Satellite Image Classification

## Overview

Satellite Image Classification is a dataset for classification tasks - it contains Objects annotations for 2,000 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 [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Declassified satellite images provide an important worldwide record of land-surface change. With the success of the first release of classified satellite photography in 1995, images from U.S. military intelligence satellites KH-7 and KH-9 were declassified in accordance with Executive Order 12951 in 2002. The data were originally used for cartographic information and reconnaissance for U.S. intelligence agencies. Since the images could be of historical value for global change research and were no longer critical to national security, the collection was made available to the public. Keyhole (KH) satellite systems KH-7 and KH-9 acquired photographs of the Earth’s surface with a telescopic camera system and transported the exposed film through the use of recovery capsules. The capsules or buckets were de-orbited and retrieved by aircraft while the capsules parachuted to earth. The exposed film was developed and the images were analyzed for a range of military applications. The KH-7 surveillance system was a high resolution imaging system that was operational from July 1963 to June 1967. Approximately 18,000 black-and-white images and 230 color images are available from the 38 missions flown during this program. Key features for this program were larger area of coverage and improved ground resolution. The cameras acquired imagery in continuous lengthwise sweeps of the terrain. KH-7 images are 9 inches wide, vary in length from 4 inches to 500 feet long, and have a resolution of 2 to 4 feet. The KH-9 mapping program was operational from March 1973 to October 1980 and was designed to support mapping requirements and exact positioning of geographical points for the military. This was accomplished by using image overlap for stereo coverage and by using a camera system with a reseau grid to correct image distortion. The KH-9 framing cameras produced 9 x 18 inch imagery at a resolution of 20-30 feet. Approximately 29,000 mapping images were acquired from 12 missions. The original film sources are maintained by the National Archives and Records Administration (NARA). Duplicate film sources held in the USGS EROS Center archive are used to produce digital copies of the imagery.

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.

The satellite image of Canada is a composite of several individual satellite images form the Advanced Very High Resolution Radiometre (AVHRR) sensor on board various NOAA Satellites. The colours reflect differences in the density of vegetation cover: bright green for dense vegetation in humid southern regions; yellow for semi-arid and for mountainous regions; brown for the north where vegetation cover is very sparse; and white for snow and ice. An inset map shows a satellite image mosaic of North America with 35 land cover classes, based on data from the SPOT satellite VGT (vegetation) sensor.

Satellite images are essentially the eyes in the sky. Some of the recent satellites, such as WorldView-3, provide images with a spatial resolution of 0.3 meters. This satellite with a revisit time of under 24 hours can scan a new image of the exact location with every revisit.

Spatial resolution explained Spatial resolution is the size of the physical dimension that can be represented on a pixel of the image. Or in other words, spatial resolution is a measure of the smallest object that the sensor can resolve measured in meters. Generally, spatial resolution can be divided into three categories:

– Low resolution: over 60m/pixel. (useful for regional perspectives such as monitoring larger forest areas)

– Medium resolution: 10‒30m/pixel. (Useful for monitoring crop fields or smaller forest patches)

– High to very high resolution: 0.30‒5m/pixel. (Useful for monitoring smaller objects like buildings, narrow streets, or vehicles)

Based on the application of the imagery for the final product, a choice can be made on the resolution, as labor intensity from person-hours to computing power required increases with the resolution of the imagery.

This dataset contains satellite imagery of 4,454 power plants within the United States. The imagery is provided at two resolutions: 1m (4-band NAIP iamgery with near-infrared) and 30m (Landsat 8, pansharpened to 15m). The NAIP imagery is available for the U.S. and Landsat 8 is available globally. This dataset may be of value for computer vision work, machine learning, as well as energy and environmental analyses.Additionally, annotations of the specific locations of the spatial extent of the power plants in each image is provided. These annotations were collected via the crowdsourcing platform, Amazon Mechanical Turk, using multiple annotators for each image to ensure quality. Links to the sources of the imagery data, the annotation tool, and the team that created the dataset are included in the "References" section.To read more on these data, please refer to the "Power Plant Satellite Imagery Dataset Overview.pdf" file. To download a sample of the data without downloading the entire dataset, download "sample.zip" which includes two sample powerplants and the NAIP, Landsat 8, and binary annotations for each.Note: the NAIP imagery may appear "washed out" when viewed in standard image viewing software because it includes a near-infrared band in addition to the standard RGB data.

Multispectral satellite image data from the upper Sacramento River in northern California were acquired on October 18, 2017, to support research on remote sensing of rivers, particularly retrieval of water depth, and to facilitate efforts to characterize salmon habitat conditions and geomorphic change along the upper Sacramento River. These data were collected by the WorldView-3 (WV3) satellite, operated by DigitalGlobe and obtained through the EnhancedView license program administered by the National Geospatial-Intelligence Agency (NGA); the image data remain copyright of DigitalGlobe (2018). DigitalGlobe performed initial radiometric and geometric processing of the image. The data were acquired from the WorldView-3 satellite from an orbit with an altitude of 617 km and have a spatial resolution (pixel size) of 1.36 m. The data set consists of 8 spectral bands spanning the visible and near infrared wavelength range from 400-954 nanometers. The image pixel values represent raw digital counts and conversion to radiance, atmospheric correction, and reflectance retrieval have not been performed for the image included in this data release. The image is in a GeoTIFF format with pixel values stored as 16-bit unsigned integers. The image provided in this data release is a subset focused on the reach of the Sacramento River where it is joined by its tributary Cottonwood Creek. Supporting field data from this reach were collected in coordination with the acquisition of the remotely sensed data.

Images and 2-class labels for semantic segmentation of Sentinel-2 and Landsat RGB satellite images of coasts (water, other)

Images and 2-class labels for semantic segmentation of Sentinel-2 and Landsat RGB satellite images of coasts (water, other)

Description

4088 images and 4088 associated labels for semantic segmentation of Sentinel-2 and Landsat RGB satellite images of coasts. The 2 classes are 1=water, 0=other. Imagery are a mixture of 10-m Sentinel-2 and 15-m pansharpened Landsat 7, 8, and 9 visible-band imagery of various sizes. Red, Green, Blue bands only

These images and labels could be used within numerous Machine Learning frameworks for image segmentation, but have specifically been made for use with the Doodleverse software package, Segmentation Gym**.

Two data sources have been combined

Dataset 1

• 1018 image-label pairs from the following data release**** https://doi.org/10.5281/zenodo.7335647
• Labels have been reclassified from 4 classes to 2 classes.
• Some (422) of these images and labels were originally included in the Coast Train*** data release, and have been modified from their original by reclassifying from the original classes to the present 2 classes.
• These images and labels have been made using the Doodleverse software package, Doodler*.

Dataset 2

• 3070 image-label pairs from the Sentinel-2 Water Edges Dataset (SWED)***** dataset, https://openmldata.ukho.gov.uk/, described by Seale et al. (2022)******
• A subset of the original SWED imagery (256 x 256 x 12) and labels (256 x 256 x 1) have been chosen, based on the criteria of more than 2.5% of the pixels represent water

File descriptions

• classes.txt, a file containing the class names
• images.zip, a zipped folder containing the 3-band RGB images of varying sizes and extents
• labels.zip, a zipped folder containing the 1-band label images
• overlays.zip, a zipped folder containing a semi-transparent overlay of the color-coded label on the image (red=1=water, bllue=0=other)
• resized_images.zip, RGB images resized to 512x512x3 pixels
• resized_labels.zip, label images resized to 512x512x1 pixels

References

*Doodler: Buscombe, D., Goldstein, E.B., Sherwood, C.R., Bodine, C., Brown, J.A., Favela, J., Fitzpatrick, S., Kranenburg, C.J., Over, J.R., Ritchie, A.C. and Warrick, J.A., 2021. Human‐in‐the‐Loop Segmentation of Earth Surface Imagery. Earth and Space Science, p.e2021EA002085https://doi.org/10.1029/2021EA002085. See https://github.com/Doodleverse/dash_doodler.

**Segmentation Gym: Buscombe, D., & Goldstein, E. B. (2022). A reproducible and reusable pipeline for segmentation of geoscientific imagery. Earth and Space Science, 9, e2022EA002332. https://doi.org/10.1029/2022EA002332 See: https://github.com/Doodleverse/segmentation_gym

***Coast Train data release: Wernette, P.A., Buscombe, D.D., Favela, J., Fitzpatrick, S., and Goldstein E., 2022, Coast Train--Labeled imagery for training and evaluation of data-driven models for image segmentation: U.S. Geological Survey data release, https://doi.org/10.5066/P91NP87I. See https://coasttrain.github.io/CoastTrain/ for more information

****Buscombe, Daniel, Goldstein, Evan, Bernier, Julie, Bosse, Stephen, Colacicco, Rosa, Corak, Nick, Fitzpatrick, Sharon, del Jesús González Guillén, Anais, Ku, Venus, Paprocki, Julie, Platt, Lindsay, Steele, Bethel, Wright, Kyle, & Yasin, Brandon. (2022). Images and 4-class labels for semantic segmentation of Sentinel-2 and Landsat RGB satellite images of coasts (water, whitewater, sediment, other) (v1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7335647

*****Seale, C., Redfern, T., Chatfield, P. 2022. Sentinel-2 Water Edges Dataset (SWED) https://openmldata.ukho.gov.uk/

******Seale, C., Redfern, T., Chatfield, P., Luo, C. and Dempsey, K., 2022. Coastline detection in satellite imagery: A deep learning approach on new benchmark data. Remote Sensing of Environment, 278, p.113044.

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.

Metadata: NOAA GOES-R Series Advanced Baseline Imager (ABI) Level 1b RadiancesMore information about this imagery can be found here.This satellite imagery combines data from the NOAA GOES East and West satellites and the JMA Himawari satellite, providing full coverage of weather events for most of the world, from the west coast of Africa west to the east coast of India. The tile service updates to the most recent image every 10 minutes at 1.5 km per pixel resolution.The infrared (IR) band detects radiation that is emitted by the Earth’s surface, atmosphere and clouds, in the “infrared window” portion of the spectrum. The radiation has a wavelength near 10.3 micrometers, and the term “window” means that it passes through the atmosphere with relatively little absorption by gases such as water vapor. It is useful for estimating the emitting temperature of the Earth’s surface and cloud tops. A major advantage of the IR band is that it can sense energy at night, so this imagery is available 24 hours a day.The Advanced Baseline Imager (ABI) instrument samples the radiance of the Earth in sixteen spectral bands using several arrays of detectors in the instrument’s focal plane. Single reflective band ABI Level 1b Radiance Products (channels 1 - 6 with approximate center wavelengths 0.47, 0.64, 0.865, 1.378, 1.61, 2.25 microns, respectively) are digital maps of outgoing radiance values at the top of the atmosphere for visible and near-infrared (IR) bands. Single emissive band ABI L1b Radiance Products (channels 7 - 16 with approximate center wavelengths 3.9, 6.185, 6.95, 7.34, 8.5, 9.61, 10.35, 11.2, 12.3, 13.3 microns, respectively) are digital maps of outgoing radiance values at the top of the atmosphere for IR bands. Detector samples are compressed, packetized and down-linked to the ground station as Level 0 data for conversion to calibrated, geo-located pixels (Level 1b Radiance data). The detector samples are decompressed, radiometrically corrected, navigated and resampled onto an invariant output grid, referred to as the ABI fixed grid.McIDAS merge technique and color mapping provided by the Cooperative Institute for Meteorological Satellite Studies (Space Science and Engineering Center, University of Wisconsin - Madison) using satellite data from SSEC Satellite Data Services and the McIDAS visualization software.

Merged Satellite Flood Images

## Overview

Merged Satellite Flood Images is a dataset for object detection tasks - it contains Flood annotations for 440 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 [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

This dataset contains remote sensing data for every village in the state of Bihar, India. For most of these villages, the data contains the corresponding electrification rate as reported by the Garv data platform from the Indian government as of July 2017. This dataset contains satellite imagery, political boundaries, lights at night imagery, rainfall measurements, and vegetation indices data for 45,220 villages and the electrification rate data for 32,817 of those villages. This dataset may be of particular interest to those investigating how electricity access maps to infrastructure and agricultural production. This dataset was compiled as part of the Summer 2017 Duke University Data+ team, titled "Electricity Access in Developing Countries from Aerial Imagery."

Inventory of satellite images from DigitalGlobe’s WorldView-2 satellite purchased by the NOAA Coral Reef Ecosystem Program for Timor-leste from Jan 26, 2010 to August 10, 2014. Images were acquired for purposes of deriving seafloor depths and benthic habitat classes for the nearshore waters (< 20m depths) of Timor-leste. In addition to the inventory, the dataset includes the regions of interest (ROI) defined by NOAA used to define the geographic areas to acquire the satellite images from DigitalGlobe, the boundary extent of the available images within each ROI, and the footprint (extent) of each image. The data within each satellite image is clipped to the ROI. The ROI, boundary, and footprint files are provided in shapefile format, and the inventory is provided as a text file, which corresponds to the footprints shapefile.

PACE NetCDF images with 8 days.

"PACE's data will help us better understand how the ocean and atmosphere exchange carbon dioxide. In addition, it will reveal how aerosols might fuel phytoplankton growth in the surface ocean. Novel uses of PACE data will benefit our economy and society. For example, it will help identify the extent and duration of harmful algal blooms. PACE will extend and expand NASA's long-term observations of our living planet. By doing so, it will take Earth's pulse in new ways for decades to come."

PACE NetCFD images dataset: - source: https://oceancolor.gsfc.nasa.gov/l3/order/ - start date: 2024-03-05 - end date: 2024-10-05 - sensor: PACE-OCI - product: Phytoplankton Carbon

All rights, and licenses go to the original data provider: NASA

This data was collected during NASA space apps challenge 2024

What is this dataset?

Nearly 10,000 km² of free high-resolution and matched low-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.

Those locations are also enriched with typically under-represented locations in ML datasets: sites of humanitarian interest, illegal mining sites, and settlements of persons at risk.

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).

We accompany this dataset with a paper, datasheet for datasets and an open-source Python package to: rebuild or extend the WorldStrat dataset, train and infer baseline algorithms, and learn with abundant tutorials, all compatible with the popular EO-learn toolbox.

Why make this?

We hope to foster broad-spectrum applications of ML to satellite imagery, and possibly develop the same power of analysis allowed by costly private high-resolution imagery from free public low-resolution Sentinel2 imagery. We illustrate this specific point by training and releasing several highly compute-efficient baselines on the task of Multi-Frame Super-Resolution.

Licences

• The high-resolution Airbus imagery is distributed, with authorization from Airbus, under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).
• The labels, Sentinel2 imagery, and trained weights are released under Creative Commons with Attribution 4.0 International (CC BY 4.0).
• The source code (will be shortly released on GitHub) under 3-Clause BSD license.

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.

We observe that satellite imagery is a powerful source of information as it contains more structured and uniform data, compared to traditional images. Although computer vision community has been accomplishing hard tasks on everyday image datasets using deep learning, satellite images are only recently gaining attention for maps and population analysis. This workshop aims at bringing together a diverse set of researchers to advance the state-of-the-art in satellite image analysis.

To direct more attention to such approaches, we propose DeepGlobe Satellite Image Understanding Challenge, structured around three different satellite image understanding tasks. The datasets created and released for this competition may serve as reference benchmarks for future research in satellite image analysis. Furthermore, since the challenge tasks will involve "in the wild" forms of classic computer vision problems, these datasets have the potential to become valuable testbeds for the design of robust vision algorithms, beyond the area of remote sensing.

Contained within the 5th Edition (1978 to 1995) of the National Atlas of Canada is a map that shows Canada as seen from space in August, 1990 using uninterrupted 1.1 kilometer resolution imagery; final colors are adjusted to approximate those of the land cover portrayed.

The cost of acquiring a satellite data was highest for the images from the GeoEye-1 satellite at 25 U.S. dollars per square kilometer of the image. Most of the satellite data have a minimum order quantities based on the company and the cost depends mostly on the spatial resolution of the satellite image.

Most of the satellites are commercially owned and provide users with data as an end product based on the requirement. Processing smaller patches of the raw images obtained from a satellite to an end product are not profitable. Hence, there is a minimum order limit of 25 to 50 square kilometers based on the requested product.

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