Hyperspectral imaging feasibility in the prediction of melanoidin parameters in whey protein-β-glucan conjugates

The Maillard reaction is a non-enzymatic browning process between reducing sugars and protein amino groups, which plays a crucial role in food chemistry by enhancing protein functionality and generating antioxidant and antimicrobial compounds. However, under unfavorable conditions, it may produce toxic or carcinogenic substances, emphasizing the need for careful monitoring. In this study, the feasibility of hyperspectral imaging was evaluated in monitoring melanoidin formation during the Maillard reaction. Predictive models based on artificial neural networks accurately estimated key reaction parameters, including advanced Maillard products, antioxidant activity, and degree of glycosylation, with accuracies of 94.27, 95.88, and 87.06%, respectively. Partial least squares regression also showed strong predictive performance (R 2 > 96%), while support vector regression was less accurate (R 2 = 75-78%). These results demonstrate high possibility of hyperspectral imaging combined with machine learning as a rapid non-destructive tool for monitoring Maillard reaction parameters, with potential applications in optimizing thermal processing and ensuring food quality.