The invention relates to a retinal vessel segmentation method and device based on a multi-scale attention network, and a storage medium. The method comprises the steps: 1, obtaining a data set; 2, preprocessing the data set, and sequentially carrying out gray processing, adaptive histogram equalization and gamma correction processing on the data set; 3, constructing a retinal vessel segmentation model; 4, training a retinal vessel segmentation model; 5, performing retinal blood vessel segmentation: preprocessing a to-be-segmented fundus retina image and then inputting the to-be-segmented fundus retina image into the trained retinal vessel segmentation model to obtain a segmented output image; 6, splicing the segmented output images to obtain an original image, and taking an average pixel value of overlapped parts to obtain a retinal blood vessel segmentation result. According to the method, the feature maps of all the layers are fused, and better feature representation is obtained. An attention module is added, which pays attention to those areas that contribute more to the result to obtain a more accurate result.