Hyperspectral image unmixing with Nonnegative Matrix Factorization

Nonnegative Matrix Factorization (NMF) is a feature extraction method that has been also successfully applied to hyperspectral imaging. Several computational approaches have been proposed to improve NMF-based spectral unmixing. In this paper, we are concerned with several important issues related to the above application, i.e. which nonnegatively constrained algorithm is the most efficient for NMF-based hyperspectral unmixing, how to estimate the number of pure spectra, and how to accelerate the learning stage. The experiments demonstrate comparative studies, carried out for the hyperspectral images measured by the AVIRIS system.