An improved weight-calculation non-local sparse unmixing for hyperspectral imagery

Abstract

Spatial sparse unmixing techniques have been known as a series of effective way in improving the unmixing accuracy with the integration of spatial correlations of imagery. To better utilize the non-local spatial information, spatial sparse unmixing methods based on non-local means such as nonlocal sparse unmixing (NLSU) have been proposed. However, the non-local spatial correlations in NLSU represented by weights between similar windows in the estimated abundances are always changing and not so reliable during the process of optimization. To obtain more precise and fixed spatial relationships, the improved weight-calculation non-local sparse unmixing algorithm is proposed in this paper by replacing the weight acquisition source from the variable estimated abundances to original hyperspectral imagery. The experimental results using two groups of simulated hyperspectral datasets indicate that the IW-NLSU outperforms the previous spatial sparse unmixing methods.