ATR-FTIR spectroscopy coupled with deep learning for the identification and quantitative detection of Panax...

The high medicinal value, strong market demand, and premium price of Panax notoginseng have led to frequent adulteration in Chinese retail markets, posing serious risks to consumer health and food safety. However, existing adulteration detection methods are often limited by low efficiency and high costs, restricting their applicability for large-scale quality screening of Panax notoginseng products. This study aims to develop a rapid and cost-effective approach integrating attenuated total reflection-Fourier transform infrared (ATR-FTIR) spectroscopy with deep learning for accurate identification and quantification of adulterants in Panax notoginseng main root powder (MRP), including Panax notoginseng fibrous root powder (FRP), Curcumae rhizoma powder (CRP) and rice powder (RP). Experimental results demonstrated that convolutional neural network (CNN) and Transformer outperformed traditional machine learning algorithms, achieving four-class identification accuracies of 97.77 % and 98.66 %, respectively. For regression prediction of adulterant concentrations, R² values ranged from 0.9475 to 0.9789 for CNN models and 0.9387 to 0.9848 for Transformer models with feature wavelength selection. Additionally, for predicting the content of primary active components in adulterated samples, R² values varied from 0.9445 to 0.9726 (CNN) and 0.9602 to 0.9720 (Transformer). These findings highlight the potential of ATR-FTIR spectroscopy combined with deep learning as a powerful tool for detecting adulteration in medicinal materials.