Aiming at the problems of low resolution of image segmentation, imprecise segmentation, and less defective sample components leading to model overfitting and insufficient generalization ability due to the complex background and fuzzy target in the process of transmission line inspection. In this study, an image segmentation algorithm based on the global-local attention mechanism of GL-Unet is designed based on the U-shaped network structure. Firstly, the Swin Transformer Block structure is used in the encoder to learn the deep feature representation, which makes up for the defects of the convolution operation and can better carry out the semantic feature learning locally and globally; secondly, the Transformer Block with Global--Local transformer block (GLTB) is introduced to construct the decoder, which captures the defective global contextual information through local branching and global branching. And the contextual information is further fused through the cross-shaped contextual interaction module. Finally, the feature refinement head (FRH) is embedded in the tail of the network to effectively fuse the spatial details and contextual information as well as further optimize the feature map to improve the accuracy of the segmentation network. The experimental results show that, compared with the baseline model, the mIoU score of the GL-Unet image segmentation network designed in this study reaches 0.912, and the comprehensive performance of mIoU, mPA, and acc is advantageous compared with other mainstream segmentation algorithms, which can better satisfy the needs of the actual inspection scenarios of transmission lines.