Anthocyanin-based colorimetric sensor for aflatoxin B1 (AFB1) detection in wheat: Performance optimization via near-infrared (NIR) spectral characterization and structurally improved recurrent neural network (RNN)

The highly toxic aflatoxin B1 (AFB1) in wheat poses a severe threat to food security and public health. In this study, a rapid and eco-friendly detection method for AFB1 residues in wheat was developed by integrating a natural anthocyanin-based colorimetric sensor array (CSA) with near-infrared spectroscopy (NIRS). A Deep-RNN regression model was specifically constructed for the high-dimensional spectral sequence features of the natural anthocyanin-based CSA-NIRS coupled system, to achieve precise extraction of the core response characteristics correlated with AFB1. The results demonstrated that NIRS effectively expanded the feature dimension of CSA; models built on NIR spectral features achieved significantly higher recognition accuracy than traditional RGB-based models, and the optimized Deep-RNN model reached a quantitative prediction accuracy of over 0.90 for AFB1. This coupled system breaks through the limitations of conventional detection technologies, and provides a novel and reliable solution for on-site rapid screening of mycotoxins in cereal grains.