HidalgoTailings-UAV: A Multiband and Multilabel UAV Dataset of Mine-Tailings Environments in Hidalgo, Mexico

HidalgoTailings-UAV is a geospatial UAV dataset developed for pixel-level semantic and multilabel analysis of mine-tailings environments in Hidalgo, Mexico. The dataset contains 4,531 spatially aligned 256 × 256 pixel tiles distributed across six UAV photogrammetric study areas: CEUNI/Jales de Jaltepec, Omitlán, La Sabina, Venta Prieta, SAGO/La Alberca, and Zimapán Zim. The dataset is distributed as six site-specific ZIP archives. Each archive contains a data_set directory organized into three main components: images, masks, and multi_classes. The images directory contains the geospatial raster tiles used as model inputs. These tiles combine RGB information with normalized spatial and elevation-derived information obtained from UAV photogrammetric products. The masks directory contains the individual binary semantic masks. The valid populated annotation directories use the English class names: tailings, slope, crack, water, vegetation, trees, road, building, and car. Some archives may retain legacy Spanish-named directories generated during dataset preparation; these directories are empty and should be ignored. The multi_classes directory contains the combined multilabel masks corresponding to each image tile. Multilabel annotations are encoded using an overlapping bit-mask representation: tailings = 1, slope = 2, crack = 4, water = 8, vegetation = 16, trees = 32, road = 64, building = 128, and car = 256. Multiple classes may therefore coexist at the same pixel. For example, a pixel simultaneously labeled as tailings, water, and road is encoded as 1 + 8 + 64 = 73. The combined multiclass masks require an integer representation capable of preserving the value 256, such as uint16. Input raster tiles and their corresponding individual and multiclass masks share the same tile identifier, allowing direct correspondence between input data and semantic annotations. The dataset is intended for research in UAV photogrammetry, semantic segmentation, multilabel segmentation, GeoAI, geographic transfer, site-held-out validation, class imbalance analysis, multiband feature evaluation, and geospatial reconstruction of predicted mine-tailings masks. Study-site locations and approximate center coordinates The six study areas are located in Hidalgo, Mexico. Approximate center coordinates in geographic coordinates (WGS 84) are provided for dataset traceability: CEUNI / Jales de Jaltepec (Pachuca–Mineral de la Reforma region): 20.105591° N, 98.712630° W Omitlán: 20.180913° N, 98.646886° W La Sabina (Zimapán region): 20.727723° N, 99.384645° W Venta Prieta (Pachuca region): 20.076232° N, 98.760003° W SAGO / La Alberca (Zimapán region): 20.738031° N, 99.401566° W Zimapán Zim: 20.728279° N, 99.392710° W These coordinates correspond to approximate dataset/site centers and are provided for geographic traceability only; they should not be interpreted as certified property boundaries or legal facility limits. The released geospatial raster products use WGS 84 / UTM Zone 14N (EPSG:32614). The six sites contribute the following number of tiles to the complete dataset: CEUNI / Jales de Jaltepec: 654 tiles Omitlán: 361 tiles La Sabina: 763 tiles Venta Prieta: 1,387 tiles SAGO / La Alberca: 557 tiles Zimapán Zim: 809 tiles This gives a total of 4,531 tiles. The dataset supports reproducible comparison of segmentation architectures, including U-Net and MobileNetV3-Large + LR-ASPP, as well as studies of geographic domain shift, leave-one-site-out validation, class imbalance, band contribution, and georeferenced reconstruction of predicted masks.