Multi-spectra combined with Bayesian optimized machine learning algorithms for rapid and non-destructive detection of adulterated functional food Panax notoginseng powder

Abstract

Panax notoginseng powder (PNP) is a widely consumed functional food and has shown promise in cardiovascular protection. However, its high price makes it often a target for economic adulteration. This study aims to build a rapid evaluation method for common adulterants of PNP using near-infrared (NIR) and/or visible (VIS) combined with Bayesian optimized machine learning algorithms. The results showed that the machine learning algorithms combined with a multi-spectra fusion strategy exhibited excellent performance in distinguishing different types of adulterated PNP, especially the deep learning algorithms (ANN and LSTM) with 100 % test accuracy. Furthermore, machine learning algorithms coupled with VIS spectra had an obvious advantage in predicting the proportion of PNP adulteration, with all algorithms having a prediction accuracy R2p of over 0.99. Overall, multi-spectra combined with Bayesian optimized machine learning algorithms enables a rapid and accurate evaluation of PNP adulterants, which can also apply to other foods.