Skin Disease Detection Based on Improved Conformer

Skin diseases are a global health problem, and the accuracy of skin disease diagnosis is undoubtedly considered an important factor in the treatment and recovery of patients. However, the original Conformer model has limitations in extracting multi-scale features from images. To address this problem, we introduce the wavelet transform convolution (WTConv2d) and Squeeze-and-Excitation Attention (SEAttention) modules. The former enhances the model's ability to extract multi-scale features from images, and the latter improves the expression of key features by adaptively adjusting channel weights. Experiments on the DermNet dermatosis dataset show that the improved WHC model outperforms the traditional Conformer model in terms of accuracy, recall and F1 score after the same number of training iterations. This paper provides new ideas and methods for the automated classification of dermatosis.