The images of HGIS bushing terminal blocks in substations often contain rich details, such as different materials of bushings, complex wiring structures, and possible background interferences. These details make contour extraction more complex. In order to improve the accuracy of extracting the contour of substation HGIS bushing terminal, an attention mechanism based method for extracting the contour of substation HGIS bushing terminal is proposed. Using edge guided operator template matching technology, combined with sliding window and nonlinear grayscale transformation algorithm, preprocess the HGIS bushing terminal image of the substation, enhance the edge information of the image, and reduce the impact of lighting and noise. In order to accurately extract the contour of wiring terminals, a deep learning network architecture based on attention mechanism was designed. The network gradually enriches feature information through an encoder, and the decoder restores high-quality contour clear images. In addition, the convolutional attention module (CBAM) and attention mechanism sub network in the network can accurately focus on the contour area of the terminal block, suppress non target features, and thus achieve accurate extraction of the terminal block contour. The experimental results show that the proposed method can accurately extract the contour of the HGIS bushing connection terminals in substations, providing strong support for subsequent image analysis and recognition tasks.