Fusion of multimode near-infrared detections using multiblock data analysis for prediction of mixed sugar content

Abstract

This study combined NIR spectral data from different detection modes to enhance the quantification performance of the instrument owing to the benefits of various detection modes in the near-infrared (NIR) spectroscopic technique. Mixed sugar samples, used for the demonstration, composed of glucose, fructose, maltose, and sucrose, were composed based on a mixture design. NIR spectra were recorded using portable Vis-NIR (360.9–1078.3 nm) and benchtop NIR (400–2500 nm) spectrometers using both transmittance and transflectance detection modes. A multiblock principal component analysis (MB-PCA) was applied to exploratorily analyze the multiblock data. Multiblock regression models, including concatenated partial least squares (C-PLS), serial-PLS (S-PLS), and multiblock-PLS (MB-PLS), were employed to quantify the sugar sample concentrations. MB-PCA could easily distinguish between the sugars, accounting for 98.51 % and 100.00 % for the portable Vis-NIR and benchtop NIR spectra on the first two principal components (PCs). The spectral fusion using the multiblock data analysis could improve the predictive performance. The best results were based on the use of MB-PLS, with  values of R² of 1.00, Q² of 0.96, and 1.00; root mean square error of calibration (RMSEC) of 0.02 and 0.05; root mean square error of cross validation (RMSECV) of 0.22, and 0.06; ratio of prediction to deviation (RPD) of 5.03 and 18.43; and relative standard deviation (RSD) values of 5.63 and 1.54 using the portable Vis-NIR and benchtop NIR spectral data, respectively.