Abstract: Aiming at the problem of uneven deformation of airborne LiDAR point clouds caused by severe terrain undulations in mountainous areas of ultra-high voltage transmission channels, resulting in mismatch between vegetation and conductor spatial benchmarks, and a sharp increase in the false correlation rate of segmentation algorithms, a warning of tree obstacles in ultra-high voltage transmission channels under unmanned aerial vehicle point cloud segmentation is proposed. Construct a point cloud collection scheme that combines ground simulation routes with differential fusion solutions, design adaptive dispersion discrimination and semantic clustering segmentation, and achieve precise separation of conductors, vegetation, and towers. Establish a catenary multiphysics coupled deformation model to dynamically calculate the shortest safe distance, combine vegetation growth rate and meteorological factors to construct multi-level dynamic risk thresholds, and achieve hazard classification warning. Through actual testing of 500kV transmission lines, this method has a false alarm rate of less than 20% for tree obstacle hazard warning in transmission channels, and can achieve high-precision warning of tree obstacle hazards.