Decomposition of Mixed Pixels Based on Self-Organizing Map and Fuzzy Membership

The mixed-pixels exist in the remote sensing images popularly,and decomposition of these mixed pixels into endmembers and their abundances are very meaningful for high-accuracy ground object recognition and quantitative remote sensing.A new method,which combines self-organizing map(SOM) neural network and fuzzy membership in the fuzzy theory,is proposed for decomposing mixed pixels in multispectral hyperspectral remote sensing images.It trains the SOM in a supervised way firstly,and then decomposes the mixed pixels based on fuzzy model.The decomposed result satisfies two constraints which are demanded for the problem of the decomposition of mixed pixels automatically:abundances non-negative constraint and abundances summed-to-one constraint.Experimental results demonstrate that the proposed method can be used for both linear spectral mixture and nonlinear spectral mixture,achieves good decomposed results and has strong anti-noise ability.