Abstract: A retina vessel segmentation algorithm that integrates CLTransformer with cross-scale attention is proposed in this paper to address various issues encountered by the existing methods, such as the nocorrect segmentation of the optic disc, the blurring of main vessel texture, and the breaking of microvascular branches. Firstly, a lightweight residual encoder-decoder module is developed for encoding and decoding, which enables the coarse-grained extraction of vessel texture features. Secondly, a multi-scale feature selection module is applied at the point of connection between the encoder and the decoder to fuse the coarse-grained features of different levels. Thirdly, a cross-layer Transformer module is deployed at the bottom of the network to cross-fuse deep semantic information, thus refining vessel feature contours. Finally, a fusion loss function is used to conduct supervision on the training and testing of the algorithm. According to the results of experiments conducted on the DRIVE, STARE, and CHASE_DB1 datasets, the accuracy reaches 97.10%, 97.66%, and 97.62%, respectively; the specificity reaches 98.64%, 99.03%, and 98.72%, respectively; and the F₁ score reaches 83.05%, 84.07%, and 81.18%, respectively.