Vision Inspection of Power Lines with Deep Learning

Abstract

The purpose of this technical paper is to introduce a novel approach to inspecting power lines using computer vision and deep learning algorithms. Traditional inspection methods are often time-consuming and costly and can be dangerous for human inspectors. This paper presents a new workflow for power line inspection that leverages machine learning algorithms to automate the process. The proposed workflow for vision inspection of power lines involves capturing high-resolution images of power lines using drones or other unmanned vehicles. These images are then processed using a deep learning algorithm, which is trained to identify potential condition such as good, problem, and unknown for various components in the power lines. The workflow utilizes cutting edge object detector models, such as YOLOv5 and YOLOv8 to analyze the images, and output the prediction of bounding boxes, abjectness scores and probabilities for each detected object. This paper presents a novel approach to power line inspection that utilizes deep learning algorithms to identify and localize objects not only by using axis-aligned bounding boxes but also with oriented bounding boxes, which can significantly reduce the time and cost associated with traditional inspection methods. The use of drones and other unmanned vehicles also increases safety by eliminating the need for human inspectors to climb power line structures. The vision inspection workflow presented in this paper has the potential to revolutionize the power line inspection industry. By utilizing deep learning algorithms and unmanned vehicles, inspection times can be reduced, costs can be lowered, and human safety can be improved. This approach can also lead to more efficient maintenance and repair, prolonging the lifespan of power lines and reducing the risk of power outages. Future research can explore the potential of using this workflow for other types of infrastructure inspection as well.