A Fast Spatial-Spectral NMF for Hyperspectral Unmixing

This letter proposes a fast yet efficient method to solve the hyperspectral unmixing problem in challenging unsupervised context, i.e., when endmember spectral signatures are unknown. First, coarse approximation of image is computed by spatially averaging neighboring pixels, which significantly reduces amount pixels be handled. reduced set unmixed derive solutions problem, estimates and corresponding low-resolution abundance maps. Then, plain resolution maps estimated from based on signatures. A sparsity promoting prior exploiting low map complements conventional data fitting term promote spatial smoothness while mitigating loss details edge areas. Finally, least square optimization solved obtain actual previous step. Numerical experiments show that proposed performs well compared state-of-the-art approaches literature.