The effect of pre-processing steps on geometric based hyperspectral unmixing algorithms

Abstract

Hyperspectral unmixing is a process to find number of surface materials (called endmember), estimation of their signatures, their abundance fractions in each pixel on the scene. Geometric based algorithms are developed for hyperspectral unmixing problem in the literature. The distribution of spectra (points in n-dimensional scatterplot) can be used to estimate endmember signatures geometrically. Both denoising and the use of spatial information in the hyperspectral image as a pre-processing step may cause the endmember signatures to be estimated closer to the truth. For this purpose, Spatio-spectral total variation (SSTV) is used for denoising and Spatial pre-processing (SPP) is used for the use of spatial information. Experiments on real data have shown that these pre-processing steps lead to closer endmember signature estimates.