Abstract
INTRODUCTION: Power transmission and distribution lines operate in complex natural and electromagnetic environments, making them susceptible to risks such as tree encroachment, ice accumulation, insulator damage, and foreign object interference. Traditional inspection methods, including manual patrols and helicopter surveys, are inefficient, costly, and unsafe, and they lack real-time monitoring and large-scale coverage capabilities.
OBJECTIVES: This study aims to improve the accuracy, reliability, and real-time capability of distribution network inspections by developing an integrated target recognition and 3D reconstruction system based on BeiDou satellite communication and edge computing.
METHODS: An enhanced YOLOv8n-based target detection model incorporating C2f modules, lightweight convolution, and CBAM attention mechanisms is designed for real-time component and obstacle recognition. Point cloud clustering and graph embedding techniques are employed to reconstruct the three-dimensional structure of transmission lines and locate obstacles spatially. BeiDou short message communication is utilized to ensure stable data transmission under strong electromagnetic interference.
RESULTS: Experimental results show that the proposed system achieves superior performance in terms of mAP, F1 score, and small-object detection compared with baseline models. The system maintains stable communication in high-electromagnetic-interference environments and enables accurate 3D reconstruction and real-time fault perception across complex inspection scenarios.
CONCLUSION: The proposed BeiDou- and edge-computing-based inspection system effectively overcomes the limitations of traditional methods, significantly enhancing detection accuracy, communication reliability, and spatial perception for distribution network patrols, thereby improving grid safety and operational efficiency.