北京测绘2025，Vol.39Issue(6)：824-829,6.DOI:10.19580/j.cnki.1007-3000.2025.06.012
基于图卷积神经网络的车载激光点云分类
Classification of vehicle-mounted laser point cloud based on graph convolutional neural network
摘要
Abstract
This paper proposed an automatic classification method for point cloud data collected by vehicle-mounted laser scanning,which combined the strengths of segmentation algorithms and graph convolutional networks(GCN).In the initial data processing stage,the noisy density-based spatial clustering of applications with noise(DBSCAN)method was utilized to meticulously classify the point cloud data into multiple point clusters.These point clusters served as nodes in a graph,where edges connected adjacent nodes to form the graph structure.A GCN was introduced to perform semi-supervised classification on the graph nodes,achieving annotations for all points.Experiments demonstrate that the proposed method excels in practical applications.By grouping point cloud data with the DBSCAN algorithm,it effectively reduces the amount of data that the algorithm needs to process,thereby enhancing processing efficiency.Meanwhile,the semi-supervised GCN model,leveraging its robust contextual analysis capability,maintains high classification accuracy even with limited annotated data.In simple scenarios,the performance of the proposed method is comparable to that of the Pointnet++model;in complex environments,it still sustains performance close to Pointnet++.关键词
车载点云/点云分类/图卷积网络/标注样本Key words
vehicle-mounted point cloud/point cloud classification/graph convolutional network/annotated sample分类
天文与地球科学引用本文复制引用
张金帅,杨俊康..基于图卷积神经网络的车载激光点云分类[J].北京测绘,2025,39(6):824-829,6.基金项目
浙江省自然科学基金(LTGG23D010001) （LTGG23D010001）
