Double-Coding Skin Disease Segmentation Based on Attention Mechanism

Abstract

Aiming at the problems of complex structure and lack of feature information of skin lesions, an image segmentation method based on attention mechanism and double coding network is proposed. Firstly, the dual-coded branch network is used to extract image feature information to improve the ability of network information capture. Secondly, the dual attention module is used to encode the global context information, which suppresses irrelevant features and highlights relevant features. Finally, in the coding part, depth separable convolution and cooperative attention are used to reconstruct the image. Experimental results show that on ISIC2018 data set, the accuracy, Dice similarity coefficient and Jaccard index of this method reach 96.59%, 92.98% and 82.65%, respectively. Compared with other networks, the accuracy and boundary segmentation effect are obviously improved.