Rui Ye, Kaifeng Deng, Lei Yu, Peiyuan Gao, Lei Chen, Jinxin Yang, and Yaping Zhang
Overhead transmission lines, elevation interval threshold, veg-etation point clouds, subspace extension strategy, point cloud classification
In the health monitoring of overhead transmission line equipment, intelligent point cloud classification confronts two critical challenges: discontinuous extraction of transmission lines caused by large terrain elevation differences, and high costs associated with acquiring labelled data. To address these issues, this paper proposes an airborne laser point cloud classification method for overhead transmission lines based on subspace extension and adaptive multidimensional feature mining. Initially, large and small elevation interval threshold point clouds are extracted via an elevation statistics histogram approach integrated with a subspace extension strategy, and then a surface is constructed via Delaunay triangulation to filter ground and vegetation points from the small elevation interval point cloud, yielding a purified point cloud. After geometric features (linearity, curvature, and elevation) are computed from this point cloud via principal component analysis, the particle swarm optimisation algorithm adaptively identifies the optimal thresholds for these features to achieve coarse classification of overhead transmission lines. Ultimately, the subspace extension strategy is reapplied to refine classification results. Extensive experiments on point cloud datasets across diverse scenarios show the classification metrics of the proposed method outperform current mainstream unsupervised methods, with over 2.5% higher intersection over union, over 5% higher classification accuracy, and a significant reduction in classification time.
Important Links:
Go Back
