Key points are not available for this paper at this time.
The traditional practices of foot patrol and manned airborne survey for inspecting electric utility assets are now deemed slow, costly, albeit subject to inaccuracy and hazard. As a result, current efforts turned to unmanned aerial vehicles (UAVs) with onboard cameras and equipped with computer vision technologies. In this paper, the authors address two main challenges of a computer vision system for unmanned aerial vehicles in utility asset inspection: (i) the scarcity of data; and (ii) the detection of six power line components, namely transformer bank, high voltage bushing, low voltage bushing, arrester, radiator fins, and cutoff fuse. In addition, the new curated dataset contains images of the subject of interest taken in pole-mounted and pad-mounted assemblies inside the university using an unmanned aerial vehicle. Furthermore, flipping, injecting different brightness adjustments, and other data augmentations to the dataset simulate the diverse environmental conditions. Finally, the authors re-trained the You Only Look Once (YOLOv5) pre-trained checkpoints on the collected custom data. The experimental results reveal that the proposed system is accurate and precise in detecting and classifying the power line components, with fewer missed detections.
Sumagayan et al. (Sun,) studied this question.