Bespoke Real‐Time Multispectral Imaging for Apple Browning Evaluation Using Physics‐Informed Neural Networks

ABSTRACT

Food materials often exhibit time‐dependent optical changes during quality degradation, such as enzymatic browning in fresh‐cut apple slices. Real‐time optical monitoring of such processes requires high accuracy and low data‐acquisition overhead. In this study, we present a proof‐of‐concept that combines a multispectral imaging system with physics‐informed neural network modeling for continuous quantitative assessment and prediction of enzymatic browning in apples. The proposed system incorporates a compact 16‐channel snapshot multispectral camera to facilitate stable and accurate colorimetric estimation during the browning progression. Under controlled laboratory conditions and a limited sample set, the proposed Weibull‐constrained physics‐informed neural network improved temporal extrapolation compared with both the conventional Weibull model and a data‐driven neural network. These results demonstrate the potential of a multispectral imaging system coupled with a physics‐informed neural network for quantitative monitoring and prediction of early‐stage browning progression.