Analysis of endmember extraction process in geological site

In hyperspectral data analyses, a lot of endmember extraction algorithms based on linear mixture model (LMM) have been developed. These algorithms search vertices in data cloud after projecting observed vectors onto a subspace or a hyperplane. Vertices are often considered pure materials, which are also called endmembers. In real data, after projected onto the subspace with reduced dimensions, however, vertex spectra sometimes don't exhibit the spectral features of pure materials in geological studies. Moreover, spectra considered as endmembers are not located around at vertex positions in this space. This work analyzes this fact by conducting experiments on well-known hyperspectral data, cuprite mining site, acquired by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). We also describe that, in this scene, we can get relatively meaningful spectra as endmembers in a geological sense by a band selection.