Extraction of Power Line Based on Point Cloud by Using Deep Learning

In recent years, with the vigorous development of green energy, the characteristics of the electricity industry, such as large scale, long chain, wide coverage, and strong social dependence, have become increasingly prominent. The massive power transmission network urgently needs an efficient, fast, and data-intelligent power corridor inspection system to ensure the safe operation and stable transmission of power energy. Efficient and accurate extraction of power line elements is the core and difficulty of the power corridor inspection system. This paper proposes an efficient and precise method for extracting power line elements based on the RandLA-Net network model tailored for LiDAR point cloud data. The method first preprocesses the point cloud (including denoising, down-sampling, and ground point filtering), and then integrates point cloud spatial and color information to train and predict based on the RandLA-Net network model. Experimental results demonstrate that this method outperforms traditional machine learning algorithms such as random forest and extreme gradient boosting in power line extraction accuracy, with F1 scores for pole lines and conductors reaching up to 99% and 98% respectively, and intersection over union rates reaching up to 99% and 97%. The RandLA-Net deep learning model shows significant potential for application optimization in power line element extraction tasks.