An approach based on self-organizing map and fuzzy membership for decomposition of mixed pixels in hyperspectral imagery

Spectral unmixing, which decomposes the mixed pixel into typical ground signatures (endmembers) and their fractional proportions (abundances) is a meaningful job for high-accuracy ground object recognition and quantitative remote sensing analysis. In this paper, a method for decomposition of mixed pixels which combines competitive neural network and fuzzy clustering, termed self-organizing map and fuzzy membership (SOM&FM) is proposed. The proposed method only demands some data samples as prior knowledge to train the SOM neural network in a supervised way. And the unmixing is based on the fuzzy model, which satisfies the abundances non-negative constraint (ANC) and the abundances summed-to-one constraint (ASC) automatically. Experimental results on synthetic and real hyperspectral data demonstrate that the proposed method can be used for both linear and nonlinear spectral mixture situations, and has good unmixing performances.