YOAM: An Efficient Object Detection Network for Protecting Overhead Transmission Lines

To ensure the sadety of power transmission lines,monitoring cameras are installed at high altitudes for target detection and recognition, allowing for early warnings of mechanical hazards. Real-time performance and accutacy are critical indicators for evaluating a model. To enhance the accuracy and real-time performance of the YOLOv8 model in detecting and recognizing large targets, distant small targets, and targets in complex backgrounds,this paper proposes a new model called YOAM based on two improvements to YOLOv8. First, a grouped depthwise separable convolution is designed to make the network model more lightweight. Second, an improved attention mechanism is designed based on the channel attention mechanism, allowing the attention mechanism to remain lightweight while focusing more on useful information and suppressing irrelevant feature information. To better test the model's performance, we designed a new dataset. Finally, experiments show that YOAM outperforms the YOLOv5 and YOLOv8 models on the existing test sets, achieving an mAP value of 96.7%. Additionally, on the newly constructed dataset, YOAM successfully detected all targets in the images, while YOLOv5 and YOLOv8 showed cases of missed detections.