Linear-Quadratic NMF-Based Urban Hyperspectral Data Unmixing With Some Known Endmembers

摘要

Thanks to their high spectral resolution, hyperspectral remote sensing data provide a wide range of information for urban surface characterization. In this work, a new method is proposed to unmix urban hyperspectral remote sensing data by exploiting some known urban endmember spectra. The proposed approach is designed for the linear-quadratic mixing model involved in the considered images. It uses a multiplicative linear-quadratic nonnegative matrix factorization (LQNMF) and can be considered as a partial/informed LQNMF approach. Experiments are carried out on realistic synthetic data to evaluate the performance of the proposed approach. Obtained results prove that the designed method yields better performance than methods from the literature.