Comparison of approaches for determining end-members in hyperspectral data

A very useful analysis approach for hyperspectral data has been linear unmixing which is a projection into a coordinate system where the coordinates are the constituent or end member spectra of the scene. The most useful technique is to determine the spectra from the image. Once these spectra are found, the image cube can be “unmixed” into fractional abundances of each material in each pixel. Several autonomous methods, ORASIS, the first real time autonomous algorithm; N-FINDR, an algorithm that determines end members by inflating a simplex, and Iterative Error Analysis (LEA) which finds end members by iterative unmixing are compared in this paper. Computer implementations of the three autonomous algorithms are evaluated using hyperspectral AVIRIS data of Cuprite, Nevada. The fractional abundance maps produced by the three algorithms are compared to a mineral map made by the USGS based upon an AVIRIS scene, and the end member spectra are compared to library spectra