Hyperspectral Image Unmixing Based On Iterative Spectral Mixture Analysis

For the low spatial resolution, mixed pixels are prevalent in the Hyperspectral Image（HSI），which seriously affects the feature identification and classification accuracy. To solve the problem,Spectral Unmixing proposed to identify feature types and estimate abundance fraction is aquantitative remote sensing technology. The raditional spectral unmixing method devided intoendmember extraction for and pixel unmixing is based on accuracy endmembers that constructmixed pixel, on the contrary, blind unmixing directly access endmember and abundance withoutthe information.This paper analyzes and compares the pros and cons of traditional spectral unmixing andblind unmixing, proposes a Hyperspectral image unmixing method base on iterative spectralmixture analysis in combination of the tow methods.At first, a common spectral mixture model and a pixel unmixing method with excessiveendmembers are introduced in this paper, then the mixed pixel decomposition method based oniterative spectral mixture analysis is validated with real hyperspectral data; Second, non-negativematrix factorization is analyzed, and weighted non-negative matrix factorization whose weightedmatrix is settled is proposed as a Blind Unmixing algorithm; Finally, the non-negative matrixfactorization algorithm conbines the mixed pixel decomposition method based on iterativespectral mixture analysis algorithm, the endmember extraction algorithm called iterative erroranalysis, and the endmember update method based on statistical endmember to a blind unmixingmethod which is validated with real and simulation hyperspectral data.