The end-member extraction is the foundation of the hyperspectral remote sensing information extraction and analysis and is also the key to pixel unmixing. In view of hyperspectral remote sensing data characteristics of the study area, the authors carried out the digital image processing of the radiance correction, minimum noise fraction (MNF) and pixel purity index (PPI) and, on such a basis, extracted the end-member spectra by using two-dimensional scatter diagram and three-dimensional scatter diagram, and conducted the research on the discrimination of end-member attributes. The extraction and analysis of rocks and minerals constitute the foundation for the recognition of rocks and minerals and directly affect the accuracy of the results.