Research on comparative detection of two categories based on normal bolts and bolts with missing pins

Pins are widely used in power transmission lines, and their main function is to fix bolts in connecting fittings. Bolts and pins are extremely small in size and have complex backgrounds, which makes detection very difficult. This paper compares multiple classic and latest target detection networks to explore the impact of single-category and dual-category annotations on the detection accuracy of bolt missing defects, thereby improving the detection accuracy of bolt and pin missing in complex transmission line scenarios. This method first uses the latest YOLOv8 target detection model to learn and train bolt pin missing to achieve bolt pin missing detection. Secondly, through multiple classic target detection networks, including Faster RCNN, SSD, and YOLOv5 classic target detection models, images containing single bolt defect targets and defect targets containing normal bolts and bolt pin missing are learned and trained respectively. Finally, the detection results of multiple models are compared to analyze the impact of single-category and dual-category on the accuracy of bolt pin missing detection. The experimental results show that the YOLOv8 network trained with two categories can more accurately detect bolt pin missing targets that were not detected by the original single-category training network, effectively improving the detection accuracy of bolt pin missing.