Improved YOLOv8 for lightweight insulator defect detection method

This paper proposes YOLOv8-SDEI, an efficient insulator defect detection model based on YOLOv8, to address low frame rates and insufficient accuracy. It enhances feature fusion with a multi-scale fusion and shared parameter detection head, introduces an Efficient Multi-Scale Attention (EMA) mechanism for better focus on key areas, and uses depthwise separable convolution to reduce parameters and computational complexity. Inner-IoU is adopted as the bounding box loss function to improve small target detection precision. Experiments on a high-voltage transmission line dataset show the model maintains 90.6% mAP, reduces computational cost by 2.2G, and increases accuracy by 1.6%. With a model size of 4.5MB, it achieves 83.1% precision and 94% recall, improving by 1.6% and 0.8% respectively. YOLOv8-SDEI accurately detects insulator defects, reduces computational load, and is suitable for real-time deployment on embedded devices, providing a novel solution for insulator defect detection in transmission networks.