Retinal Vessel Segmentation Based on Dual Channel and Transformer
A local and global dual branch segmentation model to address issues such as breakage and insufficient segmentation in retinal fundus blood vessel segmentation results is proposed.The local branch first performs average segmentation on the image,extracts local feature information from the segmented image,while the global branch extracts global information.Feature Fusion Module(FFM)connects the two branches to make the network more closely connected and share global and local branch information.In this research,a Transformer module with 6 layers is added at the bottom layer of the local branch to efficiently extract and utilize abstract notifications at the bottom layer,thereby reducing the loss of feature information and improving segmentation accuracy of the network.The accuracy of the proposed model for public datasets DRIVE and STARE are respectively 98.39％and 98.76％,representing a significant improvement in segmentation accuracy compared to traditional models.
万方数据
维普
