The traditional processing method of “sweating processing” is one of the key origin factors contributing to the quality formation of authentic Chinese herbal medicines, which is closely related to unique quality and clinical efficacy. Magnolia officinalis cortex (MOC) is the dried bark of Magnolia officinalis Rehd. et Wils to treat gastrointestinal diseases. The 2025 edition of Chinese Pharmacopoeia stipulates that MOC must undergo sweating processing until the inner surface turns purplish-brown or brown. However, quality assessment methods for MOC remain constrained by subjective biases or complex instruments, thereby impacting reproducibility and cost-effectiveness. Therefore, this study proposed a method based on digital images and deep learning to analyze color characteristics and monitor dynamic changes during the sweating processing of MOC. In this study, one-class classification was used to correctly authenticate and recognize the category of MOC samples at different stages of sweating. Pearson correlation analysis examined obvious correlations between color frequency histograms and indicator components of MOC samples. Furthermore, the study established regression prediction models including magnolol and honokiol based on partial least squares regression (PLSR) and successive projections algorithm-PLSR (SPA-PLSR). Importantly, this study confirmed the great potential of residual neural network-50 (ResNet-50) based on convolutional neural network (CNN) in understanding the quality formation during sweating processing of MOC by utilizing more stable SPA-PLSR ( R 2 Pred>0.75, RPD>2). In conclusion, the method provided new ideas and approaches to evaluate quality control of sweating processing for MOC and offered a reference for monitoring the processing of other Chinese herbal medicines.
Zhang et al. (Fri,) studied this question.