The extensive coverage and complex terrain of power transmission lines pose significant challenges for traditional manual inspection methods, necessitating more efficient approaches to meet contemporary power industry demands. In recent years, the use of unmanned aerial vehicles (UAVs) equipped with LiDAR for power inspection has received widespread attention. However, the point cloud data obtained from inspections are massive, requiring more intelligent point cloud automation processing methods. Therefore, this paper proposes an automatic semantic segmentation method based on deep learning networks for LiDAR point clouds of power corridors to extract targets such as power lines and trees for hazard analysis. For vegetation and power line distance measurement in safety hazard detection, a projection-based power line fitting algorithm is designed. Additionally, a projection-based tree height detection algorithm is proposed for tree height measurement. Experiments conducted on LiDAR point cloud data provided by power grid companies demonstrate that the proposed semantic segmentation algorithm achieves an average accuracy of 84.92% across six different power corridor scenarios, with key point clouds like power lines, vegetation, and poles exceeding 90% accuracy. The tree height measurement error is confined within ±0.5 meters, and the distance detection algorithm accurately identifies danger points based on predefined safety margins.