Research on Image Segmentation Based on Deep Learning Algorithm using Fusion Attention Mechanism

Skin lesion segmentation is a prerequisite for the intelligent and automated development of skin disease diagnosis, and its segmentation effect directly affects the quality of subsequent skin disease identification. Therefore, achieving accurate segmentation of skin lesion areas is of great significance to the intelligent diagnosis of skin diseases. In view of the problems of different sizes of lesion areas, low contrast with surrounding skin, and blurred edges, this paper proposes a skin lesion segmentation method that integrates attention mechanism. This method consists of ParNet block (PNA) and FaPN segmentation network. The PNA module is placed after each convolution layer in the FaPN network, which can not only enhance the model feature extraction ability, but also alleviate the loss of detail information caused by continuous convolution, which is very helpful for accurate lesion segmentation. To verify the performance of the improved model AFaPN, this paper compares it with other advanced models on the public dataset ISIC2016. The results show that the algorithm proposed in this paper is better than other popular algorithms.