Near-infrared spectroscopy-based quantification of sunflower oil and pea protein isolate in dense mixtures for novel plant-based products

Abstract

Techniques to quantify oil and protein in plant-based products are laborious and environmentally harmful. This study explores the potential of near-infrared (NIR) spectroscopy as an alternative method for rapid and non-destructive quantification of oil and protein in mixtures with known content of sunflower oil and pea protein isolate (PPI). Accurate calibrations with partial least square regression (PLSR) were possible for sunflower oil (Root Mean Standard Error of the Test (RMSET) set, 0.33%; coefficient of determination of the test set (R2t), 0.99) and PPI content (RMSET, 1.24%; R2t, 0.99). Prediction of an extrapolated validation set was less accurate for sunflower oil (Root Mean Standard Error of the Validation (RMSEV) set, 2.65%; coefficient of determination of the validation set (R2v), 0.70), but was successful for PPI content (RMSEV, 2.43%; R2v, 0.93). Wavelength range reduction for CH/NH bonds improved the predictions of the PLSR models for oil but not for PPI. Wavelength selection using the Covariance Selection (CovSel) algorithm reduced the number of wavelengths from 2151 to 3 without loss of prediction accuracy within the calibration range. Wavelength range reduction gave the best model for oil; wavelength selection was best for PPI. NIR spectroscopy is a promising tool to determine oil and protein in plant-based foods rapidly and with sufficient accuracy.