Mould deterioration monitoring of Citri Reticulatae Pericarpium using Vis/NIR imaging and an improved Inception ResNet.

Introduction: Mould deterioration is a critical quality risk for Citri Reticulatae Pericarpium (CRP) during storage, reducing its commercial value and potentially compromising the safety of dried food and food-medicine homologous products. Conventional visual inspection is subjective and may fail to identify early deterioration. Methods: Visible and near-infrared (Vis/NIR) multispectral imaging was combined with deep learning to develop a rapid and non-destructive method for monitoring mould deterioration in CRP. Spectral images were acquired using a self-developed Vis/NIR imaging system equipped with 26 LED centre wavelengths. An improved Inception ResNet integrating multi-scale convolution, residual learning, and attention-based feature refinement was developed. Its performance was compared with support vector machine, XGBoost, multilayer perceptron, 1D CNN, ResNet, and the original Inception ResNet. Results: The improved Inception ResNet combined with Savitzky-Golay preprocessing achieved the best performance, with five-fold cross-validation accuracy, precision, recall, and F1 score of 96.13 ± 0.82%, 96.21 ± 0.79%, 96.13 ± 0.82%, and 96.15 ± 0.80%, respectively. On the independent external validation set, the corresponding values were 94.27 ± 0.89%, 94.39 ± 0.85%, 94.27 ± 0.89%, and 94.22 ± 0.87%. Spectral analysis showed distinct deterioration-related responses in the visible and near-infrared regions, associated with surface colour variation, moisture redistribution, and internal quality degradation. Discussion: These findings demonstrate that Vis/NIR multispectral imaging coupled with the improved Inception ResNet provides an effective and interpretable approach for rapid, non-destructive CRP quality screening and mould-deterioration monitoring.