Research on Detection Method of Transmission Line Strand Breakage Based on Improved YOLOv8 Network Model

Abstract

Aiming at the existing object detection models has difficulty effectively capturing relevant features when detecting small targets such as wire breakage faults in transmission lines, resulting in low detection accuracy. An improved algorithm, named DRS-YOLO, has been developed based on YOLOv8 for detecting broken wires in transmission lines. Firstly, by adding a deformable convolution module (C2f_DCNv3) and mixed attention convolution module (RFASEConv) to the backbone network of YOLOv8n, the receptive field of the model is expanded to improve the detection accuracy of small targets in complex backgrounds. Secondly, the Inner-SIOU loss based on auxiliary bounding boxes is adopted as the loss function to enhance the feature extraction efficiency of the model and further improve its accuracy in detecting wire breakage in transmission lines. Finally, DRS-YOLO is analysed in comparison with Faster-RCNN, SSD, YOLOv9-c, YOLOv5n and YOLOv8n. The results showed that compared with other detection methods, DRS-YOLO has higher detection accuracy smaller parameters and computational complexity in detecting wire breakage faults in transmission lines. Its average detection accuracy (mAP) is 92.5%, the recall is 85.8%, and the Precision is 95.4%. Compared with the original YOLOv8n network, it has improved by 7.6%, 4%, and 2.7% respectively, while the parameter quantity is 2.76M, which is 8% lower than the original YOLOv8n. Compared with other existing object detection models, the DRS-YOLO model has achieved good results in mAP, accuracy, and recall, and can efficiently and accurately complete the task of detecting broken wires in transmission lines.

Keywords

• Transmission line strand breakage detection
• hybrid attention conv
• deformable conv
• unmanned aerial vehicle
• YOLOv8