Abstract: Spectral endmember extraction is an important pretreatment for the further analysis of hyperspectral data. Regarding many kinds of endmember extraction algorithms, N-FINDR algorithm is widely utilized for its full-automation and better endmember extraction performance. However, the order of the samples has a certain effect on the endmember extraction, and traditional N-FINDR algorithm also needs to reduce the dimensionality based on the number of the endmembers, which will limit its application. In the actual hyperspectral data, the incompact clustering of the same species presented in the high dimensional space also increases the difficulty of endmember extraction. So this paper proposed an improved stop rule and the pretreatment of the features, and utilizing Support Vector Machine (SVM) to conduct the second endmember extraction. Experiments show that the improved stop rule further increased the volume of the convex polyhedron composed of the endmembers. The pretreatment of the features and the second SVM endmember extraction increase the separability of the data and the precision of the extracted endmembers respectively.