New effective spectral matching measures for hyperspectral data analysis
The successful implementation of Spectral Matching Measures (SMMs) often plays a crucial role in material discrimination and classification using hyperspectral dataset. The commonly exploited SMMs, such as Spectral Angle Mapper (SAM), Spectral Information Divergence (SID), and their hybrid, i.e., SIDSAMtan, show limited discrimination ability while discriminating spectrally similar materials. This study presents three new effective SMMs named Dice Spectral Similarity Coefficient (DSSC), Kumar–Johnson Spectral Similarity Coefficient (KJSSC), and a hybrid of DSSC and KJSSC, i.e., KJDSSCtan, for accurate discrimination of spectrally similar materials. A wide range of hyperspectral datasets of minerals and vegetation acquired under laboratory and real atmospheric conditions were used to compare and evaluate the performance of newly proposed and existing SMMs using Relative Spectral Discrimination Power (RSDPW) statistics. We also assessed the discrimination ability of the proposed and existing SMMs using spectra of selected minerals and vegetation species with an added component of random noise and linearly synthesized mixed spectra. An in-depth comparison and evaluation of different SMMs demonstrated that the discrimination power of the proposed SMMs is significantly higher than existing SMMs. The proposed SMMs also outperform existing SMMs when discriminating noisy and linearly synthesized mixed counterparts. The KJSSC and DSSC show similar efficacy in discriminating spectra of minerals and vegetation, whereas their hybrid measure, i.e., KJDSSCtan shows significantly higher spectral discrimination ability. Therefore, the newly proposed hybrid measure, i.e., KJDSSCtan is recommended over existing SMMs for successful material discrimination and classification using hyperspectral data.