Linear pixel unmixing is a straightforward and efficient approach to the spectral decomposition of hyperspectral remotely sensed scenes

which involves only a single pixel with the same mixing proportions.In this paper we develop a method of selective endmembers for pixel unmixing.The proposed algorithm compares pixel spectral with endmember spectral and take the neighboring pixel into account when selecting the endmember matrix(EM).Experimental results show that the selective endmember algorithm achieves improvement over conventional linear algorithms.