AI Recognition and TCM Diagnosis System of Tongue Images Based on Deep Learning

Tongue diagnosis, a core component of "inspection" in Traditional Chinese Medicine TCM, is valuable for disease screening and constitution identification. Traditional tongue diagnosis relies on physicians' subjective experience, leading to poor standardization, low efficiency, and limited regional popularization. To solve these problems, this paper designs an AI intelligent recognition and TCM diagnosis system of tongue images based on deep learning. The system uses public TCM Tongue Dataset and self-built clinical datasets, with input quality optimized via preprocessing including cropping, normalization and data augmentation. A lightweight tongue feature extraction model is constructed by improving ResNet with attention mechanism, reducing redundant parameters. Model training is optimized through transfer learning, batch normalization, and learning rate decay. Comparative experiments show the improved model achieves 93.7% tongue feature recognition accuracy and 89.2% TCM constitution identification accuracy on the test set, 14.3 and 12.8 percentage points higher than traditional CNN. This system provides a feasible solution for standardized and intelligent TCM tongue diagnosis.