Research on Image Recognition Technology of Traditional Chinese Medicine Based on Deep Transfer Learning

A data set containing 25,083 images of 80 kinds of TCM is constructed in this paper to solve the complications of identifying various traditional Chinese medicines quickly and accurately. Combined with the traditional Convolutional Neural Network, PyTorch and Keras frameworks are employed to manually build a DenseNet model for the recognition of Chinese herbal pieces. Besides, the method of dropout is adopted to solve the problem of overfitting. The highest recognition rate of the traditional DenseNet-based model can reach 96.78%. Therefore, the network model such as densitynet201 is selected for pre-training, and an adaptive average pool layer is added to amplify the receiving field of the kernel, thereby further improving the recognition accuracy and convergence speed of the model, and combining the idea of transfer learning. Finally, the classification output is performed, and the highest recognition rate of the transfer learning-based model can reach 97.34%, which is helpful to further improve the accurate recognition rate of the model, effectively solve the intelligent recognition of Chinese herbal pieces, and lay the theoretical foundation for the intelligentizing of the traditional Chinese medicine industry.