Portable NIR spectroscopy for on-site origin discrimination and alkaloid quantification in Stephania tetrandra roots

Abstract

Given the critical importance of field harvesting, market supervision, and quality control during plants processing, the demand for rapid and reliable screening tools has become increasingly urgent to ensure quality and safety throughout the entire medicinal plants supply chain. In this study, we developed a novel portable near-infrared (NIR) spectroscopy-based system integrated with multiple chemometric techniques for the rapid and non-destructive discrimination of geographical origin and quantitative prediction of alkaloid components in medicinal plant roots. Using Stephania tetrandra as a case study, we validated the effectiveness and practicality of this in-field testing approach. Our findings demonstrated that the LightGBM classifier, using an S–G + 2nd der. preprocessing method, achieved perfect accuracy (100%) and an F1 score of 1.0 for geographical origin discrimination. For alkaloidal components, Partial Least Squares Regression (PLSR), Back Propagation Neural Network (BPNN), and Random Forest (RF) models were employed to predict the concentrations of magnoflorine, fangchinoline and tetrandrine. The optimal models identified for each alkaloid component were CARS-MSC-RF for magnoflorine, Full-MSC-BPNN for fangchinoline, and Full-SNV + 2nd der.-BPNN for tetrandrine. To further validate model performance, a paired T-test was conducted between the predicted and actual values, yielding p-values greater than 0.05, indicating no significant difference. These results underline the potential of portable NIR spectroscopy as an efficient and reliable tool for geographical origin discrimination and alkaloidal components quantification, crucial for quality control in medicinal plants industry.