Abstract: Edge computing devices have been widely used in unmanned aerial vehicle (UAV) power transmission line inspections. However, the limited power and computational capabilities of edge devices hinder the detection speed and accuracy. To address these challenges, we improve the YOLOv7 target detection algorithm and propose an edge-end defect detection method for transmission lines based on the LEE-YOLOv7. Firstly, the Mosaic-9 data enhancement method is used to improve the inputs in the training phase to improve the generalization ability of the network. Then, the LCnet network is introduced to transform the backbone network part, to reduce the redundant parameters and to lighten the network. Subsequently, the Meta-ACON activation function and Wise-IoU loss function are applied to optimize the network. Finally, during the inference process, the Deepstream technology is utilized to invoke the TensorRT module, achieving model reconstruction, optimization, and acceleration. The experimental results demonstrate that the LEE-YOLOv7 achieves an average accuracy of 92.3% in identifying ten common types of power transmission line defects with varying sizes. Compared to the original YOLOv7 algorithm, the proposed approach improves accuracy by 2.8% and increases detection speed by 38 frames per second, reaching 53 frames per second. After optimizing the inference process using Deepstream and TensorRT on the Nvidia Jetson Xavier NX edge module, the approach achieves an average accuracy of 91.2% and 79 frames per second for defect detection, satisfying the requirements of accurate real-time transmission line defect detection at the edge end.