Lightweight-Based Defect Detection for Small Target Insulators

Identifying defects in insulators, especially in small targets, remains a significant challenge due to the diverse types of insulators and complex transmission line backgrounds. This paper aims to enhance the accuracy of defect detection in insulators while minimizing the model’s parameter count for stable power system operation. To achieve this, we propose improvements to the feature fusion mechanism in the YOLOv8 network. These enhancements focus on better integrating multi-scale feature information, particularly emphasizing the detection sensitivity towards small targets. Additionally, these modifications lead to a reduction in the model’s parameter count while improving the detection accuracy of small targets. Furthermore, we introduce the SIOU loss function to the improved model, further enhancing its accuracy. Through experiments, our improved model achieves a mean average precision (mAP) of 89.2%, surpassing YOLOv8n by 2.7%. Remarkably, the model’s parameter count is significantly reduced to 2,796,249, demonstrating a balance between higher model accuracy and fewer parameters.