Characterization of Industrial Smoke Plumes from Remote Sensing Data

<strong>Characterization of Industrial Smoke Plumes from Remote Sensing Data</strong><br> This data set contains imaging data acquired by ESA's Sentinel-2 Earth-observing satellite constellation for a sample of industrial sites that were picked based on emission information provided by the European Pollutant Release and Transfer Register. The images contain scenes of mainly industrial sites, some of which are actively emitting smoke plumes. This data set was created to investigate whether it would be possible to train a deep learning model to automatically identify and segment smoke plumes from remote sensing image data. Please refer to the acknowledgements section for more on information on this project. <br> <strong>Description</strong> Each image is provided in the GeoTIFF file format, contains a total of 13 bands and georeferencing information, and has a shape of 120 x 120 pixels (corresponding to a square area with an edge length of 1.2 km on the ground). The bands are extracted from Sentinel-2 Level-2A products, except for band 10, which has been extracted from the<br> corresponding Level-1C product (this band has not been utilized in the underlying work). This repository contains a total of 21,350 images. Based on manual annotation, the image sample was split into a sample of 3,750 <em>positive</em> images that contain industrial smoke plumes, and 17,600 <em>negative</em> images that do not contain smoke plumes. Furthermore, this repository contains a collection of JSON files that hold manual segmentation labels for smoke plumes present in 1,437 images. Segmentation labels were generated using label-studio. Please note that polygon edge coordinates have to be scaled by a factor of 1.2 to fit the images. <br> <strong>Content</strong> The following tarballs are contained in this repository: README.md - this file images.tar.gz [6.0GB] - contains 21,350 GeoTIFF images segmentation_labels.tar.gz [350KB] - contains 1,437 JSON files <strong>Acknowledgement</strong> If you use this data set, please cite our publication: <em>Mommert, M., Sigel, M., Neuhausler, M., Scheibenreif, L., Borth, D., "Characterization of Industrial Smoke Plumes from Remote Sensing Data", Tackling Climate Change with Machine Learning workshop at NeurIPS 2020.</em> Please refer to this publication for additional information on the data set. The code used for this publication is available at github. This data set contains modified Copernicus Sentinel data acquired in 2019, processed by ESA. <strong>Responsible Author</strong> Michael Mommert<br> University of St. Gallen, Institute of Computer Science<br> Chair Artificial Intelligence and Machine Learning<br> michael.mommert ( at ) unisg.ch