Detection of Aflatoxin B1 in Maize Silage Based on Hyperspectral Imaging Technology

Aflatoxin B1 (AFB1) is widely present in maize silage feed and poses strong toxicity, seriously threatening livestock production and food safety. To achieve efficient and accurate non-destructive detection of AFB1, this study proposes a quantitative prediction method based on hyperspectral imaging technology. Using the full-spectrum bands after SG, SNV, MSC, FD, SD, and SNV + FD, MSC + FD, SNV + SD, MSC + SD preprocessing, the characteristic wavelengths selected by CARS, BOSS, and RF feature selection methods, and the augmented bands generated by Mixup data augmentation as input features, three models were developed for AFB1 content prediction: a linear WPLSR_SD_Mixup_QPE model, a nonlinear SVR_SD_Mixup_PCA model, and a deep learning CNN_SD_Mixup_WMSE_SA model. The results demonstrated that SD preprocessing was the most suitable for AFB1 detection in maize silage, and the Mixup data augmentation method effectively improved model performance. Among the models, SVR_SD_Mixup_PCA achieved the best performance, with an R p 2 of 0.9458, RMSEP of 3.1259 μg/kg, and RPD of 4.2969, indicating high prediction accuracy and generalization capability. This study fills the gap of hyperspectral image technology fused with artificial intelligence algorithm in the application of quantitative detection of AFB1 content in maize silage and provides a new technical method and theoretical basis for nondestructive testing of corn silage feed.