A Novel Geo-Stat Endmember Extraction Algorithm

The blind hyperspectral unmixing chain can be divided into three sub-problems. First is subspace identification, second is endmember extraction and third is abundance estimation. Endmember extraction is a very challenging problem in this chain. There are mainly three approaches to extract endmembers: Sparse, Statistical and Geometrical. The sparse approach is useful only when we have a spectral library. In the absence of the spectral library, either a statistical approach or geometrical approach is useful. In this paper, a novel Geo-Stat Endmember Extraction (GSEE) algorithm is proposed which uses statistical and geometrical features of the data for endmember extraction. The proposed algorithm uses covariance based two bands selection to find convex set. The proposed framework and other state-of-art algorithms are applied to synthetic and real data. The performance of the proposed and other algorithms is compared using standard evaluation parameters. The simulation results show that the proposed algorithm outperforms.