, source repo
• landcover-orinoquia -> Land cover mapping of the Orinoquía region in Colombia, in collaboration with Wildlife Conservation Society Colombia. An #AIforEarth project
• RoadDetections dataset by Microsoft

Google datasets

• open-buildings -> A dataset of building footprints to support social good applications covering 64% of the African continent. Read Mapping Africa’s Buildings with Satellite Imagery

Google Earth Engine (GEE)

Since there is a whole community around GEE I will not reproduce it here but list very select references. Get started at https://developers.google.com/earth-engine/

• Various imagery and climate datasets, including Landsat & Sentinel imagery
• 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
• awesome-google-earth-engine
• Awesome-GEE
• awesome-earth-engine-apps
• How to Use Google Earth Engine and Python API to Export Images to Roboflow -> to acquire training data
• ee-fastapi is a simple FastAPI web application for performing flood detection using Google Earth Engine in the backend.
• How to Download High-Resolution Satellite Data for Anywhere on Earth
• wxee -> Export data from GEE to xarray using wxee then train with pytorch or tensorflow models. Useful since GEE only suports tfrecord export natively

Image captioning datasets

• RSICD -> 10921 images with five sentences descriptions per image. Used in Fine tuning CLIP with Remote Sensing (Satellite) images and captions, models at this repo
• RSICC -> 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
• RSRCC -> Remote Sensing Remote Change Captioning dataset of paired before/after satellite images with natural language descriptions of semantic changes such as new construction, demolition, road changes, vegetation changes, and residential development
• ChatEarthNet -> A Global-Scale Image-Text Dataset Empowering Vision-Language Geo-Foundation Models, utilizes Sentinel-2 data with captions generated by ChatGPT

Weather Datasets

• NASA (make request and emailed when ready) -> https://search.earthdata.nasa.gov
• NOAA (requires BigQuery) -> https://www.kaggle.com/datasets/noaa/goes16/home
• Time series weather data for several US cities -> https://www.kaggle.com/datasets/selfishgene/historical-hourly-weather-data
• DeepWeather -> improve weather forecasting accuracy by analyzing satellite images

Cloud datasets

• Planet-CR -> A Multi-Modal and Multi-Resolution Dataset for Cloud Removal in High Resolution Optical Remote Sensing Imagery, 3m resolution, with paper
• The Azavea Cloud Dataset which is used to train this cloud-model
• Sentinel-2 Cloud Cover Segmentation Dataset on Radiant mlhub
• cloudsen12 -> see video
• WHUS2-CD+ -> 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
• HRC_WHU -> High-Resolution Cloud Detection Dataset comprising 150 RGB images and a resolution varying from 0.5 to 15 m in different global regions
• AIR-CD -> a challenging cloud detection data set called AIR-CD, with higher spatial resolution and more representative landcover types
• Landsat 8 Cloud Cover Assessment Validation Data

Forest datasets

• OpenForest -> A catalogue of open access forest datasets
• awesome-forests -> A curated list of ground-truth forest datasets for the machine learning and forestry community
• ReforesTree -> A dataset for estimating tropical forest biomass based on drone and field data
• yosemite-tree-dataset -> a benchmark dataset for tree counting from aerial images
• Amazon Rainforest dataset for semantic segmentation -> Sentinel 2 images. Used in the paper 'An attention-based U-Net for detecting deforestation within satellite sensor imagery'
• Amazon and Atlantic Forest image datasets for semantic segmentation -> Sentinel 2 images. Used in paper 'An attention-based U-Net for detecting deforestation within satellite sensor imagery'
• TreeSatAI -> Sentinel-1, Sentinel-2
• PureForest -> VHR RGB + Near-Infrared & lidar, each patch represents a monospecific forest

Geospatial datasets

• Resource Watch provides a wide range of geospatial datasets and a UI to visualise them

Time series & change detection datasets

• BreizhCrops -> A Time Series Dataset for Crop Type Mapping
• The SeCo dataset contains image patches from Sentinel-2 tiles captured at different timestamps at each geographical location. Download SeCo here
• SYSU-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

DEM (digital elevation maps)

• Shuttle Radar Topography Mission, search online at usgs.gov
• 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
• Awesome-DEM

UAV & Drone datasets

• Many on https://www.visualdata.io
• AU-AIR dataset -> a multi-modal UAV dataset for object detection.
• ERA -> A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos.
• Aerial Maritime Drone Dataset -> bounding boxes
• RetinaNet for pedestrian detection -> bounding boxes
• BIRDSAI: A Dataset for Detection and Tracking in Aerial Thermal Infrared Videos -> Thermal IR videos of humans and animals
• ERA: A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos
• DroneVehicle -> Drone-based RGB-Infrared Cross-Modality Vehicle Detection via Uncertainty-Aware Learning. Annotations are rotated bounding boxes. With Github repo
• 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
• Busy-parking-lot-dataset---vehicle-detection-in-UAV-video -> Vehicle instance segmentation. Unsure format of annotations, possible Matlab specific
• dd-ml-segmentation-benchmark -> DroneDeploy Machine Learning Segmentation Benchmark
• SeaDronesSee -> Vision Benchmark for Maritime Search and Rescue. Bounding box object detection, single-object tracking and multi-object tracking annotations
• aeroscapes -> semantic segmentation benchmark comprises of images captured using a commercial drone from an altitude range of 5 to 50 metres.
• ALTO -> Aerial-view Large-scale Terrain-Oriented. For deep learning based UAV visual place recognition and localization tasks.
• HIT-UAV-Infrared-Thermal-Dataset -> A High-altitude Infrared Thermal Object Detection Dataset for Unmanned Aerial Vehicles
• caltech-aerial-rgbt-dataset -> synchronized RGB, thermal, GPS, and IMU data
• Leafy Spurge Dataset -> Real-world Weed Classification Within Aerial Drone Imagery
• Agriculture-Vision 2021 Dataset
• UAV-HSI-Crop-Dataset -> dataset for "HSI-TransUNet: A Transformer based semantic segmentation model for crop mapping from UAV hyperspectral imagery"
• UAVVaste -> COCO-like dataset and effective waste detection in aerial images
• BSB-Aerial-Dataset -> A panoptic segmentation dataset of aerial imagery from Brasilia, Brazil.
• ELDOR -> large-scale UAV benchmark for monitoring illegal gold mining in the Amazon rainforest, with manually annotated orthomosaic imagery covering 2,500+ hectares and pixel-level semantic labels for mining activities and ecological structures, supporting segmentation and classification tasks
• Songdo Vision -> GitHub. 5,419 high-altitude bird's-eye-view drone frames (4K, 140-150 m altitude) with 274,190 vehicle bounding boxes in 4 classes (car/van, bus, truck, motorcycle) over an urban road network in Songdo, South Korea, in COCO, YOLO and Pascal VOC formats, CC BY 4.0. Paper: Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery
• AVID -> GitHub. 32,040 weakly aligned RGB-infrared frame pairs from ground-based surveillance of birds and drones, with 35,952 bounding-box annotations and the AVID-C perturbation benchmark. Paper: AVID: A Weakly Aligned RGB-Infrared Dataset and Perturbation Benchmark for Tiny Aerial Targets

Other datasets

Object Detection & Classification

• RSOD-Dataset -> dataset for object detection in PASCAL VOC format. Aircraft, playgrounds, overpasses & oiltanks
• VHR-10_dataset_coco -> Object detection and instance segmentation dataset based on NWPU VHR-10 dataset. RGB & SAR
• MAR20 -> Military Aircraft Recognition dataset
• RSAPS-ASD -> A Remote Sensing Airport Panoptic Segmentation with Airplane States dataset, constructed in "Airplane State Discrimination from Single-temporal High-Resolution Remote Sensing Images"
• Sewage-Treatment-Plant-Dataset -> object detection
• TGRS-HRRSD-Dataset -> High Resolution Remote Sensing Detection (HRRSD)
• OGST -> Oil and Gas Tank Dataset
• SearchAndRescueNet -> Satellite Imagery for Search And Rescue Dataset, with example Faster R-CNN model
• UBC-dataset -> a dataset for building detection and classification from very high-resolution satellite imagery with the focus on object-level interpretation of individual buildings
• Building_Dataset -> High-speed Rail Line Building Dataset Display
• RID -> Roof Information Dataset for CV-Based Photovoltaic Potential Assessment. With paper
• APKLOT -> A dataset for aerial parking block segmentation
• SAR-ACD -> SAR-ACD consists of 4322 aircraft clips with 6 civil aircraft categories and 14 other aircraft categories
• SODA -> 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.
• urban-tree-detection-data -> Dataset for training and evaluating tree detectors in urban environments with aerial imagery
• Satellite imagery datasets containing ships -> A list of radar and optical satellite datasets for ship detection, classification, semantic segmentation and instance segmentation tasks
• Roofline-Extraction -> dataset for paper 'Knowledge-Based 3D Building Reconstruction (3DBR) Using Single Aerial Images and Convolutional Neural Networks (CNNs)'
• Building-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'
• OnlyPlanes -> Synthetic dataset and pretrained models for Detectron2
• SV248S -> Single Object Tracking Dataset, tracking Vehicle, Large-Vehicle, Ship and Airplane
• NWPU-MOC -> A Benchmark for Fine-grained Multi-category Object Counting in Aerial Images
• Vehicle Perception from Satellite -> a large-scale benchmark for traffic monitoring from satellite
• SARDet-100K -> Large-Scale Synthetic Aperture Radar (SAR) Object Detection
• Urban Vehicle Segmentation Dataset (UV6K)
• ShipRSImageNet -> A Large-scale Fine-Grained Dataset for Ship Detection in High-Resolution Optical Remote Sensing Images
• VME: A Satellite Imagery Dataset and Benchmark for Detecting Vehicles in the Middle East and Beyond
• VHRV: Very High-Resolution Benchmark Dataset for Vessel Detection
• VesselBench-800K -> 800,000 512x512 optical and SAR images, including synthetic sources, for multimodal vessel detection, counting and density estimation. Paper: VesselBench-800K: A Large-scale Perception Benchmark for Multimodal Vessel Detection, Counting, and Density Estimation

Land Use & Land Cover

• land-use-land-cover-datasets
• RSD46-WHU -> 46 scene classes for image classification, free for education, research and commercial use
• RSSCN7 -> Dataset of the article "Deep Learning Based Feature Selection for Remote Sensing Scene Classification"
• geonrw -> orthorectified aerial photographs, LiDAR derived digital elevation models and segmentation maps with 10 classes. With repo
• Attribute-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.
• open_earth_map -> a benchmark dataset for global high-resolution land cover mapping
• Mumbai-Semantic-Segmentation-Dataset
• GAMUS -> A Geometry-aware Multi-modal Semantic Segmentation Benchmark for Remote Sensing Data
• openWUSU -> WUSU is a semantic understanding dataset focusing on urban structure and the urbanization process in Wuhan
• RSE_Cross-city -> Cross-City Matters: A Multimodal Remote Sensing Benchmark Dataset for Cross-City Semantic Segmentation using High-Resolution Domain Adaptation Networks
• Chesapeake Roads Spatial Context (RSC)
• So2Sat-POP-DL -> Dataset discovery: So2Sat Population dataset covering 98 EU cities
• HouseTS -> Long-term, Multimodal Housing Dataset Across 30 U.S. Metropolitan Area. Uses NAIP. With paper
• 10,000 Crop Field Boundaries across India -> using Airbus SPOT
• OpenEarthMap-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
• Tokyo Land Use Land Cover Dataset -> 0.5-m resolution images, two kinds of 10-m resolution LCPs, and two kinds of 30-m resolution LCPs

Change Detection

• S2Looking -> A Satellite Side-Looking Dataset for Building Change Detection, paper
• Haiming-Z/MtS-WH-reference-map -> a reference map for change detection based on MtS-WH
• MtS-WH-Dataset -> Multi-temporal Scene WuHan (MtS-WH) Dataset
• SZTAKI -> A Ground truth collection for change detection in optical aerial images taken with several years time differences
• DSIFN -> change detection dataset, it consists of six large bi-temporal high resolution images covering six cities in China
• Road-Change-Detection-Dataset
• 3DCD -> infer 3D CD maps using only remote sensing optical bitemporal images as input without the need of Digital Elevation Models (DEMs)
• TUE-CD -> A change detection detection for building damage estimation after earthquake
• Hi-UCD -> ultra-High Urban Change Detection for urban semantic change detection
• LEVIR-CC-Dataset -> A Large Dataset for Remote Sensing Image Change Captioning
• GDCLD -> A globally distributed dataset of coseismic landslide mapping via multi-source high-resolution remote sensing images
• BANet change dataset - RS image to cadastral map
• Indian Cities Change Detection (ICCD) Dataset
• FOTBCD-datasets -> FOTBCD: A Geographically Diverse Dataset for Building Change Detection from High-Resolution Aerial Imagery
• TERRA-CD -> Temporal Remote-sensing Repository for Analyzing Change Detection: 5,221 paired Sentinel-2 images from 232 cities across the USA and Europe with 4-class land cover, 3-class vegetation change, and 13-class semantic change masks for multi-class and semantic change detection

SAR-Specific Datasets

• HRSID -> high resolution sar images dataset for ship detection, semantic segmentation, and instance segmentation tasks
• LS-SSDD-v1.0-OPEN -> Large-Scale SAR Ship Detection Dataset
• WHU-SEN-City -> A paired SAR-to-optical image translation dataset which covers 34 big cities of China
• SAR_vehicle_detection_dataset -> 104 SAR images for vehicle detection, collected from Sandia MiniSAR/FARAD SAR images and MSTAR images
• AIR-PolSAR-Seg -> a challenging PolSAR terrain segmentation dataset
• QXS-SAROPT -> Optical and SAR pairing dataset from the paper: The QXS-SAROPT Dataset for Deep Learning in SAR-Optical Data Fusion
• SynthWakeSAR -> A Synthetic SAR Dataset for Deep Learning Classification of Ships at Sea, with paper
• SAR2Opt-Heterogeneous-Dataset -> SAR-optical images to be used as a benchmark in change detection and image transaltion on remote sensing images
• OpenSARWake -> A SAR ship wake rotation detection benchmark dataset.
• NASTaR -> The NovaSAR Automated Ship Target Recognition (NASTaR) comprises 3415 ship patches extracted from NovaSAR S-band imagery, with labels aligned to AIS (Automatic Identification System) data

Specialized Applications

• MUSIC4HA -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Hot Area
• MUSIC4GC -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Golf Course
• MUSIC4P3 -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Photovoltaic Power Plants (solar panels)
• ABCDdataset -> damage detection dataset to identify whether buildings have been washed-away by tsunami
• Thermal power plans dataset
• SolarDK -> A high-resolution urban solar panel image classification and localization dataset
• BDAPPV — Aerial Images of Rooftop Photovoltaic Installations -> aerial images of rooftop PV installations in France and Belgium, with segmentation masks and installation metadata. Includes imagery from Google and IGN, supporting segmentation and classification benchmarks as well as distribution-shift evaluation across imagery sources
• Oil and Gas Infrastructure Mapping (OGIM) database -> includes locations and facility attributes of oil and gas infrastructure types that are important sources of methane emissions
• Overhead Wind Turbine Dataset - NAIP
• CAFOSat -> a large-scale, high-resolution dataset designed to support infrastructure-aware mapping of Concentrated Animal Feeding Operations (CAFOs) across the United States, using NAIP imagery
• CloudTracks: A Dataset for Localizing Ship Tracks in Satellite Images of Clouds -> the dataset consists of 1,780 MODIS satellite images hand-labeled for the presence of more than 12,000 ship tracks.
• Digital Typhoon Dataset -> aimed at benchmarking machine learning models for long-term spatio-temporal data
• BirdSAT -> 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.
• RSHaze+ -> remote sensing dehazing datasets in PhDnet: A novel physic-aware dehazing network for remote sensing images
• GMSEUS -> A comprehensive ground-mounted solar energy dataset with sub-array design metadata in the United States
• MultiviewRS -> List of remote sensing (RS) multi-view datasets for exploring multi-view learning
• OpenSatMap -> for large-scale map construction and downstream tasks like autonomous driving

Agricultural & Environmental

• AgroFusion -> GitHub. A pan-European multimodal agricultural mapping dataset for approximately 200,000 validated LUCAS points, combining in-situ field photos and land-cover and land-use annotations with Sentinel-1, Sentinel-2, and Landsat-8 pixel time series, ERA5-Land variables, elevation, soil properties, and annual embeddings.
• FBIS-73M -> GitHub. Large-scale, multi-resolution agricultural field instance segmentation dataset used to train Delineate Anything v2, with 1,478,096 satellite image patches at 0.25–10 m resolution and 74,015,889 field masks. Paper: Delineate Anything v2: A Global Foundation Model for Field Delineation
• EuroCropsV2 -> harmonized European crop-parcel data distributed as 153 GeoParquet files by NUTS region and year, plus 18 multi-year stack layers for crop-rotation analysis
• Hyperspectral Change Detection Dataset Irrigated Agricultural Area
• CNN-RNN-Yield-Prediction -> soybean dataset
• FireRisk -> A Remote Sensing Dataset for Fire Risk Assessment with Benchmarks Using Supervised and Self-supervised Learning
• TimeMatch -> dataset for cross-region adaptation for crop identification from SITS in four different regions in Europe
• Landsat 8 Cloud Cover Assessment Validation Data
• Remote Sensing Satellite Video Dataset for Super-resolution
• SpatioTemporalYield -> covers the USA’s top five corn-producing states: Iowa, Illinois, Indiana, Nebraska, and Minnesota.
• Palm Tree Dataset
• ts-satfire -> A Multi-Task Satellite Image Time-Series Dataset for Wildfire Detection and Prediction
• GTPBD -> Global Terraced Parcel and Boundary Dataset
• AI4SmallFarms -> 439,001 field polygons across 62 tiles in Vietnam and Cambodia for crop field boundary delineation in smallholder farms, with multitemporal Sentinel-2 composites, data
• Crop Yield Prediction MODIS -> processed 2003–2015 MODIS surface-reflectance, surface-temperature, and land-cover samples for crop-yield regression, distributed as nine-channel yearly arrays and spectral histograms. Paper: DFYP: A Dynamic Fusion Framework with Spectral Channel Attention and Adaptive Operator Learning for Crop Yield Prediction
• GeoFM Agriculture Benchmark -> representative multi-temporal crop-segmentation and change-detection chips plus fine-tuned SatMAE, Prithvi, and SpectralGPT checkpoints for four US agricultural regions. Paper: Benchmarking Geospatial Foundation Models for Agriculture Applications
• CatCrops Dataset -> approximately 32 GB of parcel-level Sentinel-2 L2A time series and ancillary crop, irrigation, variety, elevation, and location data for early crop-type classification. Paper: Early Crop Type Classification and Mapping Combining Remote Sensing and Ancillary Data
• Copernicus High Resolution Layer Croplands -> annual 10 m European crop-type and cropping-pattern products from 2017 onward, including main-crop emergence, harvest and duration, secondary crops, bare-soil periods, fallow land and cropping-season layers derived primarily from Sentinel-1/2 imagery.

Hyperspectral & Multi-modal

• AeroRIT -> A New Scene for Hyperspectral Image Analysis
• Data-CSHSI -> Open source datasets for Cross-Scene Hyperspectral Image Classification, includes Houston, Pavia & HyRank datasets
• HySpecNet-11k -> a large-scale hyperspectral benchmark dataset
• STARCOP dataset: Semantic Segmentation of Methane Plumes with Hyperspectral Machine Learning Models
• Toulouse Hyperspectral Data Set
• Toulouse Hyperspectral Data Set
• Multi-modality-image-matching -> image matching dataset including several remote sensing modalities
• PanCollection -> Pansharpening Datasets from WorldView 2, WorldView 3, QuickBird, Gaofen 2 sensors

Benchmark & Foundation Models

• EORSSD-dataset -> Extended Optical Remote Sensing Saliency Detection (EORSSD) Dataset
• ERA-DATASET -> A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos
• SSL4EO-S12 -> a large-scale dataset for self-supervised learning in Earth observation
• AIR-CD -> a challenging cloud detection data set called AIR-CD, with higher spatial resolution and more representative landcover types
• HRC_WHU -> High-Resolution Cloud Detection Dataset comprising 150 RGB images and a resolution varying from 0.5 to 15 m in different global regions
• University1652-Baseline -> A Multi-view Multi-source Benchmark for Drone-based Geo-localization
• benchmark_ISPRS2021 -> A new stereo dense matching benchmark dataset for deep learning
• WHU-Stereo -> A Challenging Benchmark for Stereo Matching of High-Resolution Satellite Images
• GeoPile pretraining dataset -> compiles imagery from other datasets including RSD46-WHU, MLRSNet and RESISC45 for pretraining of Foundational models
• pangaea-bench -> A Global and Inclusive Benchmark for Geospatial Foundation Models
• VRSBench: A Versatile Vision-Language Benchmark Dataset for Remote Sensing Image Understanding
• SeeFar -> Satellite Agnostic Multi-Resolution Dataset for Geospatial Foundation Models
• dynnet -> DynamicEarthNet: Daily Multi-Spectral Satellite Dataset for Semantic Change Segmentation
• Awesome-Remote-Sensing-Relative-Radiometric-Normalization-Datasets
• AISD -> Aerial Imagery dataset for Shadow Detection## Kaggle Kaggle hosts over > 200 satellite image datasets, search results here. The kaggle blog is an interesting read.

Kaggle - Amazon from space - classification challenge

• https://www.kaggle.com/c/planet-understanding-the-amazon-from-space/data
• 3-5 meter resolution GeoTIFF images from planet Dove satellite constellation
• 12 classes including - cloudy, primary + waterway etc
• 1st place winner interview - used 11 custom CNN
• FastAI Multi-label image classification
• Multi-Label Classification of Satellite Photos of the Amazon Rainforest
• Understanding the Amazon Rainforest with Multi-Label Classification + VGG-19, Inceptionv3, AlexNet & Transfer Learning
• amazon-classifier -> compares random forest with CNN
• multilabel-classification -> compares various CNN architecutres
• Planet-Amazon-Kaggle -> uses fast.ai
• deforestation_deep_learning
• Track-Human-Footprint-in-Amazon-using-Deep-Learning
• Amazon-Rainforest-CNN -> uses a 3-layer CNN in Tensorflow
• rainforest-tagging -> Convolutional Neural Net and Recurrent Neural Net in Tensorflow for satellite images multi-label classification
• satellite-deforestation -> Using Satellite Imagery to Identify the Leading Indicators of Deforestation, applied to the Kaggle Challenge Understanding the Amazon from Space

Kaggle - DSTL segmentation challenge

• https://www.kaggle.com/c/dstl-satellite-imagery-feature-detection
• 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
• WorldView 3 - 45 satellite images covering 1km x 1km in both 3 (i.e. RGB) and 16-band (400nm - SWIR) images
• 10 Labelled classes include - Buildings, Road, Trees, Crops, Waterway, Vehicles
• Interview with 1st place winner who used segmentation networks - 40+ models, each tweaked for particular target (e.g. roads, trees)
• ZF_UNET_224_Pretrained_Model 2nd place solution ->
• 3rd place soluton -> which explored pansharpening & calculating reflectance indices, with arxiv paper
• Deepsense 4th place solution
• Entry by lopuhin using UNet with batch-normalization
• Multi-class semantic segmentation of satellite images using U-Net using DSTL dataset, tensorflow 1 & python 2.7. Accompanying article
• Deep-Satellite-Image-Segmentation
• Dstl-Satellite-Imagery-Feature-Detection-Improved
• Satellite-imagery-feature-detection
• Satellite_Image_Classification -> using XGBoost and ensemble classification methods
• Unet-for-Satellite
• building-segmentation -> TensorFlow U-Net implementation trained to segment buildings in satellite imagery

Kaggle - DeepSat land cover classification

• https://www.kaggle.com/datasets/crawford/deepsat-sat4 & https://www.kaggle.com/datasets/crawford/deepsat-sat6
• DeepSat-Kaggle -> uses Julia
• deepsat-aws-emr-pyspark -> Using PySpark for Image Classification on Satellite Imagery of Agricultural Terrains

Kaggle - Airbus ship detection challenge

• https://www.kaggle.com/c/airbus-ship-detection/overview
• Rating - medium, most solutions using deep-learning, many kernels, good example kernel
• Detecting ships in satellite imagery: five years later…
• I believe there was a problem with this dataset, which led to many complaints that the competition was ruined
• Lessons Learned from Kaggle’s Airbus Challenge
• Airbus-Ship-Detection -> This solution scored 139 out of 884 for the competition, combines ResNeXt50 based classifier and a U-net segmentation model
• Ship-Detection-Project -> uses Mask R-CNN and UNet model
• Airbus_SDC
• Airbus_SDC_dup -> Project focused on detecting duplicate regions of overlapping satellite imagery. Applied to Airbus ship detection dataset
• airbus-ship-detection -> CNN with REST API
• Ship-Detection-from-Satellite-Images-using-YOLOV4 -> uses Kaggle Airbus Ship Detection dataset
• Image Segmentation: Kaggle experience -> Medium article by gold medal winner Vlad Shmyhlo

Kaggle - Ships in Google Earth

• https://www.kaggle.com/datasets/tomluther/ships-in-google-earth
• 794 jpegs showing various sized ships in satellite imagery, annotations in Pascal VOC format for object detection models
• /kaggle-ships-in-satellite-imagery-with-YOLOv8

Kaggle - Classify Ships in San Franciso Bay using Planet satellite imagery

• https://www.kaggle.com/datasets/rhammell/ships-in-satellite-imagery
• 4000 80x80 RGB images labeled with either a "ship" or "no-ship" classification, 3 meter pixel size
• shipsnet-detector -> Detect container ships in Planet imagery using machine learning
• DeepLearningShipDetection
• Ship-Detection-Using-Satellite-Imagery

Kaggle - Planesnet classification dataset

• https://www.kaggle.com/datasets/rhammell/planesnet -> Detect aircraft in Planet satellite image chips
• 20x20 RGB images, the "plane" class includes 8000 images and the "no-plane" class includes 24000 images
• Dataset repo and planesnet-detector demonstrates a small CNN classifier on this dataset
• ergo-planes-detector -> An ergo based project that relies on a convolutional neural network to detect airplanes from satellite imagery, uses the PlanesNet dataset
• Using AWS SageMaker/PlanesNet to process Satellite Imagery
• Airplane-in-Planet-Image -> pytorch model

Kaggle - CGI Planes in Satellite Imagery w/ BBoxes

• https://www.kaggle.com/datasets/aceofspades914/cgi-planes-in-satellite-imagery-w-bboxes
• 500 computer generated satellite images of planes
• Faster RCNN to detect airplanes
• aircraft-detection-from-satellite-images-yolov3

Kaggle - Swimming pool and car detection using satellite imagery

• https://www.kaggle.com/datasets/kbhartiya83/swimming-pool-and-car-detection
• 3750 satellite images of residential areas with annotation data for swimming pools and cars
• Object detection on Satellite Imagery using RetinaNet

Kaggle - Draper challenge to place images in order of time

• https://www.kaggle.com/c/draper-satellite-image-chronology/data
• Rating - hard. Not many useful kernels.
• 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.
• Kaggle interviews for entrants who used XGBOOST and a hybrid human/ML approach
• deep-cnn-sat-image-time-series -> uses LSTM

Kaggle - Dubai segmentation

• https://www.kaggle.com/datasets/humansintheloop/semantic-segmentation-of-aerial-imagery
• 72 satellite images of Dubai, the UAE, and is segmented into 6 classes
• dubai-satellite-imagery-segmentation -> due to the small dataset, image augmentation was used
• U-Net for Semantic Segmentation on Unbalanced Aerial Imagery -> using the Dubai dataset
• Semantic-Segmentation-using-U-Net -> uses keras
• unet_satelite_image_segmentation

Kaggle - Massachusetts Roads & Buildings Datasets - segmentation

• https://www.kaggle.com/datasets/balraj98/massachusetts-roads-dataset
• https://www.kaggle.com/datasets/balraj98/massachusetts-buildings-dataset
• Official published dataset
• Road_seg_dataset -> subset of the roads dataset containing only 200 images and masks
• Road and Building Semantic Segmentation in Satellite Imagery uses U-Net on the Massachusetts Roads Dataset & keras
• Semantic-segmentation repo by fuweifu-vtoo -> uses pytorch and the Massachusetts Buildings & Roads Datasets
• ssai-cnn -> This is an implementation of Volodymyr Mnih's dissertation methods on his Massachusetts road & building dataset
• building-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
• Road detection using semantic segmentation and albumentations for data augmention using the Massachusetts Roads Dataset, U-net & Keras
• Image-Segmentation) -> using Massachusetts Road dataset and fast.ai

Kaggle - Deepsat classification challenge

Not 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?

• 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
• Sat6 405,000 image patches each of size 28x28 and covering 6 landcover classes - barren land, trees, grassland, roads, buildings and water bodies.

Kaggle - High resolution ship collections 2016 (HRSC2016)

• https://www.kaggle.com/datasets/guofeng/hrsc2016
• Ship images harvested from Google Earth
• HRSC2016_SOTA -> Fair comparison of different algorithms on the HRSC2016 dataset

Kaggle - SWIM-Ship Wake Imagery Mass

• https://www.kaggle.com/datasets/lilitopia/swimship-wake-imagery-mass
• An optical ship wake detection benchmark dataset built for deep learning
• WakeNet -> 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

Kaggle - Understanding Clouds from Satellite Images

In this challenge, you will build a model to classify cloud organization patterns from satellite images.

• https://www.kaggle.com/c/understanding_cloud_organization/
• 3rd place solution on Github by naivelamb
• 15th place solution on Github by Soongja
• 69th place solution on Github by yukkyo
• 161st place solution on Github by michal-nahlik
• Solution by yurayli
• Solution by HazelMartindale uses 3 versions of U-net architecture
• Solution by khornlund
• Solution by Diyago
• Solution by tanishqgautam

Kaggle - 38-Cloud Cloud Segmentation

• https://www.kaggle.com/datasets/sorour/38cloud-cloud-segmentation-in-satellite-images
• Contains 38 Landsat 8 images and manually extracted pixel-level ground truths
• 38-Cloud Github repository and follow up 95-Cloud dataset
• How to create a custom Dataset / Loader in PyTorch, from Scratch, for multi-band Satellite Images Dataset from Kaggle
• Cloud-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
• Segmentation of Clouds in Satellite Images Using Deep Learning -> semantic segmentation using a Unet on the Kaggle 38-Cloud dataset

Kaggle - Airbus Aircraft Detection Dataset

• https://www.kaggle.com/airbusgeo/airbus-aircrafts-sample-dataset
• One hundred civilian airports and over 3000 annotated commercial aircrafts
• detecting-aircrafts-on-airbus-pleiades-imagery-with-yolov5
• pytorch-remote-sensing -> Aircraft detection using the 'Airbus Aircraft Detection' dataset and Faster-RCNN with ResNet-50 backbone in pytorch

Kaggle - Airbus oil storage detection dataset

• https://www.kaggle.com/airbusgeo/airbus-oil-storage-detection-dataset
• Oil-Storage Tank Instance Segmentation with Mask R-CNN with accompanying article
• Oil Storage Detection on Airbus Imagery with YOLOX -> uses the Kaggle Airbus Oil Storage Detection dataset
• Oil-Storage-Tanks-Data-Preparation-YOLO-Format

Kaggle - Satellite images of hurricane damage

• https://www.kaggle.com/datasets/kmader/satellite-images-of-hurricane-damage
• https://github.com/dbuscombe-usgs/HurricaneHarvey_buildingdamage

Kaggle - Austin Zoning Satellite Images

• https://www.kaggle.com/datasets/franchenstein/austin-zoning-satellite-images
• classify a images of Austin into one of its zones, such as residential, industrial, etc. 3667 satellite images

Kaggle - Statoil/C-CORE Iceberg Classifier Challenge

Classify 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%.

• https://www.kaggle.com/c/statoil-iceberg-classifier-challenge/data
• An interview with David Austin: 1st place winner
• radar-image-recognition
• Deep-Learning-Project -> uses keras
• iceberg-classifier-challenge solution by ShehabSunny -> uses keras
• Analyzing Satellite Radar Imagery with Deep Learning -> by Matlab, uses ensemble with greedy search
• 16th place solution
• fastai solution

Kaggle - Land Cover Classification Dataset from DeepGlobe Challenge - segmentation

• https://www.kaggle.com/datasets/balraj98/deepglobe-land-cover-classification-dataset
• Satellite Imagery Semantic Segmentation with CNN -> 7 different segmentation classes, DeepGlobe Land Cover Classification Challenge dataset, with repo
• Land Cover Classification with U-Net -> Satellite Image Multi-Class Semantic Segmentation Task with PyTorch Implementation of U-Net, uses DeepGlobe Land Cover Segmentation dataset, with code
• DeepGlobe Land Cover Classification Challenge solution

Kaggle - Next Day Wildfire Spread

A Data Set to Predict Wildfire Spreading from Remote-Sensing Data

• https://www.kaggle.com/datasets/fantineh/next-day-wildfire-spread
• https://arxiv.org/abs/2112.02447

Kaggle - Satellite Next Day Wildfire Spread

Inspired by the above dataset, using different data sources

• https://www.kaggle.com/datasets/satellitevu/satellite-next-day-wildfire-spread
• https://github.com/SatelliteVu/SatelliteVu-AWS-Disaster-Response-Hackathon

Kaggle - Spacenet 7 Multi-Temporal Urban Change Detection

• https://www.kaggle.com/datasets/amerii/spacenet-7-multitemporal-urban-development
• SatFootprint -> building segmentation on the Spacenet 7 dataset

Kaggle - Satellite Images to predict poverty in Africa

• https://www.kaggle.com/datasets/sandeshbhat/satellite-images-to-predict-povertyafrica
• Uses satellite imagery and nightlights data to predict poverty levels at a local level
• Predicting-Poverty -> Combining satellite imagery and machine learning to predict poverty, in PyTorch

Kaggle - NOAA Fisheries Steller Sea Lion Population Count

• https://www.kaggle.com/competitions/noaa-fisheries-steller-sea-lion-population-count -> count sea lions from aerial images
• Sealion-counting
• Sealion_Detection_Classification

Kaggle - Arctic Sea Ice Image Masking

• https://www.kaggle.com/datasets/alexandersylvester/arctic-sea-ice-image-masking
• sea_ice_remote_sensing

Kaggle - Overhead-MNIST

• A Benchmark Satellite Dataset as Drop-In Replacement for MNIST
• https://www.kaggle.com/datasets/datamunge/overheadmnist -> kaggle
• https://arxiv.org/abs/2102.04266 -> paper
• https://github.com/reveondivad/ov-mnist -> github

Kaggle - Satellite Image Classification

• https://www.kaggle.com/datasets/mahmoudreda55/satellite-image-classification
• satellite-image-classification-pytorch

Kaggle - EuroSAT - Sentinel-2 Dataset

• https://www.kaggle.com/datasets/raoofnaushad/eurosat-sentinel2-dataset
• RGB Land Cover and Land Use Classification using Sentinel-2 Satellite
• Used in paper Image Augmentation for Satellite Images

Kaggle - Satellite Images of Water Bodies

• https://www.kaggle.com/datasets/franciscoescobar/satellite-images-of-water-bodies
• pytorch-waterbody-segmentation -> UNET model trained on the Satellite Images of Water Bodies dataset from Kaggle. The model is deployed on Hugging Face Spaces

Kaggle - NOAA sea lion count

• https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count
• noaa -> UNET, object detection and image level regression approaches

Kaggle - miscellaneous

• https://www.kaggle.com/datasets/reubencpereira/spatial-data-repo -> Satellite + loan data
• https://www.kaggle.com/datasets/towardsentropy/oil-storage-tanks -> Image data of industrial oil tanks with bounding box annotations, estimate tank fill % from shadows
• https://www.kaggle.com/datasets/airbusgeo/airbus-wind-turbines-patches -> Airbus SPOT satellites images over wind turbines for classification
• https://www.kaggle.com/datasets/aceofspades914/cgi-planes-in-satellite-imagery-w-bboxes -> CGI planes object detection dataset
• https://www.kaggle.com/datasets/atilol/aerialimageryforroofsegmentation -> Aerial Imagery for Roof Segmentation
• https://www.kaggle.com/datasets/andrewmvd/ship-detection -> 621 images of boats and ships
• https://www.kaggle.com/datasets/alpereniek/vehicle-detection-from-satellite-images-data-set
• https://www.kaggle.com/datasets/sergiishchus/maxar-satellite-data -> Example Maxar data at 15 cm resolution
• https://www.kaggle.com/datasets/cici118/swimming-pool-detection-algarves-landscape
• https://www.kaggle.com/datasets/donkroco/solar-panel-module -> object detection for solar panels
• https://www.kaggle.com/datasets/balraj98/deepglobe-road-extraction-dataset -> segment roads
• https://www.kaggle.com/datasets/towardsentropy/oil-storage-tanks -> Image data of industrial Oil Storage Tanks with bounding box annotations
• https://www.kaggle.com/competitions/widsdatathon2019/ -> Palm oil plantations
• https://www.kaggle.com/datasets/siddharthkumarsah/ships-in-aerial-images -> Ships/Vessels in Aerial Images
• https://www.kaggle.com/datasets/jangsienicajzkowy/afo-aerial-dataset-of-floating-objects -> Aerial dataset for maritime Search and Rescue applications
• https://www.kaggle.com/datasets/yaroslavnaychuk/satelliteimagesegmentation -> Segmentation on Gaofen Satellite Image, extracted from GID-15 dataset

Competitions

Competitions are an excellent source for accessing clean, ready-to-use satellite datasets and model benchmarks.

• https://codalab.lisn.upsaclay.fr/competitions/9603 -> object detection from diversified satellite imagery
• https://www.drivendata.org/competitions/143/tick-tick-bloom/ -> detect and classify algal bloom
• https://www.drivendata.org/competitions/81/detect-flood-water/ -> map floodwater from radar imagery
• https://platform.ai4eo.eu/enhanced-sentinel2-agriculture -> map cultivated land using Sentinel imagery
• https://www.diu.mil/ai-xview-challenge -> multiple challenges ranging from detecting fishing vessals to estimating building damages
• https://competitions.codalab.org/competitions/30440 -> flood detection
• https://www.drivendata.org/competitions/83/cloud-cover/ -> cloud cover detection
• https://www.drivendata.org/competitions/78/overhead-geopose-challenge/page/372/ -> predicts geocentric pose from single-view oblique satellite images
• https://www.drivendata.org/competitions/60/building-segmentation-disaster-resilience/ -> building segmentation
• https://captain-whu.github.io/DOTA/ -> large dataset for object detection in aerial imagery
• https://spacenet.ai/ -> set of 8 challenges such as road network detection

• https://spaceml.org/repo/project/6269285b14d764000d798fde -> ML for floods
• https://spaceml.org/repo/project/60002402f5647f00129f7287 -> lightning and extreme weather
• https://spaceml.org/repo/project/6025107d79c197001219c481/true -> ~1TB dataset for precipitation forecasting
• https://spaceml.org/repo/project/61c0a1b9ff8868000dfb79e1/true -> Sentinel-2 image super-resolution