In recent years, with the rapid development and widespread application of image recognition technology, deep learning based object detection algorithms have been studied and applied to distribution lines to some extent. Aiming at the problems of low efficiency of traditional target detection algorithm in distribution line scene image detection and bottleneck of cloud server prediction service computing power, the deployment of improved PPYOLOe model algorithm on Edge device is proposed to realize the real-time detection task on distribution lines. Firstly, the use of ESE attention mechanism after the backbone network and CA attention mechanism after the neck feature extraction network has improved the model’s feature extraction ability in channels and positions; Secondly, the feature matrices output by the backbone network were fused, which improved the calculation speed of the model and also improved its robustness; Finally, the obtained model was quantified and deployed on the NVIDIA Jetson Xavier of the Aarch64 architecture, achieving edge prediction services. The experimental results show that on the locally collected dataset, the average detection accuracy (mAP) of the algorithm reaches 89.33%. After lightweight, the inference acceleration ratio reached 2.24, with an average detection accuracy of 85.84%, and the processing frames per second (FPS) on Xavier was 4.5.