Advanced Autonomous Frameworks and Deep Learning Architectures for UAV-Based Power System Inspection

Abstract

The regular inspection and structural verification of high-voltage energy transmission lines constitute a critical operational imperative for maintaining the stability, safety, and reliability of global energy infrastructures. Historically, the energy sector has relied on manual, labor-intensive inspection methodologies, which are inherently characterized by high operational costs, prolonged execution timelines, and severe safety hazards for maintenance personnel required to navigate treacherous topographies. In recent years, Unmanned Aerial Vehicles (UAVs) have catalyzed a profound paradigm shift in power line maintenance, evolving from remotely piloted platforms to fully autonomous, edge-intelligent robotic systems. This comprehensive research paper provides an exhaustive analysis of the contemporary technological ecosystem surrounding UAV-based power system inspections. By systematically evaluating the integration of sophisticated hardware payloads—including LiDAR, thermal sensors, and edge-computing units such as the NVIDIA Jetson series—this paper delineates the transition from cloud-dependent processing to real-time onboard inference. Furthermore, the paper extensively reviews multi-objective path planning algorithms, ranging from Particle Swarm Optimization (PSO) to Fairness-Aware Multi-Objective Multi-UAV Path Planning (FAMO-MUPP), which govern safe navigation in complex transmission corridors. A major focus of this research is a deep-dive architectural analysis of convolutional neural networks (CNNs), specifically tracking the evolutionary trajectory of the YOLO (You Only Look Once) family of models from YOLOv4-tiny through the state-of-the-art YOLOv11. By dissecting ablation studies, feature extraction augmentations, attention mechanisms, and benchmark evaluations on the Chinese Power Line Insulator Dataset (CPLID) and the Insulator Defect Image Dataset (IDID), this paper highlights the critical trade-offs between computational complexity and detection precision. The synthesized findings provide an authoritative reference for optimizing edge AI deployments and advancing the frontier of intelligent grid operation and maintenance.