机载激光点云密度对单木分割精度的影响  

The influence of airborne laser point cloud density on individual tree segmentation accuracy

作　　者：贾越 夏永华[2,3] 赵昌福 伍福万 赵曲皑 王帅 JIA Yue;XIA Yong-hua;ZHAO Chang-fu;WU Fu-wan;ZHAO Qu-ai;WANG Shuai(College of Land and Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;Application Engineering Research Center of Spatial Information Surveying and Mapping Technology in Plateau Mountainous Areas,Yunnan Universities,Kunming University of Science and Technology,Kunming 650093,China;College of City,Kunming University of Science and Technology,Kunming 650051,China;Yunnan Phosphating Group Co.,Ltd.,Kunming 650600,China;Yunnan Geological Engineering Second Survey Institute Co.,Ltd.,Kunming 650032,China)

机构地区：[1]昆明理工大学国土资源工程学院,昆明650093 [2]昆明理工大学云南高校高原山区空间信息测绘技术应用工程研究中心,昆明650093 [3]昆明理工大学城市学院,昆明650051 [4]云南磷化集团有限公司,昆明650600 [5]云南地质工程第二勘察院有限公司,昆明650032

出　　处：《兰州大学学报(自然科学版)》2025年第2期215-221,共7页Journal of Lanzhou University(Natural Sciences)

基　　金：国家自然科学基金项目(41861054);云南地质工程第二勘察院项目(KKF0201856026);昆明市城市地下空间规划管理办公室技术服务项目(KKF0201956004)。

摘　　要：点云密度是影响单木分割的重要因素之一,为了探究不同点云密度下单木分割方法的适用性和性能表现,针对阔叶林和针叶林两个不同类型的森林环境,采用一种非均匀最远点(NFPS)采样方法与传统的体素下采样方法,将样地点云数据重采样至5个不同的密度等级,获取不同点云密度数据集,即原始点云数量的100%、 50%、 25%、 12%、 6%,以点云分割(PCS)和冠层高度模型(CHM)算法,对森林中的独立树木进行有效分割.评估单木分割在不同密度下的精度时,使用精确率、召回率和F分数等指标.结果表明,NFPS采样方法在不同密度点云中的分割精度均高于体素下采样方法.当点云密度为113点/m2时,两种分割方法的3个评价指标均表现最好.阔叶林样地中NFPSCHM和NFPS-PCS的F分数值分别为88%和84%;针叶林样地中NFPS-PCS与NFPS-CHM的F分数值分别为82%和65%.NFPS采样方法较传统采样方法对单木分割精度有较好的效果,适度降低点云密度的同时也让单木分割算法精度得到了保证.Point cloud density is one of the important factors affecting individual tree segmentation.To explore the applicability and performance of individual tree segmentation methods under different point cloud densities.An non-uniform farthest point sampling(NFPS) downsampling method and voxel downsampling method were proposed for two different types of forest environments,broad-leaved forest and coniferous forest.The sampling point cloud data was resampled to five different density levels to obtain different point cloud density datasets,namely 100%,50%,25%,12%,and 6% of the original point cloud quantity.Two segmentation methods,point cloud segmentation(PCS) algorithm and canopy height mode(CHM) algorithm,were used to effectively segment independent trees in the forest.In evaluation of the accuracy of individual tree segmentation at different densities,indicators such as precision,recall,and F-score were used.The results showed that the proposed NFPS sampling method had higher segmentation accuracy than voxel downsampling method in point clouds of different densities.When the point cloud density was 113 points/m2,both segmentation methods performed best in all three evaluation metrics.The F-score of NFPS-CHM and NFPS-PCS in the broad-leaved forest plot reached 88% and 84%,respectively;The F-scores of NFPS-PCS and NFPS-CHM in coniferous forest plots were 82% and 65%,respectively.The sampling method had a better effect on the accuracy of individual tree segmentation compared to traditional sampling methods.While moderately reducing the point cloud density,it also ensured the accuracy of the individual tree segmentation algorithm.

关 键 词：机载激光点云 体素下采样 非均匀最远点采样 点云密度 单木分割 

分 类 号：P258[天文地球—测绘科学与技术]