Enhancing Weak Feature Detection in Convolutional Analysis of Traditional Chinese Medicine Tongue Diagnosis

Tongue diagnosis holds a significant position in traditional Chinese medicine, where it involves assessing certain illnesses within the human body by observing tongue images. Utilizing machine learning techniques to train on tongue images can enable automated analysis of tongue diagnosis, offering the potential to significantly enhance the efficiency and accuracy of medical diagnostics. In this study, we address two critical issues with existing deep convolutional models:1. The loss of weak features in low-brightness regions. 2. The fragmentation of feature extraction in the convolution process for tongue images. Through experiments conducted on 14 different convolutional models, we observed an improvement in accuracy ranging from approximately 10% to 15%. This enhancement notably elevates the accuracy of machine learning in this field.