Self-Adaptive Sliding Map Convolution Multi-Scale Feature Fusion Classification Method for LiDAR Point Clouds of Transmission Lines in Complex Terrain Environment

In the process of smart grid construction, the safe and stable operation of transmission lines is the key to ensuring reliable power supply. However, LiDAR point cloud classification of transmission lines in complex terrain environments faces severe challenges. Traditional classification methods are difficult to accurately distinguish various types of ground objects, resulting in poor accuracy and reliability. Therefore, this article innovatively proposes an adaptive sliding graph convolution multi-scale feature fusion classification method. This method utilizes a dynamic neighborhood aware sliding graph convolution operator based on entropy analysis and terrain roughness adjustment, which can flexibly construct neighborhoods based on local features of point clouds and terrain complexity, and accurately extract features. At the same time, a terrain adaptive multi-level feature pyramid architecture is adopted to integrate micro, meso, and macro scale features, fully understanding complex terrain scenes. The self built STPL-3D dataset covers various terrains, device types, and point cloud density distributions, indicating that this method is significantly superior to traditional and existing deep learning methods in terms of overall accuracy, F1 score, and other indicators. In cross canyon line icing detection, the false alarm rate is greatly reduced; the response time of the tree obstacle warning system has been significantly shortened and the accuracy has been significantly improved. In practical engineering applications, it has excellent performance and provides efficient and reliable technical support for the detection of smart grid transmission lines, effectively promoting the development of smart grids.