Research on Retinal Vessel Segmentation Algorithm Based on Deep Learning

Abstract

In order to solve the problems of low precision and large error in the segmentation task of retinal blood vessel by computer, this paper improved the UNet++ network structure and proposed an algorithm model DAUNet++ (Deformable attention UNet++) that can effectively extract retinal blood vessel structure. Firstly, the deformation residual module is designed to construct the encode structure to enhance the feature extraction capability of the network for target details. At the same time, the attention mechanism is used to remove redundant features in the original network decoding module group, and the feature enhancement module is designed to enhance the performance of features. In order to verify the robustness of the optimized network model, DRIVE dataset was used for experimental tests. The test results showed that the accuracy, sensitivity and specificity of the optimized network model reached 97.07%, 83.25% and 98.36%. The experimental results show that the network model designed in this paper has a good performance in retinal vessel segmentation task and has certain competitiveness compared with other existing methods.