An Unmixing Algorithm Based on a Large Library of Shortwave Infrared Spectra

The unmixing algorithm described in this paper has been motivated by a "spectral library" of pure shortwave infrared reflectance spectra that we started building in the early 1990's. The library currently consists of 493 samples, representing 60 nominally pure materials (mostly minerals, but also water, dry vegetation and several man-made materials). The algorithm, implemented in software called The Spectral Assistant (TSA), is designed to analyse quickly tens to hundreds of thousands of spectra measured from drill core or chips using CSIRO's HyLogger and HyChips instruments, and other commercial reflectance spectrometers. Individual samples typically are composed of a small number of minerals. Therefore, in order to avoid overfitting, the TSA algorithm utilises fast subset selection procedures to identify the most likely minerals in the mixture. Other novel aspects of the algorithm include the simultaneous fitting of the low frequency background with mineral identification (which provides greater model flexibility) and the combined fitting being carried out in penalised canonical variate space (which has certain optimality properties under an idealised model). The performance of the algorithm is illustrated on a few key examples. Discussion includes its wider applicability, its limitations and possible future extensions and modifications.