Coupling subband decomposition and independent component regression for quantitative NIR spectroscopy

In this paper, a new method is proposed for coupling digital bandpass filtering with independent component regression (ICR) to improve the quality of the raw absorbance spectra in quantitative NIR spectroscopy. The proposed model, referred to as SDICR, is based on a subband decomposition independent component analysis (SDICA) model coupled with ICR regression. The SDICA is used to select the optimal parameters of the digital bandpass filter and to determine the most independent subcomponents of the original sources, so the standard ICA methods can be used. The efficiency of the proposed model is validated using mixtures composed of glucose, urea and triacetin in a phosphate buffer solution in a non-controlled environment. The proposed model decreases the standard error of prediction (SEP) from 29.1 mg/dL for ICR to only 18.6547 mg/dL using 10 subbands.