An Improved Bilinear Mixture Model Considering Adjacency and Shade Effects

Bilinear mixture models have shown their effectiveness for nonlinear spectral unmixing in literature. However, the influence of adjacency and shade effects is commonly overlooked. In this paper, an improved bilinear mixture model accounting for this issue is presented. Neighboring pixels of a pixel are linearly combined to represent the adjacency effect which works in both the first-order and second-order scattering parts. Moreover, abundances and the first-order scattering proportions are handled separately to consider the shadows. Model parameters are estimated by alternating proj ection gradient method. Experimental results on both simulated and real hyperspectral data verify the rationality of the proposed model.