Transmission Line Foreign Object Detection Network Based on Improved YOLOv8

Aiming at the problems of low accuracy and poor robustness of transmission line foreign object detection in complex environments, a transmission line foreign object detection algorithm based on YOLOv8 is proposed. First replace the standard convolution in the backbone of YOLOv8n with a sensory wild attention convolution that incorporates SE attention, and embed it into the C2f module to construct a new lightweight backbone network; Secondly, a double-layer routing attention mechanism is added to the neck to improve the accuracy of recognizing small foreign objects; Finally, Shape-IoU is introduced to improve the regression rate and accuracy of the boundaries. Experiments based on the foreign object dataset expanded by Poisson mixing and geometric data enhancement methods show that the improved YOLOv8-CBS has an average accuracy of 92.1%, an accuracy of 94.6%, and a detection speed of 156 f/s, which are 3.5%, 4.3%, and 11.0%, respectively, compared to the baseline model, and the model volume is only 3.2 MB.