An Improved YOLOv8 UAV Transmission Line Inspection Image Recognition Detection Network

In the process of UAV transmission line inspection, the aerial image quality is uneven, the background of real transmission lines is complex, the types of targets to be detected on line are diverse, the scale of targets varies greatly and the size difference is large, which leads to the difficulty of recognition and detection. In order to solve the problems, an improved YOLOv8 UAV transmission line inspection image recognition and detection network is proposed. Firstly, the multi-head self-attention mechanism was introduced into the backbone network to effectively extract the context information of the detection target, which improved the detection ability of transmission line targets under complex backgrounds. Secondly, a small object detection layer is added to the detection head, and the bounding box loss function is optimized, which enriches the network scale information and reduces the false detection rate and missed detection rate. Experimental results show that the proposed method can accurately detect transmission line targets. The average precision rate, average recall rate and average F1-score of six types of targets of transmission lines are 93.34%, 93.86% and 94.93%, respectively, showing strong accuracy and robustness.