Assessment of macronutrients and alpha-galactosides of texturized vegetable proteins by near infrared hyperspectral imaging

Near infrared (950−1654 nm) hyperspectral imaging (NIR-HSI) was applied for the first time for the prediction of proximate composition (proteins, carbohydrates, lipids, ashes) and alpha-galactosides (verbascose, stachyose, raffinose) content of texturized vegetable proteins (TVP) obtained from different raw materials (four formulations) intended for plant-based meat analogues (PBMA) production. After exploration by Principal Component Analysis, the dataset was split into calibration and test sets and analyzed by Partial Least Squares Regression. In calibration and cross-validation, the R2 was between 0.92 and 0.98, with low error values. The figures of merit of the prediction confirm those results and the good performance of the models. Pixel-by-pixel prediction allowed the tracking of the non-uniform distribution of chemical components. Overall, NIR-HSI showed potentiality to be applied as a tool for rapid, accurate, and non-destructive quality control of TVP, which is fundamental as they strongly affect the nutritional and textural properties of PBMA.