Edge-Optimized Defect Detection for Substation Insulators

Insulator self-detonation defect detection is of great significance to ensure the stable operation of power system and prevent large-scale power outage. Traditional manual inspection and helicopter inspection methods have the limitations of high cost, low efficiency and strong environmental dependence. In this paper, a lightweight edge-optimized insulator self-detonation detection algorithm is proposed. By replacing the backbone network of YOLOv4 with MobileNetv3, depth separable convolution is used to reduce the number of parameters, and an improved loss function is designed to improve the detection performance. The DE-BiFPN module is further introduced to enhance the multi-scale feature fusion capability, and the C3TR module is combined to capture global information. Experimental results show that the detection accuracy of the improved model on the test set ([email protected]) is 1.65% higher than that of the original YOLOv4, the loss value converges to 0.71%, and the improved model exhibits strong robustness under complex background and small-target scenarios. Compared with Mobilenet-SSD, the proposed model reduces overfitting problems and significantly improves defect detection accuracy, which verifies its high efficiency and practicability on edge devices.