A Study on Skin Lesions Classification Based on Improved EfficientN etB0 Network

Skin cancer is the most common type of cancer, and early detection and treatment of skin lesions can greatly improve the cure rate of skin cancer. However, existing skin disease assisted classification networks have some problems such as large number of parameters and complex operations, which are not convenient for deployment on mobile devices. This paper proposes a lightweight convolution neural network method for skin lesion classification. The method replaces the standard convolution module in the first stage of the network based on the lightweight EfficientN etBO with the Ghost module, and adds the CBAM module between the 1 *1 convolution layer and the pooling layer in the ninth stage, to form an improved EfficientNetBO network model. The accuracy of the improved EfficientN etBO on the ISIC2018 skin disease dataset HAMI0000 is 89.38%, the Macro_P, Macro_R, Fl-Score is 0.88, 0.86, 0.87 respectively. The number of parameters is only 5.19M, while the single inference time is 71ms. All of these indicators make the model suitable for deploying skin disease classification models on mobile devices to assist clinical diagnosis.