Counterfeit Citri Reticulatae Pericarpium Identification: Multi-Class Detection Using Vis/NIR Spectroscopy and...

The expanding market demand and high commercial value of Citri Reticulatae Pericarpium (CRP) have led to frequent counterfeiting. Rapid and non-destructive identification of counterfeit CRP is therefore essential. This study proposes a novel detection model that integrates visible/near-infrared (Vis/NIR) spectroscopy with deep learning algorithms. Authentic CRP and four types of counterfeit CRP samples were collected, and their multispectral Vis/NIR data were acquired. Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) data augmentation enhanced model generalization, further improving classification robustness. After spectral preprocessing, deep learning models were applied to classify and detect different CRP types, aiming to identify the optimal network for counterfeit discrimination. The experimental results demonstrated that the Inception-ResNet model achieved the test set accuracy of 96.56% in identifying authentic CRP from four counterfeit categories, outperforming conventional machine learning approaches. This study provides a precise and efficient model for non-destructive authentication and adulteration analysis of CRP.