Panax notoginseng is a precious Chinese herbal medicine, its efficacy and quality are affected by the saponin content, which is closely related to the origin. In order to accurately identify the origin of Panax notoginseng and ensure its quality, a new method combining terahertz precision spectroscopy and convolutional neural network algorithm was proposed in this study. We collected 40 samples of Panax notoginseng from four origins of Honghe Autonomous Prefecture, Kunming City, Qujing City and Wenshan Autonomous Prefecture in Yunnan Province, China, and analyzed them by terahertz spectroscopy and high performance liquid chromatography. Based on the collected spectral and chromatographic data, a convolutional neural network model was constructed and trained for the origin classification of Panax notoginseng samples. The results show that the classification accuracy of terahertz spectroscopy combined with convolutional neural network model reaches 92.5%, which is significantly better than that of high performance liquid chromatography combined with convolutional neural network model (82.5%). This discovery demonstrates the potential of terahertz spectroscopy in the composition analysis and origin identification of Chinese herbal medicines, providing a new scientific tool for rapid, non-destructive and accurate identification of Chinese herbal medicines.