Powerline extraction from aerial and mobile LiDAR data using deep learning
Vaibhav Kumar,Aritra Nandy,Vishal Soni,Bharat Lohani
DOI: https://doi.org/10.1007/s12145-024-01310-w
2024-04-26
Earth Science Informatics
Abstract:Accurate powerline classification from LiDAR point clouds is essential for efficient monitoring and management of power distribution networks. Currently, the classification is being done through manual labeling or some semi-automatic methods which are time-consuming and resource intensive. In this study, we explore three deep learning architectures, namely KPConv, PointCNN, and RandLA-Net, for powerline segmentation in aerial and mobile LiDAR datasets. We utilize manually labeled aerial and mobile LiDAR datasets from Surry in Canada and Kerala in India, respectively. Impressive results are obtained, demonstrating high powerline classification accuracy and precision. KPConv outperformed the other architectures, achieving over 98% accuracy for ALS data and surpassing 94% for MLS data. Additionally, we investigate the impact of ground removal on powerline extraction. The results highlight a significant improvement in powerline classification accuracy, with an approximate 3–4% enhancement observed, particularly in the test sets with a higher density of powerline points. Removing ground points effectively reduces noise and clutter in LiDAR data, leading to improved segmentation outcomes. Furthermore, ground removal proves beneficial in classifying wires in challenging scenarios, such as areas with nearby objects or branch and loose wires. KPConv demonstrates superior performance in powerline classification, capturing intricate powerline details even in scenes with varying point densities. PointCNN shows limitations, especially in wire point classification near other objects. RandLA-Net exhibits higher speed but slightly lower accuracy compared to KPConv. The findings contribute to advancing powerline classification and provide valuable insights for more efficient and automated management of power distribution networks.
geosciences, multidisciplinary,computer science, interdisciplinary applications