Automated extraction of dominant endmembers from hyperspectral image using SUnSAL and HySime

Abstract

Linear spectral unmixing (LSU) is widely used technique, in the field of remote sensing (RS), for the accurate estimation of number of endmembers, their spectral signatures and fractional abundances. Large data size, poor spatial resolution, not availability of pure endmember signatures in dataset, mixing of materials at various scales and variability in spectral signature makes linear spectral unmixing as a challenging and inverse-ill posed task. Mainly there are three basic approaches to manage the linear spectral unmixing problem: geometrical, statistical and sparse regression. First two approaches are kind of blind source separation (BSS). Third approach assumes the availability of some standard publicly available spectral libraries, which contains signatures of many materials measured on the earth surface using advance spectra radiometer. The problem of linear spectral unmixing, in semi supervised manner, is simplified to finding the optimal subset of spectral signatures from the library known in advance. In this paper, the concept of soft thresholding is incorporated along with the sparse regression for automatic extraction of endmember signatures and their fractional abundances. Our simulation results conducted for both standard publicly available synthetic fractal dataset and real hyperspectral dataset, like cuprite image, shows procedural improvement in spectral unmixing.