Research on 3D point cloud modeling and evaluation technology for transmission lines and corridor environments

This study utilizes ground-based LiDAR to collect high-resolution point cloud data and establishes a unified annotation system covering 11 object classes. The dataset is preprocessed through sampling, augmentation, normalization, and class balancing. An improved

PointTransformer model is proposed with enhanced feature encoding and spatial awareness to improve classification accuracy and efficiency.

To evaluate model performance, a manually annotated benchmark dataset was established as ground truth, and the model predictions were quantitatively compared with reference labels.

Based on classification results, key components such as towers and conductors are reconstructed using reverse modeling and a model reuse algorithm, enabling asset-to-model mapping and real-time digital representation.

Experiments on 300 GB of data show over 90% overall accuracy, with 99.1% accuracy for conductors. The model demonstrates strong generalization across 330 – 1100 kV scenarios, providing effective support for intelligent monitoring and digital management of transmission lines..