Near-infrared spectroscopy combined with machine learning for rapid identification of Atractylodis rhizoma decoction pieces

As a naturally occurring plant source of essential oil, Atractylodis rhizoma (AR) is of significant economic and therapeutic importance. In modern medical use, it is preferable to process the material into flakes of dried AR. Fake products often pass off as authentic AR, and products from non-primary production areas pass off as pri-mary production areas to pursue high profits. In this study, near-infrared spectroscopy (NIRS) was developed to better identify the authenticity, botanical sources, and geographical origins of AR. The impacts of pretreatment, selection of characteristic wavenumbers, and parameter optimization on model performance were compared and analyzed. Five different types of machine learning methods were used. The results showed that the extreme learning machine (ELM) had the best effect in identifying the authenticity of AR, while the back propagation neural network (BPNN) had advantages in determining the sources of plants. The support vector classification (SVC) had great potential to pinpoint the geographical origins of Atractylodes lancea (Thunb.) DC. and Atrac-tylodes chinensis (DC.) Koidz. The feasibility of direct spectral acquisition without crushing the sample was also demonstrated. Therefore, NIRS combined with machine learning is a fast, effective, and feasible method to identify the authenticity, botanical sources, and geographical origins of AR.