Medical image segmentation can assist doctors in medical diagnosis. In practical applications, medical image acquisition devices are different, target sizes are different, shapes are irregular, and there are various artifacts in the images, resulting in some problems in medical image segmentation. U-Net is currently one of the most widely used networks in the field of medical image segmentation, where convolutional operations have local characteristics when extracting features, which in turn affects the performance of medical image segmentation. This article proposes a cup and disc segmentation method using dual branches and Transformer, and designs three modules to improve the expression ability of features. Among them, the scale perception feature fusion module is used to collect semantic and positional information of the optic disc and cup from high-level features; The recognition module is used to capture information about the optic disc and cup hidden in low-level features; The graph convolutional domain feature fusion module integrates high-level semantic features and low-level features. By using Drishti-GS1 The effectiveness of the algorithm was verified through experiments on the RIM-ONE-r3 and REFUGE datasets.