Non-local sparse spectral unmixing for remote sensing imagery

Abstract

Sparse unmixing is promising acted as a semi-supervised fashion due to the availability of the spectral library. However, the spatial information hasn't been well taken into consideration. In this paper, a novel sparse spectral unmixing algorithm based on non-local means filter (NLSSU) for remote sensing imagery is proposed. Compared with the conditional sparse unmixing methods, the proposed algorithm considers not only spectral information but also spatial-contextual information which is often ignored. The proposed method was tested to function with both simulated and real imagery. Experimental results demonstrate that the proposed approach outperforms traditional sparse unmixing algorithms.