Figures from this paperAbstractFigures from this paperAbstractFigures from this paperAbstractFigures from this paper

Abstract

Insulators play the role of electrical isolation and mechanical support in transmission lines, and once a failure occurs, it will seriously threaten the stable operation of transmission lines. Due to the limitations of traditional methods in detecting insulator fault points in complex backgrounds. Therefore, we propose an insulator fault detection network (SMI-YOLO) based on the improved YOLOv5, which is suitable for the detection of small targets in complex backgrounds. First, a deep information extraction network (SHNet) is proposed to enhance the feature representation of the target and effectively counter the interference generated by the complex background. Then, to adaptively fuse the features in different scale feature layers, a multi-scale feature-consistent fused feature pyramid network (MA- FPN) is designed, which enhances the model’s detection ability for multi-scale small targets. Subsequently, considering that the original decoupling head ignores the redundant information generated in the network, an information reconstruction decoupling head (IRHead) is proposed to improve the information utilization in both channel and spatial dimensions, which further improves the accuracy of insulator fault detection. Experimental results show that SMI-YOLO has a higher detection accuracy than other models, and can accurately and efficiently accomplish the insulator fault detection task, which is significant for maintaining the safe and stable operation of transmission lines.