Minimum Volume Constrained Linear Spectral Unmixing Algorithm

The mixels in the hypersepectral images directly influence the accuracy of target recognition.A large number of spectral unmixing methods are all based on the convex simplex geometry and the hypothesis of the pure pixels' existence.Actually,this hypothesis is very hard to be met in practice.We present the minimum volume constrained linear spectral unmixing algorithm,which isn't based on the presence of the pure pixels,and calculates the endmember matrix with the quadratic programming method in the reduced dimensional space.Then we estimate the abundance and extract the endmembers with least square method.Experimental results demonstrate that the proposed scheme for decomposition of mixels overall outperforms the MVC-NMF algorithm in the mass.