Minerals in the Wild: A Paired Hyperspectral-XRF Dataset for Elemental Composition Estimation

Minerals in the Wild is a paired hyperspectral imaging and X-ray fluorescence dataset for elemental composition estimation. It contains calibrated short-wave infrared (SWIR) hyperspectral image cubes and matched XRF elemental compositions for 1,132 natural rock specimens collected across Europe. Each hyperspectral cube contains 273 spectral bands covering 996.34–2504.28 nm and corresponds one-to-one with a specimen row in the XRF table. The released XRF table contains 51 harmonised elemental mass-fraction columns. The hyperspectral cubes are provided as NumPy float32 files. Background locations within each rectangular specimen crop are represented by NaN values and should be excluded using NaN-aware processing. The deposit includes four ZIP archives containing the hyperspectral cubes, the matched XRF table, an asset manifest, and SHA-256 checksums for file verification. Documentation, metadata and usage examples are available at: https://github.com/EleftheriaTtl/minerals-in-the-wild