Rapid detection of beef oxidation indicators using hyperspectral imaging and SHAP-enhanced CNN

Beef oxidation is a critical factor affecting meat quality, shelf life, and consumer acceptance, and its rapid non-destructive assessment remains challenging under different storage conditions. In this study, hyperspectral imaging (HSI) combined with chemometric and deep-learning approaches was used to predict beef oxidation indicators under atmospheric and high-pressure storage conditions. Beef longissimus dorsi muscles obtained from eight cattle were divided into 24 physical sub-samples, and spectral/reference measurements collected across five storage time points (0, 2, 4, 6, and 8 days) generated a total of 120 sample-time observations for model development. After spectral preprocessing and feature wavelength selection, principal component regression (PCR), multiple linear regression (MLR), and one-dimensional convolutional neural network (1D-CNN) models were established and compared. The optimal CNN models were further interpreted using SHAP to identify spectrally important wavelengths associated with oxidation-related changes. Results showed that the CNN-based models generally outperformed PCR and MLR, demonstrating superior capability in capturing nonlinear relationships between spectral features and oxidation indicators. The CNN model exhibits significant advantages in fitting the nonlinear relationships between spectral features and oxidation indicators. Under different conditions, the Rp values for Atmospheric Pressure (400–1000 nm), Atmospheric Pressure (900–1700 nm), High Pressure (400–1000 nm), and High Pressure (900–1700 nm) groups reached 0.9729, 0.8780, 0.9524, and 0.9639, respectively. SHAP analysis highlighted several informative wavelength regions that were consistent with oxidation-related biochemical changes in beef. These findings indicate that the proposed HSI-CNN-SHAP framework provides a promising and interpretable strategy for rapid, non-destructive evaluation of beef oxidation under different preservation conditions.