Constrained Nonnegative Matrix Factorization Used for Spectral Unmixing

The mixels in the hyperspectral images directly influence the accuracy of target recognition. The ICE algorithm doesn´t extract the endmembers based on the hypothesis of the pure pixels´ existence, and gets good performance in the spectral unmixing application. After analyzing the theory of the the ICE algorithm and nonnegative matrix factorization, the method of hyperspectral image unmixing via endmembers´ sum of squared distance constrained nonnegative matrix factorization was presented. Experimental results demonstrated that the proposed scheme for decomposition of mixels outperforms the ICE algorithm..