A lightweight transmission line foreign object detection algorithm based on improved YOLO11n

In response to the challenges posed by low detection precision and large parameters and computational in existing models for identifying foreign objects on transmission lines, a lightweight transmission line foreign object detection model based on improved you only look once 11 nano (YOLO11n) is proposed. First, the lightweight StarNet network is used to improve the backbone network of YOLO11n, which significantly decreases the parameters and computational complexity of the model by utilising the star operation. Second, the cross stage partial with kernel size 2 (C3k2) from YOLO11n is fused with the StarNet network's star blocks, forming C3k2_Star, thereby boosting the model's feature extraction performance. Finally, the original detection head in YOLO11n is substituted with a lightweight detection head, reducing the parameters and improving the model's accuracy using shared convolution. Experimental results reveal that the improved model improves the mean average precision on the transmission line foreign object detection dataset by 1.9% compared with the baseline model, and the parameters and computation are decreased by 32.73% and 31.75%, respectively, and a detection speed of 117.7 frames per second. The improved model achieves the synergistic optimisation of detection accuracy and lightweight in the transmission line foreign object detection task, significantly reduces the hardware resource requirement, and furnishes a reliable technical guarantee for real-time warning of transmission line safety hazards.