Drone-Based Image Analysis For Automated Inspection Of Transmission Lines

Abstract

To keep the grid reliable, constant monitoring and inspection of transmission lines are needed, but traditional foot patrols and helicopter inspections are time- and resource-intensive and pose risks to both people and equipment. Uncrewed aerial vehicles (UAVs), or “drones,” equipped with high-resolution visual, thermal, and light detection and ranging (LiDAR) sensors are proving to be an effective solution for capturing aerial imagery of towers, conductors, insulators, and surrounding vegetation corridors. This manuscript summarizes the latest advances in image analysis algorithms and the development of architectures for automated inspection of transmission lines using drones, focusing on deep learning object detection models such as Faster R-CNN, the Single Shot Multibox Detector (SSD), the You Only Look Once (YOLO) series of models, and the detection transformer (DETR). An overview of the methodological framework, from the selection of the UAV platform to the configuration of the sensor payload, flight-path planning, image acquisition, pre-processing, and defect classification using image models, is presented. Based on the literature, lightweight architectures such as YOLOv8, which can achieve detection accuracy above 0.93 and achieve inference speeds of more than 55 frames per second, are suitable for deployment in onboard or near-real-time applications, whereas transformer-based detectors are better suited to more time-consuming applications. Major difficulties include small-object detection in cluttered backgrounds, a lack of annotated fault imagery, variations in illumination and weather during measurement, and limited onboard computational power. The manuscript concludes with a summary of existing challenges and a vision for future directions, including multimodal sensor fusion, collaborative joint inference in edge-cloud schemes, and the development of standardized benchmark datasets for transmission line asset inspection.