Transmission wire plays an important part of the power system, and defects in transmission wire will affect the operation of the power system. A transmission wire defect detection method based on the improved YOLOv7 algorithm is proposed. First, a method of automatically expanding the data set is proposed in this paper, which can use a small number of pictures to establish a data set of transmission wire defects. Then, a lightweight self-attention backbone network is proposed, which replaces the YOLOv7’s backbone network, and uses the BiFPN to perform feature fusions. The experimental results show that the improved YOLOv7 algorithm proposed improves the accuracy from the original 89.4% to 97.5%. At the same time, the detection speed is increased by about 60.49%, and the FPS is improved from the original 52.36 frames to 84.03 frames; real-time detection of transmission lines can be achieved, which reduces the misdetection and omission rate of transmission wire defects, as well as improves the detection speed and enhances the inspection efficiency.