Systematic approach towards extracting endmember spectra from hyperspectral image using PPI and SMACC and its evaluation using spectral library

Abstract

This study focuses on systematic approach towards extraction of endmember spectra from hyperspectral image. The study demonstrates the effect of systematic preprocessing like atmospheric correction, radiometric correction (bad band and columns correction), and geometric correction on hyperspectral image. The study also focuses on the selection of the method for extracting endmember spectra depending upon the land-cover classes present in the study area. Two algorithms for extracting endmember spectra, i.e., pixel purity index (PPI) and sequential maximum angle convex cone (SMACC), have been used. To validate and perform comparative analysis of the two algorithms, spectral library has been created using field spectroradiometer and used as a reference to evaluate their performance. Visually, good results have been observed between extracted endmember spectra and reference library spectra after applying the rigorous preprocessing. To further analyze the two endmember extraction algorithms, spectral angle mapper (SAM) scores have been computed for various endmember spectral classes with respect to reference spectral library. The tabulated SAM scores for the endmembers of PPI and SMACC show that SMACC is more effective in extracting endmember spectra of vegetation classes while PPI is a more effective algorithm for roads and dry soil. It has been observed that systematic approach towards extracting endmember spectra from a hyperspectral image should consist of proper preprocessing steps, ground validation with a reference spectral library, and most importantly the proper selection of algorithm as the performance of algorithms for extracting endmember spectra depends on the land-cover classes present in the study area.