The morphological changes of retinal vessels,such as bifurcation angle and dilation degree,can provide the basis for the diagnosis of fundus diseases.Using deep l earning technology to evaluate the degree of retinopathy has become the focus of current research. This paper proposes a retinal vessel segmentation method based on multi-path in put and multi-scale feature fusion to solve the problem of retinal vessel segmentation. We use multi-path input and multi-scale feature fusion to improve the U-Net model,s o that our network can effectively solve the problem of poor segmentation effect of fundus retinal image.The experimental results show that the sensitivity of the proposed algori thm is 0.8148and 0.8137on DRIVE and CHASE_DB1datasets,the specificity is 0.9848and 0.9866,and the segmentation ac curacyis 0.9699and 0.9757.