Satellite Image Classification

## Overview

Satellite Image Classification is a dataset for object detection tasks - it contains Tanks Vehicles Tents annotations for 101 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [Public Domain license](https://creativecommons.org/licenses/Public Domain).

1) Data Introduction • The Satellite Image Classification Dataset is a benchmark image classification dataset constructed using satellite remote sensing imagery. It includes a total of four land surface classes—cloudy, desert, green_area, and water—collected from various sensor-based images and Google Maps snapshots. The dataset is designed for training and evaluating image-based scene recognition models.

2) Data Utilization (1) Characteristics of the Satellite Image Classification Dataset: • The dataset was collected with the aim of automatic interpretation of satellite imagery and consists of a combination of sensor-based images and map snapshots, offering a realistic representation of real-world conditions. • All images are of fixed resolution and include diverse landform features, making the dataset suitable for classification experiments across different environments and for evaluating model generalization performance.

(2) Applications of the Satellite Image Classification Dataset: • Land surface classification model training: Can be used in experiments to classify various types of terrain such as buildings, farmland, and roads. • Research and application in geospatial information analysis: Useful for developing models that support spatial decision-making through tasks such as land use monitoring, urban structure analysis, and land surface inference.

CoastSat image classification training data

CoastSat is an open-source global shoreline mapping toolbox, available at https://github.com/kvos/CoastSat, which enables users to extract time-series of shoreline change from 30+ years of publicly available satellite imagery (Landsat 5, 7, 8 and Sentinel-2).

The automated shoreline extraction relies on a classifier (Multilayer Perceptron from scikit-learn) which labels each pixels on the images with one of four classes: sand, water, white-water and other land features.

The data used to train the classifier is stored here, the README.md file provides information on the data organisation and content of each file.

AutoTrain Dataset for project: satellite-image-classification

  Dataset Descritpion

This dataset has been automatically processed by AutoTrain for project satellite-image-classification.

  Languages

The BCP-47 code for the dataset's language is unk.

  Dataset Structure





  Data Instances

A sample from this dataset looks as follows: [ { "image": "<256x256 CMYK PIL image>", "target": 0 }, { "image": "<256x256 CMYK PIL image>"… See the full description on the dataset page: https://huggingface.co/datasets/victor/autotrain-data-satellite-image-classification.

This dataset provides annotated very-high-resolution satellite RGB images extracted from Google Earth to train deep learning models to perform instance segmentation of Juniperus communis L. and Juniperus sabina L. shrubs. All images are from the high mountain of Sierra Nevada in Spain. The dataset contains 810 images (.jpg) of size 224x224 pixels. We also provide partitioning of the data into Train (567 images), Test (162 images), and Validation (81 images) subsets. Their annotations are provided in three different .json files following the COCO annotation format.

Datasets UOPNOA and UOS2.

Each dataset contains images and labels to train and test a semantic segmentation model for crop classification / land use with satellite or aircraft imagery. The region of intereset is the northern

• UPNOA is made out of PNOA aircraft imagery and uses RGB images. (34.000 images)
• UOS2 is made out of Sentinel-2 satellite imagery and uses 13 bands or channels per image. (2.000 images)

Ground truth masks were made from SIGPAC data from the northern part of the Iberian Peninsula plateau in Spain.

Originally trained with UNet and DeepLabv3+

Please cite the original paper, which can be found at:

https://doi.org/10.3390/rs13122292

BibTex:

@article{pedrayes2021evaluation,
 title={Evaluation of Semantic Segmentation Methods for Land Use with Spectral Imaging Using Sentinel-2 and PNOA Imagery},
 author={Pedrayes, Oscar D and Lema, Dar{\'\i}o G and Garc{\'\i}a, Daniel F and Usamentiaga, Rub{\'e}n and Alonso, {\'A}ngela},
 journal={Remote Sensing},
 volume={13},
 number={12},
 pages={2292},
 year={2021},
 publisher={Multidisciplinary Digital Publishing Institute}
}

1) Data Introduction • The Land-Use Scene Classification Dataset is an image dataset built to classify land-use types in different regions based on Landsat satellite imagery.

2) Data Utilization (1) Characteristics of the Land-Use Scene Classification Dataset: • The images are collected from a diverse range of geographic environments, including urban, rural, coastal, and forested areas, making the dataset suitable for evaluating domain generalization performance. • It is based on low-resolution Landsat satellite images, yet designed to effectively distinguish various terrain and structural patterns even with limited spatial resolution.

(2) Applications of the Land-Use Scene Classification Dataset: • Development of land-use classification models: The dataset can be used to train deep learning models that automatically classify land-use types such as residential areas, roads, and farmlands from satellite imagery. • GIS-based land-use change analysis: It can support geographic information system (GIS) research to analyze land-use pattern changes over time and infer spatial utilization trends.

This dataset is part of the larger data collection, “Aerial imagery object identification dataset for building and road detection, and building height estimation”, linked to in the references below and can be accessed here: https://dx.doi.org/10.6084/m9.figshare.c.3290519. For a full description of the data, please see the metadata: https://dx.doi.org/10.6084/m9.figshare.3504413.

Imagery data from the United States Geological Survey (USGS); building and road shapefiles are from OpenStreetMaps (OSM) (these OSM data are made available under the Open Database License: http://opendatacommons.org/licenses/odbl/1.0/); and the Lidar data are from U.S. National Oceanic and Atmospheric Administration (NOAA), the Texas Natural Resources Information System (TNRIS).

This dataset consists of 1,759,830 multi-spectral image patches from the Sentinel-2 mission, annotated with image- and pixel-level land cover and land usage labels from the German land cover model LBM-DE2018 with land cover classes based on the CORINE Land Cover database (CLC) 2018. It includes pixel synchronous examples from each of the four seasons, plus an additional snowy set, spanning the time from April 2018 to February 2019. The patches were taken from 519,547 unique locations, covering the whole surface area of Germany, with each patch covering an area of 1.2km x 1.2km. The set is split into two overlapping grids, consisting of roughly 880,000 samples each, which are shifted by half the patch size in both dimensions. The images in each of the both grids themselves do not overlap.

Contents

Each sample includes:

3 10m resolution bands (RGB), 120px x 120px

1 10m resolution band (infrared), 120px x 120px

6 20m resolution bands, 60px x 60px

2 60m resolution bands, 20xp x 20px

1 pixel-level label map

2 binary masks for cloud and snow coverage

2 binary masks for easy and medium segmentation difficulties, marks areas <300px and <100px respectively

1 JSON-file containing additional meta-information

The meta.csv contains the following information about each sample:

Which season it belongs to

Which of the two grids it belongs to

Coordinates of the patch center

Whether it was acquired from Sentinel-2 Satellite A or B

Date and time of image acquisition

Snow and cloud coverage percentages

Image-level multi-class labels

Three additional image-level urbanization labels, based on the center pixel (details below)

The path to the sample

Classes

    ID
    Class




    1
    Continuous urban fabric


    2
    Discontinuous urban fabric


    3
    Industrial or commercial units


    4
    Road and rail networks and associated land


    5
    Port areas


    6
    Airports


    7
    Mineral extraction sites


    8
    Dump sites


    9
    Construction sites


    10
    Green urban areas


    11
    Sport and leisure facilities


    12
    Non-irrigated arable land


    13
    Vineyards


    14
    Fruit trees and berry plantations


    15
    Pastures


    16
    Broad-leaved forest


    17
    Coniferous forest


    18
    Mixed forest


    19
    Natural grasslands


    20
    Moors and heathland


    21
    Transitional woodland/shrub


    22
    Beaches, dunes, sands


    23
    Bare rock


    24
    Sparsely vegetated areas


    25
    Inland marshes


    26
    Peat bogs


    27
    Salt marshes


    28
    Intertidal flats


    29
    Water courses


    30
    Water bodies


    31
    Coastal lagoons


    32
    Estuaries


    33
    Sea and ocean

Urbanization classes

SLRAUM

0: None

1: Ländlicher Raum (~ rural area)

2: Städtischer Raum (~ urban area)

RTYP3

0: None

1: Ländliche Regionen (~ rural areas)

2: Regionen mit Verstädterungsansätzen (~ urbanizing areas)

3: Städtische Regionen (~ urban areas)

KTYP4

0: None

1: Dünn besiedelte ländliche Kreise

2: Kreisfreie Großstädte

3: Ländliche Kreise mit Verdichtungsansätzen

4: Städtische Kreise

Further information on the urbanization classes can be found here:

SLRAUM

https://www.bbsr.bund.de/BBSR/DE/forschung/raumbeobachtung/Raumabgrenzungen/deutschland/kreise/staedtischer-laendlicher-raum/kreistypen.html

RTYP3

https://www.bbsr.bund.de/BBSR/DE/forschung/raumbeobachtung/Raumabgrenzungen/deutschland/regionen/siedlungsstrukturelle-regionstypen/regionstypen.html

KTYP4

https://www.bbsr.bund.de/BBSR/DE/forschung/raumbeobachtung/Raumabgrenzungen/deutschland/kreise/siedlungsstrukturelle-kreistypen/kreistypen.html

License of landcover model

Bundesamt für Kartographie und Geodäsie

dl-de/by-2-0 from https://www.govdata.de/dl-de/by-2-0

© GeoBasis-DE / BKG 2022

Source of landcover model

https://gdz.bkg.bund.de/index.php/default/catalog/product/view/id/1071/s/corine-land-cover-5-ha-stand-2018-clc5-2018/

This dataset is used for urban village classification. The data source is Google Earth level-17 high-resolution remote sensing imagery (2.15m) and Tencent streetview data. The dataset contains 856 and 1714 image samples corresponding to the two categories of urban villages and non-urban villages, respectively, which are sampled in Beijing, Tianjin and Shijiazhuang. After data preprocessing, per sample contains one remote sensing image and four corresponding streetview images, and all image sizes are 224 × 224 × 3. The dataset is divided into training and test set using the ratio 7 : 3, and then the training and validation set are divided from the training set using the ratio 8 : 2.

1) Data Introduction • The Wildfire Prediction Dataset is a satellite image classification dataset constructed based on historical wildfire occurrences in Canada. The dataset consists of two classes: ‘wildfire’ (presence of wildfire) and ‘no_wildfire’ (absence of wildfire).

2) Data Utilization (1) Characteristics of the Wildfire Prediction Dataset: • Each image was extracted from satellite data using the latitude and longitude coordinates of actual wildfire locations. The dataset is designed for training deep learning models to predict the likelihood of wildfire occurrence.

(2) Applications of the Wildfire Prediction Dataset: • Development of wildfire risk prediction models: The dataset can be used to train AI models that classify whether a given region is at risk of wildfire based on satellite imagery. • Environmental monitoring and disaster response system research: Useful for building satellite-based wildfire monitoring systems that take into account climate change, terrain, and vegetation conditions.

This dataset contains satellite images of damaged and undamaged buildings in Kahramanmaraş, Turkey, following the devastating earthquake that occurred on February 6, 2023. The data can be used for training classification models to automatically distinguish between structurally impacted and unaffected buildings, which is critical for rapid disaster response and assessment.

Source: Satellite imagery collected in the aftermath of the earthquake. Classes: damaged, undamaged

⚠️ Please use this dataset responsibly, especially in contexts involving humanitarian response or sensitive geographic information.

Dataset

This dataset was created by Levrex

Contents

The dataset includes trained models, training logs, and test results on PASTIS and MTLCC semantic segmentation benchmark datasets. Both benchmark datasets these models are trained on are crop-type classification benchmarks that use time series Sentinel data as inputs. The PASTIS benchmark covers agricultural land plots in France while the MTLCC benchmark covers agricultural land plots in Germany. Code that accompanies these trained weights and records as well as code that can be used to transform benchmark inputs for use by the trained models can be found here: https://github.com/macdonaldezra/VistaFormer

Dataset consists of SAR and Optical (RGB) image pairs from Sentinel‑1 and Sentinel‑2 satellites, provided by the Technical University of Munich. Sentinel-1&2 Image Pairs, Michael Schmitt, Technical University of Munich (TUM)

We searched through images captured during the fall season in the original dataset provided by TUM, and selected images which could belong to each of the four classes: barren land, grassland, agricultural land, and urban areas. Optical images shown in the following sections give an idea of the type of images belonging to each class. We have tried to introduced as much variation as possible when selecting images for a class.

Data can be used to train a Conditional GAN. Since the images in this dataset are highly complex i.e. they are not regularized and they do not have a neat geometric pattern or orientation, it can also be used to check the robustness of a model, no matter the task.

This data set includes: (1) fine-scale snow and land cover maps from two mountainous study sites in the Western U.S., produced using machine-learning models trained to extract land cover data from WorldView-2 and WorldView-3 stereo panchromatic and multispectral images; (2) binary snow maps derived from the land cover maps; and (3) 30 m and 465 m fractional snow-covered area (fSCA) maps, produced via downsampling of the binary snow maps. The land cover classification maps feature between three and six classes common to mountainous regions and integral for accurate stereo snow depth mapping: illuminated snow, shaded snow, vegetation, exposed surfaces, surface water, and clouds. Also included are Landsat and MODSCAG fSCA map products. The source imagery for these data are the Maxar WorldView-2 and Maxar WorldView-3 Level-1B 8-band multispectral images, orthorectified and converted to top-of-atmosphere reflectance. These Level-1B images are available under the NGA NextView/EnhancedView license.

Detecting Landscape Objects on Satellite Images with Artificial Intelligence In recent years, there has been a significant increase in the use of artificial intelligence (AI) for image recognition and object detection. This technology has proven to be useful in a wide range of applications, from self-driving cars to facial recognition systems. In this project, the focus lies on using AI to detect landscape objects in satellite images (aerial photography angle) with the goal to create an annotated map of The Netherlands with all the coordinates of the given landscape objects.

Background Information

Problem Statement One of the things that Naturalis does is conducting research into the distribution of wild bees (Naturalis, n.d.). For their research they use a model that predicts whether or not a certain species can occur at a given location. Representing the real world in a digital form, there is at the moment not yet a way to generate an inventory of landscape features such as presence of trees, ponds and hedges, with their precise location on the digital map. The current models rely on species observation data and climate variables, but it is expected that adding detailed physical landscape information could increase the prediction accuracy. Common maps do not contain this level of detail, but high-resolution satellite images do.

Possible opportunities Based on the problem statement, there is at the moment at Naturalis not a map that does contain the level of detail where detection of landscape elements could be made, according to their wishes. The idea emerged that it should be possible to use satellite images to find the locations of small landscape elements and produce an annotated map. Therefore, by refining the accuracy of the current prediction model, researchers can gain a profound understanding of wild bees in the Netherlands with the goal to take effective measurements to protect wild bees and their living environment.

Goal of project The goal of the project is to develop an artificial intelligence model for landscape detection on satellite images to create an annotated map of The Netherlands that would therefore increase the accuracy prediction of the current model that is used at Naturalis. The project aims to address the problem of a lack of detailed maps of landscapes that could revolutionize the way Naturalis conduct their research on wild bees. Therefore, the ultimate aim of the project in the long term is to utilize the comprehensive knowledge to protect both the wild bees population and their natural habitats in the Netherlands.

Data Collection Google Earth One of the main challenges of this project was the difficulty in obtaining a qualified dataset (with or without data annotation). Obtaining high-quality satellite images for the project presents challenges in terms of cost and time. The costs in obtaining high-quality satellite images of the Netherlands is 1,038,575 $ in total (for further details and information of the costs of satellite images. On top of that, the acquisition process for such images involves various steps, from the initial request to the actual delivery of the images, numerous protocols and processes need to be followed.

After conducting further research, the best possible solution was to use Google Earth as the primary source of data. While Google Earth is not allowed to be used for commercial or promotional purposes, this project is for research purposes only for Naturalis on their research of wild bees, hence the regulation does not apply in this case.

The global satellite imagery and image processing services market is experiencing robust growth, driven by increasing demand across diverse sectors. The market, estimated at $15 billion in 2025, is projected to expand at a Compound Annual Growth Rate (CAGR) of 7% from 2025 to 2033, reaching approximately $25 billion by 2033. This expansion is fueled by several key factors. Firstly, advancements in satellite technology are providing higher-resolution imagery with improved accuracy and faster processing times, enabling more detailed analysis for various applications. Secondly, the rising adoption of cloud-based platforms for image processing and analytics is streamlining workflows and reducing costs for users. This is particularly crucial for smaller businesses and organizations that previously lacked access to sophisticated image processing capabilities. Thirdly, the growing need for precise geographical information across diverse sectors, including environmental monitoring, precision agriculture, urban planning, and disaster response, fuels market demand. The defense and security sector remains a significant contributor, with increasing reliance on satellite imagery for intelligence gathering and surveillance. Market segmentation reveals significant opportunities within specific application areas. The environmental sector, utilizing satellite imagery for deforestation monitoring, climate change analysis, and pollution detection, is a rapidly growing segment. Similarly, the energy and power sector leverages satellite imagery for pipeline monitoring, renewable energy resource assessment, and infrastructure management. Within image processing types, the demand for advanced data analytics is soaring, with growing adoption of artificial intelligence and machine learning for automated feature extraction and predictive analysis. While regulatory hurdles and the high initial investment cost of satellite technologies pose some challenges, the overall market outlook remains positive, driven by technological advancements, increasing data accessibility, and rising demand for location-based intelligence. Competition is intensifying amongst established players and new entrants, leading to innovation and affordability in the market.

This is the image dataset for satellite image processing which is a collection therml infrared and multispectral images .

This dataset consists of annotated high-resolution aerial imagery of roof materials in Bonn, Germany, in the Ultralytics YOLO instance segmentation dataset format. Aerial imagery was sourced from OpenAerialMap, specifically from the Maxar Open Data Program. Roof material labels and building outlines were sourced from OpenStreetMap. Images and labels are split into training, validation, and test sets, meant for future machine learning models to be trained upon, for both building segmentation and roof type classification.The dataset is intended for applications such as informing studies on thermal efficiency, roof durability, heritage conservation, or socioeconomic analyses. There are six roof material types: roof tiles, tar paper, metal, concrete, gravel, and glass.Note: The data is in a .zip due to file upload limits. Please find a more detailed dataset description in the README.md