"hyperspectral dataset"

"This benchmark provides a curated collection of 9 hyperspectral remote sensing datasets spanning 5 downstream tasks, designed to evaluate domain-independent spectral representations from foundation models. The datasets cover classification (Indian Pines, Pavia University, Houston), anomaly detection (Pavia), target detection (San Diego), change detection (Bay Area, Hermiston, Santa Barbara), and spectral unmixing (Urban). Data was acquired from four different airborne sensors (AVIRIS, ROSIS, ITRES CASI, HYDICE) with spectral configurations ranging from 102 to 242 bands. All datasets are provided in MATLAB .mat format compatible with Python (scipy) and MATLAB. The benchmark was assembled to validate the Hyper-Focus spectral foundation model, demonstrating that domain-independent spectral representations can generalize across diverse tasks without task-specific fine-tuning."