A Novel Endmember Extraction Method Using Sparse Component Analysis for Hyperspectral Remote Sensing Imagery

The spectral unmixing technique is an effective method of solving the mixed pixel problem in the hyperspectral remote sensed imagery. During the process, endmember extraction algorithm (EEA) is significant for the creation of material abundance maps. However, the traditional EEAs are not very reliable due to the low resolution of sensor and the complex diversity of land cover feature distribution. In addition, the mutually independent assumption of the endmember corresponding abundance will be affected accordingly. In order to overcome the above limitations, a novel endmember extraction method using sparse component analysis for hyperspectral remote sensing imagery (EESCA) has been presented in this paper. EESCA assumes that each pixel in the image scene is a sparse linear mixture of all endmembers. First, the hyperline clustering algorithm is incorporated to consider the subspace clustering of all pixels after the initialization of endmember mixing matrix. It enlarges differences among ground objects and helps finding endmembers with smaller spectrum divergences. After that, the K-SVD is proposed to search the real endmembers for sparse representations with coefficients summarized in the mixing matrix. The method transfers the pure endmember extraction problem into an optimization problem by minimizing the residual errors. The four state-of-the art methods are implemented to make comparisons with the performance of EESCA. The robustness of the proposed algorithm is verified through both simulated images and real satellite images. Experimental results show that the EESCA outperforms other methods in spectral angle distance and root-mean-square error, and especially could identify accurate endmembers for ground objects with smaller spectrum divergences.