Age identification of mountain-cultivated ginseng based on terahertz precision spectroscopy and convolutional neural network

Mountain-cultivated ginseng (MCG) accumulates age-dependent ginsenosides through ecological habitat simulation, while the age determination method is lacking and requires urgent resolution. This research pioneers a new non-destructive and precise age identification method for MCG by integrating terahertz (THz) spectroscopy technology with convolutional neural networks (CNN). Through comparative analysis of 54 MCG samples (5–20 years), the THz-CNN method demonstrated superior age classification accuracy (92.59%) versus the HPLC-CNN method (77.78%). The underlying mechanism lies in the non-linear relationship between the accumulation pattern of ginsenosides and the plant age. The terahertz spectrum includes comprehensive ginsenoside characteristics, including fingerprint spectral features, multiannual evolution patterns, and their relative correlations, which can be used for multi-dimensional network analysis. In HPLC analyses, inter-batch variability in the silanization degree of the chromatographic column induces deviations in ginsenoside peak areas and retention times, thereby amplifying errors in CNN-based algorithms for chronological identification. Furthermore, by implementing a feature-augmented input framework that combines enhanced spectral features with raw data, the model's classification accuracy was further elevated to 96.30% by intensifying feature learning capacity. The THz-CNN synergy establishes a novel paradigm for phytochemical maturation analysis, simultaneously advancing non-destructive detection technology and intelligent processing of spectral big data in herbal medicine research.‌.