Comparison of near infrared and Raman hyperspectral unmixing performances for chemical identification of pharmaceutical tablets

Methods for fast and accurate identification and quantification of the composition of pharmaceutical mixtures are important in many scientific and industrial applications. When this goal is approached via hyperspectral data analysis, the problem becomes one of hyperspectral unmixing, where the goal is to identify the pure materials (also called endmembers) present in a mixture, as well as their relative abundances. In this paper, we assess the performance of a state-of-the-art method (named SISAL-simplex identification via split augmented Lagrangian) recently proposed for remote sensing applications, in the context of hyperspectral unmixing of NIR and Raman spectra for pharmaceutical applications. The NIR and Raman data are complementary in nature and have different spatial resolutions. The central goal of this study is to show that these two different spectroscopic techniques allow the identification of different endmembers, and also that the higher spatial resolution of Raman data allows more accurate predictions of the endmembers spectral signatures.