Insulator Defect Detection Based on Improved YOLOv5

Insulators play an important role in power systems. Insulators are easily damaged and need to be inspected regularly. Manual inspections are inefficient. Drone inspections are gradually replacing manual inspections. In order to make the insulator defect recognition algorithm more lightweight and deploy it on drones, the YOLOv5 algorithm is lightweighted and improved. First, the YOLOv5 backbone network has many parameters and large computational complexity, so the backbone network is replaced with a lighter MobileNetV3. Then, in order to better extract the features of insulator defects, the C3 module is integrated with a common attention mechanism. It is found that spatial information is more important than channel information for insulator defect recognition. Finally, the experimental results indicate that the improved network achieves a 54% reduction in parameter count, a 0.7% improvement in precision, and a 65.8% decrease in GFLOPs compared to the original algorithm.