Detection of Insulator Defects in Transmission Lines Based on Lightweight Convolutional Neural Networks

The role of the transmission line insulator is crucial in a power system. In the past few years, due to the advancements in artificial intelligence technology, insulator defect detection based on UAV intelligent patrol technology has become a research hotspot in power equipment detection. Focusing on the challenges related to demanding real-time requirements and constrained computing resources in UAV online inspection.. Two lightweight improvement schemes of YOLOv5 are proposed. The enhanced model combines the strengths of lightweight convolutional neural networks, namely MobileNetV3 and GhostNet, with the classic YOLOv5. The experimental findings demonstrate that this improved model efficiently reduces the computational load and simplifies the algorithm’s complexity while maintaining detection accuracy. YOLOv5 MobileNetV3 model reduces the computational load by 85.4%, and the detection accuracy slightly decreases; The calculation amount of YOLOv5 GhostNet model is reduced by 49.4%, and high detection accuracy is maintained. Henceforth, the suggested model is better suited for deployment on UAV platforms, enabling real-time detection of transmission line insulator flaws.