Shadow-Aware Nonlinear Spectral Unmixing for Hyperspectral Imagery

In hyperspectral imagery, differences in ground surface structures cause a large variation the optical scattering sunlit and (partly) shadowed pixels. The complexity of scene demands general spectral mixture model that can adapt to different scenarios surface. this article, we propose physics-based model, i.e., extended shadow multilinear mixing (ESMLM) accounts for typical presence shadows nonlinear effects, by considering multiple illumination sources. Specifically, diffuse solar alters as wavelength changes, requiring wavelength-dependent modeling shadows. Moreover, allow types interactions conditions. proposed is described graph-based representation, which sums up all possible radiation paths initiated Physical assumptions are made simplify resulting material abundances four physically interpretable parameters. Additionally, shadow-removed images be reconstructed. compared with other state-of-the-art models using one synthetic dataset two real datasets. Experimental results show ESMLM performs robustly various addition, parameters contain valuable information on assist performing removal outperforms works.