Inspection and identification of transmission line insulator breakdown based on deep learning using aerial images

• Published in: Electric power systems research 2022-10, Vol.211, p.108199, Article 108199
• Main Authors: Ahmed, MD.Faiyaz, Mohanta, J.C, Sanyal, Alok
• Format: Article
• Language: English
• Subjects: Aerial images; Data augmentation; Insulator detection; Single shot multibox detector; Unmanned aerial vehicle/quadcopter; Vision based inspection
• ISSN: 0378-7796
• DOI: 10.1016/j.epsr.2022.108199

•Existing power line inspection methods are reviewed.•A new methodology has been proposed for coordinated movement of quadcopter along the transmission line monitoring.•A novel automatic autonomous vision-based power line inspection concept is proposed.•This paper designs a lightweight insulator fault detection network that can be deployed at the edge.•The accuracy of insulator fault detection is improved by enhancing the various performance metric function of the network.•A deep learning algorithm strategy is proposed to complete insulator fault detection at the edge with the help of autonomous quadcopter. The traditional methods of overhead power transmission line inspections are mostly unsuited as the height of transmission towers is too high and wide. Detection and inspection of insulators in aerial images with cluttered backgrounds is a challenging task for autonomous inspections. This manuscript mainly focuses on the development of autonomous Unmanned Aerial Vehicles (UAV/Quadcopter) that can hover over the transmission towers and capture images and videos by following pre-determined waypoints. To accomplish this, authors propose a new autonomous vision-based inspection that uses a Quadcopter as primary source of data, aerial images as the main source of information, and Deep Learning (DL) as the backbone analysis for inspection and focused on (i) insufficient training data, (ii) detection of insulators and their defects. A medium sized dataset of insulators for training and detection is created to overcome data insufficiency. The experimental results shows that the proposed deep learning architecture successfully identifies the anomalies of insulator such as, cracks, missing top caps and broken disk etc. The detection accuracy of the proposed deep learning algorithm can reach up to 93.5% with a detection speed of 58.2 frames/sec. The proposed DL algorithm has a promising potential towards smart inspection of insulators in power grids.