A Res-UNet model for transmission segmentation based on attention mechanism

Image segmentation is widely used to intuitively observe transmission lines and substations to maintain transmission stability, but the small proportion of transmission lines and complex background information in high-altitude images undoubtedly increase the difficulty of segmentation. Therefore, this paper uses the transmission image segmentation data set, and constructs a new network structure with high precision and strong anti-jamming ability on the basis of comparing several mainstream semantic segmentation models. In order to enhance the expressive ability of network features and prevent the disappearance and degradation of gradients in network training, ResNet is used as the backbone of U-Net, and attention mechanism is introduced into the sampling part of the network to help the network better reconstruct the details and spatial information of the original image. The results show that the F1 score of the model in the test set is 91.43% and the score of mIoU is 81.3%. Compared with the existing models, mIoU has been improved by nearly 5%, reaching the level of practical application.