机载LIDAR点云数据的分类方法比较分析.pdfVIP

第48卷第9期测绘与空间地理信息Vol.48,No.9

2025年9月GEOMATICSSPATIALINFORMATIONTECHNOLOGYSept.,2025

机载LIDAR点云数据的分类方法比较分析

王琦12，尚金光12，邹仁均12，喻燕萍1.2

（1.成都市勘察测绘研究院，四川成都610081；

2.成都城市地下空间勘察和数字化测绘工程技术研究中心，四川成都610081)

摘要：机载点云数据精确分类在基础测绘、实景三维、公园城市建设等领域需求较高。基于RF、SVM、CCN、PointNet

和PointNet++5种算法，对比分析了几种算法的点云分类结果，并探讨了几种算法的差异。结果表明,PointNet++的分

类效果最好，其次是PointNet、RF,SVM算法和GCN的效果较差；在分类类别较少的情况下,深度学习方法分类结果较

机器学习方法结果无明显优势。研究结果可为生产项目中点云分类算法的选择提供参考。

关键词：点云数据；分类；机器学习；深度学习

中图分类号:P228文献标识码：A文章编号：1672-5867(2025)09-0007-03

ComparativeAnalysisofClassificationMethodsfor

AirborneLiDARPointCloudData

WANGQire,SHANGJinguang,ZOURenjun,YUYnpinga

(1.ChengduInstituteofSurveyingandInvestigation,Chengdu610081,China;2.EngineeringTechnologyResearchCenterfor

UrbanUndergroundSpaceSurveyandDigitalSurveyingandMappingofChengdu,Chengdu610081,China)

Abstract:Theaccurateclassificationofairbornepointclouddatahasastrongdemandinthefieldsofbasicsurveyingandmapping,realistic

3D,andparkcityconstruction.Basedonfivealgorithms,RF,SVM,GCN,PointNet,andPointNet++,thepointcloudclassificationre-

sultsofseveralalgorithmswerecomparedandanalyzed,andthedifferencesbetweentheseveralalgorithmswerediscussed.Theresultsshow

thatPointNet++hasthebestclassificationeffect,followedbyPointNet,RF,SVMalgorithm,andGCNwithpoorperformance.Inthecase

offewerclassificationcategories,theclassificationresultsofdeeplearningmethodshavenoobviousadvantagesoverthoseofmachinelearn-

ingmethods.Theresearchresultscanprovideareferencefortheselectionofpointcloudclassificationalgorithmsinproductionprojects.

Keywords:pointclouddat