The goal of hyperspectral unmixing is to decompose an electromagneticspectral dataset measured over M spectral bands and T pixels into N constituentmaterial spectra (or "end-members") with corresponding spatial abundances. Inthis paper, we propose a novel approach to hyperspectral unmixing based onloopy belief propagation (BP) that enables the exploitation of spectralcoherence in the endmembers and spatial coherence in the abundances. Inparticular, we partition the factor graph into spectral coherence, spatialcoherence, and bilinear subgraphs, and pass messages between them using a"turbo" approach. To perform message passing within the bilinear subgraph, weemploy the bilinear generalized approximate message passing algorithm(BiG-AMP), a recently proposed belief-propagation-based approach to matrixfactorization. Furthermore, we propose an expectation-maximization (EM)strategy to tune the prior parameters and a model-order selection strategy toselect the number of materials N. Numerical experiments conducted with bothsynthetic and real-world data show favorable unmixing performance relative toexisting methods.
展开▼
