A novel method for automatic minerals identification and their abundance estimation for material type discrimination using hyperspectral data

In hyperspectral remote sensing, the conventional endmember extraction and unmixing procedures are often complex and associated with uncertainties. In this work, we have designed an algorithm that uses Crude Low Pass Filter (CLoPF) and Pearson´s Correlation Coefficient (PCC) to identify the endmember spectra from spectral library. Subsequently, a Non-Negativity Fully Constrained Least Square (NNFCLS) optimization approach was used to determine the fractional abundances of identified end-members. The efficacy of adopted procedure was estimated by Normalized Root Mean Squared Deviation (NRMSD), Spectral Angular Mapper (SAM), computation timing and appropriateness of identified candidates. It is observed that this procedure can be effectively used to resolve the mix-pixel spectra into library constituents and its fractional abundances.