Abstract This study investigated the feasibility of non-destructive acrylamide quantification in whole-grain bread using combined RGB and Thermal (RGB-T) imaging techniques after the toasting process. Bread samples were toasted at different temperatures and then photographed with an RGB-T camera at successive times from 0 min to 30 min after toasting. Image processing algorithms and machine learning models, including multiple linear regression, partial least squares regression, and random forest regression, were applied to correlate color and temperature features with acrylamide concentrations. Toasting parameters exhibited a strong correlation with acrylamide concentrations, whose values ranged from 179.9 to 1332.4 µg kg − 1 . Among the models tested, the best performance was achieved using a Principal Component Regression algorithm with an R² of 0.83 and a MAPE of 9.99%, indicating high predictive accuracy based on color and thermal image features. While color analysis alone provided limited explanatory accuracy, incorporating thermal features significantly improved model performance, indicating that bread cooling curves add significant prediction power to the analysis. The results highlight the potential of contactless RGB-T imaging as a non-invasive, real-time approach for acrylamide concentration estimation in baked goods. This method offers practical applications in food quality control and safety monitoring, paving the way for the development of automated, real-time contaminant detection systems in food processing industries.
Sepehr et al. (Wed,) studied this question.