Research on small object foreign object detection algorithm for transmission lines based on improved YOLOv8s

An enhanced YOLOv8s-based algorithm is introduced to resolve the issues of frequent missed detections and low precision in identifying small foreign objects during transmission line inspections. Firstly, construct a dataset of foreign objects in transmission lines and optimize it to reduce the impact of the environment on identification. Secondly, use two optimization methods: introducing ECA (Efficient Channel Attention) mechanism in the YOLOv8 backbone structure, strengthening the input image channels, and reducing the model parameters to make it lighter and reduce the missed detection rate. Adding deformable ConvNets v2 to the Bottleneck module in C2f module improves the model's capacity for learning offsets, thereby further improving the ability to detect small objects. Finally, ablation experiments were conducted to verify the effectiveness of the improved algorithm. The experimental findings indicate that the model achieves an accuracy of 98.571%, a recall rate of 94.812%, and an average detection precision of 97.375%. When compared to the original algorithm, it has improved by 6.856%, 6.232%, and 5.084%, respectively, and has good detection performance.