Fundus vascular images are commonly used in clinical practice for the diagnosis and monitoring of eye diseases. The morphology and structure of blood vessels could reflect the essential features of the disease. Therefore, the segmentation of fundus vascular images is of great medical significance for the diagnosis and prevention of eye diseases. Current mainstream artificial intelligence algorithms, due to convolution and pooling operation, often neglect the extracted features of the spatial information in the images, making it difficult to segment fine blood vessels and other details. This study conducted research based on the U-net model, combining a spatial attention module to refine the spatial features. It also proposed a low compensation structure to reduce the feature loss during the feature extraction process of network, thereby improving the segmentation accuracy. Experiments were conducted on the DRIVE open dataset, and the algorithm achieved a segmentation accuracy of 96.97% and an F₁ value of 74.36%. The results demonstrate that the proposed network structure exhibits better segmentation performance and more accurate identification of fine blood vessels structures.