A Sentinel-2 dataset for road hierarchy classification across the Iberian Peninsula

This dataset contains 20,000 Sentinel-2 true-color image chips for multi-class road hierarchy classification across the Iberian Peninsula, covering Spain and Portugal. The collection comprises 5,000 chips labelled motorway, 5,000 chips labelled major-road, 5,000 chips labelled minor-road, and 5,000 chips labelled no-road. Each image is provided as a 224 × 224 pixel RGB PNG file and represents a 512 m × 512 m geographic footprint. Motorway sample locations were derived from OpenStreetMap geometries tagged highway=motorway. Major-road locations were derived from geometries tagged highway=trunk or highway=primary, whereas minor-road locations were derived from geometries tagged highway=secondary or highway=tertiary. Locations for all road classes were spatially balanced across the retained road networks and selected using distance thinning. Candidates were accepted only when their complete image footprints did not overlap previously accepted samples or intersect retained road geometries assigned to another positive class. Road segments marked as tunnels were excluded from positive sample selection. Road-absent locations were generated from the local surroundings of retained road locations. They were accepted only when their complete 512 m × 512 m footprints did not intersect retained motorway, major-road, or minor-road geometries and did not overlap accepted image footprints. The no-road label therefore denotes the absence of the road categories included in the dataset definition; it does not imply the absence of all transport-related features, such as residential roads, service roads, tracks, paths, or cycleways. Image chips were generated from Sentinel-2 Level-2A imagery using bands B04, B03, and B02 as red, green, and blue channels. Imagery was retrieved through the Copernicus Data Space Ecosystem Sentinel Hub Process API for the period from 26 June 2025 to 25 June 2026. The collection was generated using a maximum cloud-cover threshold of 10 percent and least-cloud-cover mosaicking. The repository includes the image collection and a CSV metadata table containing the image filename, class label, and stable candidate identifier. The dataset can be used for supervised multi-class image classification, benchmarking of computer vision models, geographically structured validation, representation learning, road-infrastructure mapping, and research on transport infrastructure.