In the process of retinal vessel segmentation, issues such as incomplete vessel segmentation or the presence of micro-vessel fractures arise due to the complex structure of small vessels and blurred boundaries. To address these challenges, we propose an edge feature enhancement U-Net (EFE U-Net) based on edge feature enhancement. By introducing an edge feature enhancement module to replace the 2D convolution in the U-Net model, the receptive field of U-Net is expanded while enhancing the edge information in retinal vessel images. The edge feature enhancement module consists of an edge information extraction module, an edge information fusion module, and a channel and spatial attention module. Experiments on the DRIVE, CHASE_DB1, and STARE datasets demonstrate that EFE U-Net achieves sensitivities of 82.64%, 84.41% and 84.45%, the F1 scores are 83.09%, 84.49% and 82.84%. The two metrics show improvemens over the baseline model. The experimental results indicate that EFE U-Net can effectively segment retinal vessels.
Publications
- Article type
- Year
Year
Open Access
Issue
Journal of Northwest University (Natural Science Edition) 2026, 56(5): 1125-1135
Published: 25 October 2026
Downloads:0
Total 1
京公网安备11010802044758号