Efficient artificial intelligence algorithms for on-site discrimination of Angelica sinensis.

Angelica sinensis, a valuable traditional Chinese medicine with high economic and medicinal value, is used medicinally in parts such as the body, head, and tail. However, there are significant differences in the chemical composition and pharmacological effects of different parts of Angelica sinensis, and the classification characteristics are not obvious. Therefore, this study focused on the identification of the medicinal parts of Angelica sinensis using a benchtop and four types of miniature near-infrared (NIR) spectrometers for rapid classification. Using the classical machine learning algorithms, partial least squares discrimination analysis (PLSDA) and support vector machine (SVM), determining the optimal spectral modeling processing strategy for each instrument and constructing a classification model. Additionally, we constructed a convolutional neural network (CNN) classification model to identify the three medicinal parts of Angelica sinensis. Results showed that there was a difference between the predictive performance of miniature NIR spectrometers and benchtop instruments, with some miniature spectrometers being unable to accurately classify traditional Chinese medicinal materials. The proposed CNN framework exhibits versatility in handling various spectrometer configurations and reduces the computational burden associated with traditional parameter tuning procedures. Both Fourier-transform and miniature NIR spectrometers, supported by classical and modern chemometric classifiers, are suitable for evaluating the medicinal parts of Angelica sinensis. The CNN model structure is applicable for modeling with various spectrometers and eliminates the need for complex parameter optimization procedures.