MRB-YOLOv8: An Algorithm for Insulator Defect Detection

As China’s electricity consumption surges, the reliability and safety of long-distance transmission lines become increasingly crucial. Insulators, vital for grid stability, demand accurate defect identification. Existing methods fall short on small targets and complex backgrounds. An insulator defect detection method MRB-YOLOv8 is proposed. By integrating an attention mechanism and multi-scale features, the model’s focus on key features is significantly improved. The Multi-Spectral Channel Attention captures essential information across different frequency domains through a well-designed frequency selection strategy. In addition, Receptive Field Attention Convolution (RFAConv) replaces the C2f module in the backbone network, which enhances the ability to perceive the features in complex backgrounds through the weighting operation of the receptive field weights. Meanwhile, the weighted bi-directional feature pyramid network (BiFPN) and a fourth detection layer prevent feature loss during fusion, enhancing the detection accuracy of small targets. Experimental results show that, at mAP50 and mAP50:95, the improved method obtains a gain of 3.2% and 3.6%, respectively, which significantly improves the model’s capability of detecting defects such as insulator self-explosion, breakage, and flashover in the images captured by UAVs.