Research On Retinal Blood Vessel Segmentation Algorithm Based On Improved U-Ne

Retinal vessel image is important biological information.In the field of social security,the high uniqueness of retinal vessel structure makes fundus images available for personal identification;in the medical field,some ophthalmic and cardiovascular diseases can be reflected in the structure and morphological changes of retinal vessels,so fundus images can also be used for early diagnosis of diseases.The prerequisite for retinal vessels to play a role in the above fields is to segment retinal vessels.Traditional retinal image segmentation methods rely on manual design,which not only consumes a lot of human and material resources but also has poor segmentation accuracy.With the rapid development of deep learning,end-to-end segmentation methods based on deep learning have become a hot topic in current research and have made some progress.However,retinal vessel segmentation is still a challenging task.On the one hand,retinal vessels themselves have complex topological structures,and there are large differences in size and shape between arteries and veins;on the other hand,fundus images are subject to external interference during imaging,resulting in problems such as blurred images edges and low color contrast.Faced with these challenges,some segmentation algorithms have designed deep and complex network structures to improve segmentation performance.Although this improves segmentation accuracy to a certain extent,it also results in high model complexity,requiring a large number of hardware resources and carrying the risk of overfitting.Therefore,how to balance model size and segmentation quality by reducing network complexity while ensuring segmentation quality is a direction worth studying.In order to solve the above problems,the research content of this article mainly includes the following aspects:(1)In response to the problem that the low contrast and excessive interference areas in fundus images make it difficult to segment vascular contours,this paper enhances the contrast between vascular pixels and background pixels by analyzing the histogram characteristics of some retinal vascular images and adopting a contrast-limited adaptive histogram equalization algorithm to help the network recognize vascular edges.(2)A lightweight segmentation algorithm based on improved convolutional blocks and spatial group enhancement(IS-LNFN)is proposed.To reduce model complexity,this paper uses structural pruning to compress the original U-Net,significantly reducing the network parameter volume while ensuring the segmentation model’s effectiveness.To speed up the convergence speed of the network model and reduce the risk of gradient disappearance,an improved convolutional block that combines residual learning and PReLU activation function is designed to replace the original U-Net’s convolutional block,reducing the loss of original feature information by adding information flow.In addition,a spatial group enhancement module is introduced into the network to enhance vascular features,suppress irrelevant features,and improve segmentation performance by constructing the relationship between global and local features.To further improve segmentation performance,the improved U-Net is cascaded using a dual-path connection,and the vascular probability map is learned twice to generate more refined segmentation images.(3)A segmentation algorithm based on parallel attention modules and multi-scale information aggregation modules(PMIS-LNFN)is proposed.Based on the previous algorithm,this work designs a network that pays more attention to spatial and channel domain information and enhances the capture of multi-scale information through parallel attention modules and multi-scale information aggregation modules to further improve segmentation performance.