CSWin-Unet: Fusion Multi-scale Feature Insulator Segmentation Model

Insulator image segmentation is a process of separating and extracting the insulator part from the complex background in insulator image, which is of great significance to the automatic detection and fault diagnosis of power system. Convolutional neural networks have been performing well in image segmentation tasks for the past few years, and Transformer has been applied to computer vision tasks with great success. Transformer has the advantage of building global connections, but at the same time it lacks CNN's ability to generalize local information. In order to achieve fast and accurate identification of insulators under complex background, a U-shaped insulator image segmentation method is proposed in this paper. The proposed method utilizes the generalization ability of local features of CNN and the generalization ability of global features of Transformer to carry out feature extraction respectively. Then, global features and local features are fused through the feature fusion module, in which way the model can learn more robust features.