A Multi-Label Sentinel-2 Dataset for Deep Learning-Based Energy, Transport, and Storage Infrastructure Mapping in Germany

This dataset contains 10,000 Sentinel-2 true-color image chips for multi-label energy, transport, and storage infrastructure mapping across Germany. Each image is annotated independently for the presence of selected infrastructure categories, including electrical substations, power plants, storage tanks, railway infrastructure, and major road infrastructure. The labels are not mutually exclusive, so each image may contain none, one, or multiple infrastructure types. Each image is provided as a 224 × 224 pixel RGB PNG file and represents a 512 m × 512 m geographic footprint. OpenStreetMap geometries were used to derive the label definitions. Electrical substations were derived from power=substation, power plants from power=plant, storage tanks from man_made=storage_tank, railway infrastructure from railway=rail and railway=yard, and major road infrastructure from highway=motorway, highway=trunk, and highway=primary. Candidate locations were sampled from retained linear and polygonal geometries and spatially balanced across Germany. For each accepted footprint, binary labels were assigned based on intersections with the retained OpenStreetMap geometries. Locations without any positive label were generated in the local surroundings of retained infrastructure features and accepted only when their complete footprints did not intersect any of the selected label geometries. Therefore, the absence of a label denotes only the absence of the infrastructure categories included in the dataset definition and does not imply the absence of all human-made or transport-related features. Image chips were generated from Sentinel-2 Level-2A imagery using bands B04, B03, and B02 as red, green, and blue channels. The repository includes the image collection and CSV metadata tables containing filenames, stable identifiers, coordinates, and binary label annotations. The dataset can be used for supervised multi-label image classification, deep learning benchmarking, geospatial representation learning, and infrastructure mapping.