Abstract

The maintenance of power lines is challenged by the encroachment of vegetation, posing significant risks to the reliability and safety of power utilities. Traditional methods, based on manual inspection, are not only resource-intensive but also lack the necessary precision for effective and proactive maintenance. This paper aims to develop an automated, accurate, and efficient approach to vegetation management in the vicinity of power lines. It leverages advancements in data collection using LiDAR scanning technology, which despite its potential, faces computational challenges in processing large-scale 3D point clouds to accurately identify power lines and surrounding vegetation. To overcome this challenge, the proposed method deploys the RandLA-Net model for the semantic segmentation of power lines and nearby vegetation in point cloud datasets. Furthermore, the post-processing analysis of the segmented data uses clustering and rule-based thresholding to refine the identification of vegetation. Then, proximity detection is applied using spatial queries based on a KDTree structure. The results of the case study demonstrate the computational efficiency and accuracy of the proposed method, presenting a promising solution for power utilities.