Abstract:An insulator is a device designed to withstand voltage and mechanical pressure on conductors with different potentials. Due to the impact of electrical environment and power load fluctuations, insulators may be subjected to various electrical mechanical coupling stresses, which may prevent them from working properly and affect the lifespan of the entire insulator network. To address this issue, a scheme was proposed to detect insulator damage through object detection algorithms. The improved solution is based on the YOLOv5s model. Firstly, more small object detection layers have been added to the original YOLOv5s model, thereby improving the detection accuracy. In addition, an additional computational layer was introduced to extend the feature map, and the SEA (Attention and Observation) attention module was used to make the network more focused on detecting objects. SIOU is also used to replace the Loss function in YOLOv5s. The experimental results show that the improved model has significant advantages in insulator damage detection compared to the traditional YOLOv5s model. The improved model improved by 2.5%, 1.1%, and 0.8% on indicators such as mAP (mean accuracy), P (precision), and R (recall), respectively. Compared with the original YOLOv5s model and other models (such as Yolov5m, Yolov5l, etc.), it has stronger competitiveness in insulator defect detection and recognition. These improvement strategies provide effective solutions for improving the accuracy of insulator damage detection. Through these improvements, we can more accurately detect insulator damage and take necessary repair and maintenance measures as soon as possible to extend the lifespan of insulators and ensure the stable operation of the power system.