Advanced similarity metric and its application to the band selection issue of multiple spectra identification

A multiple spectra oriented similarity metric, referred to as n-dimensional solid spectral angle (NSSA), is addressed in this study. NSSA extends the traditional spectral angle metric (SAM)-calculating two spectra's angle-to the calculation of the high-dimensional solid angle jointly constituted by a set of spectra with any number of bands. Some significant inherent properties of NSSA are also discussed. Furthermore, as a merit of NSSA, an NSSA-based band add-on (BAO-NSSA) band selection method is derived, which displays an advantage in capturing spectra absorption features-particularly for similar classes-by operating a set of spectra instead of distinct band variables. Finally, an annular architecture in contrast with hierarchical architecture is presented by embedding the idea of BAO-NSSA into the issue of multiple class identification. Comprehensive analyses were conducted on three real hyperspectral data sets with similar and distinct classes. The proposed approach is shown to be effective for eliminating redundant bands and improving the accuracy of spectra identification.