Shadow-Aware Nonlinear Spectral Unmixing With Spatial Regularization

This paper presents a nonlinear spectral unmixing method jointly considering shadow effect and spatial relationships in local neighborhoods. Sunlit and shadowed spectra are modeled by considering two illumination sources, i.e., direct and diffuse solar radiation. Specifically, we model the spectrum of a material in shadowed regions according to the spectrum of the same material exposed to direct sunlight. Furthermore, we embed a weighted total variation regularization in order to keep the spatial relationships among pixels. Weighting factors take into account the similarity of neighboring pixels by considering the spectral information from the hyperspectral imagery, height information from the image-generated digital surface model (DSM), and shadow effects. The optimization problem is solved by the Alternating Direction Method of Multipliers (ADMM). Experimental results demonstrate that the proposed shadow-aware unmixing method performs better with the aid of the spatial regularization.