Developing a quantitative adulteration discrimination model for forest-grown Panax notoginseng using near-infrared spectroscopy with a dual-branch network.

The ecological cultivation of herbs is gaining popularity due to its superior quality and greater economic value, with increasing applications in functional foods. However, profit-driven motives have led to the mislabeling of conventionally grown herbs as ecologically cultivated. Species consistency presents new challenges in authenticity identification, especially in detecting adulteration. In this study, near-infrared spectroscopy was used to analyze samples of Panax notoginseng subjected to simulated adulteration under two cultivation models. A dual-branch network architecture was developed to extract multi-scale features from both spectral sequences and time-frequency images. The results revealed R2 values of 0.991 and 0.987 for the training and testing sets, with RMSEs of 0.225 and 0.347, respectively. Furthermore, external validation showed R2 values of 0.9735 and an RMSE of 0.638, demonstrating superior performance in terms of accuracy and robustness for quantitative analysis. This research offers critical scientific evidence for the quality control and standardization of P. notoginseng, and provides new approaches for the quantitative analysis of products under different cultivation models.