The Research On The Endmember Extraction Algorithms For Hyperspectral Remote Sensing Images

Hyperspectral remote sensing images usually contains hundreds of bands.Hyperspectral remote sensing image has become a hot spot in the field because of its high spectral resolution.However,its low spatial resolution and complex surface distribution make mixed pixels in hyperspectral imagery a commonly seen phenomenon.Therefore,to have accurate recognition and classification of hyperspectral image.The first step is to decompose the mixed pixels.Endmember extraction is a key step in decomposing hyperspectral mixed pixels.This paper focuses on the endmember extraction of hyperspectral image.The algorithm of endmember extraction based on single body expansion is proposed.This paper is based on the theory of linear representation and the theory of convex cone model.Proofing that if vector in vitro and in vivo is represented by vertex vector,there will be a negative coefficient.The theory gives a necessary and sufficient condition for the endmember identify.A new iterative algorithm is proposed under non-ideal situation.The experimental results show that the precision of the endmember extraction using the algorithm proposed in this paper is better than VCA algorithm.Our algorithm has high stability and low sensitivity to noise.However,the algorithm has a low efficiency.To overcome this shortcoming two improved algorithms are proposed.The first improved algorithm is based on the theory that the endmember is usually expressed in pixels with negative coefficients.Under the premise of not affecting the precision of endmember extraction.It presents an algorithm of endmember extraction based on pixel gradually decreased.The second improved algorithms combines with the theory of spectral information entropy,which believes that when spectral information entropy is smaller,the pixel is relatively pure.The total number of initial pixels is reduced by setting a threshold.A preprocessing method based on the extension of the endmember is presented.The two improved algorithms can reduce the running time of the algorithm on the premise of guaranteeing the precision of endmember extraction.