Nonlocal weighted sparse unmixing based on global search and parallel optimization

Sparse unmixing (SU) can represent an observed image using pure spectral signatures and corresponding fractional abundance from a large library is important technique in hyperspectral unmixing. However, the existing SU algorithms mainly exploit spatial information fixed neighborhood system, which not sufficient. To solve this problem, we propose nonlocal weighted algorithm based on global search (G-NLWSU). By exploring similarity of image, weights pixels are calculated to form matrix weight matrix. Specifically, G-NLWSU first searches for similar group each pixel scope then uses singular value decomposition denoise finally obtains by considering correlations between pixels. reduce execution burden G-NLWSU, parallel computing version named PG-NLWSU, integrates compute unified device architecture-based into accelerate groups nonlocally Our proposed shed new light exploitation process scenario. Experimental results conducted simulated real datasets show that PG-NLWSU superior comparison algorithms.