Abstract: The detection and analysis of fundus retinal blood vessels is of great significance for the diagnosis of many ophthalmological diseases. To extract retinal blood vessels more accurately and soundly, we propose a U-shaped network model combining multi-label loss and dual attention. Firstly, we introduce spatial pyramid pooling in the encoder part to provide multi-scale input, and incorporate dual attention residual block into the U-shaped network to improve the network's ability to extract feature information.Secondly, the feature similarity module is embedded at the bottom of the network to capture the long-range dependency between features, To effectively suppress the influence of noise in the fundus image and capture the multi-scale information of the blood vessels, the dual-pathway attention gates mechanism and densely connected atrous spatial pyramid pooling(DenseASPP) module are introduced in the skip connection part.Finally, the side output layer is set in the decoder part to generate a local prediction image corresponding to the level, and cooperate with the multi-label Dice loss function for training. Experiments on the DRIVE,STARE, and CHASE_DB1 datasets show that the sensitivity is 80.54%, 83.97%, and 82.40%, and the area under curve of the receiver operating characteristic is 98.07%, 98.50%, and 98.36%, respectively.