Optimal linear unmixing for hyperspectral image analysis

In this paper we study the linear unmixing problem for the remotely sensed hyperspectral imagery. According to the, linear mixture model, the reflectance of a pixel is considered as the linear mixture of all the materials resident in the area covered by this pixel. The abundances are subject to two constraints: sum-to-one constraint and non-negativity constraint. When endmember signatures are known, quadratic programming can be used for estimating the abundances satisfying these two constraints. When endmember signatures are partially or completely unknown, they are generated from the image scene directly using least squares criterion. Computer simulation is conducted to analyze the purity of such generated endmember signatures