西北林学院学报2024，Vol.39Issue(2)：28-35,8.DOI:10.3969/j.issn.1001-7461.2024.02.04
基于多尺度几何特征单木点云的语义分割
Semantic Segmentation of Individual Tree Point Clouds Based on Multi-scale Geometric Features
摘要
Abstract
It is difficult to accurately separate limb and leaf points from terrestrial laser scanning(TLS)tree point clouds.To solve the problem a method of semantic segmentation of individual tree point clouds based on multi-scale geometry features was proposed.First,the point cloud data of trees in the sample plots of Fraxinus mandshurica and Pinus sylvestris were segmented into individual trees,and the eigenvalues of the multi-scale point cloud covariance matrix were calculated,then the feature classifier was selected,and the optimal eigenvalues were selected according to the importance of the eigenvalues,and finally the limb and leaf points of tree were divided.By comparing the training time and accuracy of support vector machine(SVM),extreme gradient boosting(XGBoost)and random forest(RF)classifiers,the XGBoost was se-lected as the final classifier,and six optimal features were selected according to the importance of eigenval-ues.The results showed that the segmentation accuracy of limb and leaf in the two plots was more than 0.88,and Fl-Score and IOU(intersection over union)were also above 0.8.The method proposed in this study can effectively segment limb and leaf points of F.mandshurica and P.sylvestris,and has a high i-dentification accuracy.The research results provide conditions for the subsequent three-dimensional model construction and biomass estimation of individual tree.关键词
地面激光扫描/点云数据/多尺度/几何特征/语义分割Key words
terrestrial laser scanning/point cloud data/multiple scale/geometric feature/semantic seg-mentation分类
农业科技引用本文复制引用
曹荣贞,刘浩然,林文树..基于多尺度几何特征单木点云的语义分割[J].西北林学院学报,2024,39(2):28-35,8.基金项目
国家自然科学基金(31971574). （31971574）
