基于机载激光雷达技术的山区针叶林单木分割方法研究  

Research on a Method for Individual Tree Segmentation for Mountainous Coniferous Forests Using UAV-LiDAR Technology

作　　者：黄冰倩 曹霸 岳彩荣[2] 周全 HUANG Bingqian;CAO Ba;YUE Cairong;ZHOU Quan(Forestry Survey and Planning Institute of Guizhou Province,Guiyang 550003,Guizhou,China;Southwest Forestry University,Kunming 650224,Yunnan,China;Central South Academy of Inventory and Planning of NFGA,Changsha 410014,Hunan,China)

机构地区：[1]贵州省林业调查规划院,贵州贵阳550003 [2]西南林业大学,云南昆明650224 [3]国家林业和草原局中南调查规划院,湖南长沙410014

出　　处：《中南林业调查规划》2024年第3期34-39,48,共7页Central South Forest Inventory and Planning

基　　金：基于激光雷达技术的马尾松森林结构参数估测研究(黔林科合[2022]37号)。

摘　　要：以杉木为研究对象,利用无人机激光雷达点云数据,结合地面实测样地数据,探索创新现有的森林资源调查模式,提升森林资源外业调查效率,更新资源调查数据,保证数据的现势性。考虑到研究区为山地林区,地面起伏较大,选取了一种基于改进加密三角网滤波算法进行滤波和分类,并对比分析了分水岭、点云分割、层堆叠种子点三种不同的单木分割方法,完成了研究区单木位置、树高、冠幅等林木参数的提取研究,优化了激光雷达林木参数反演的技术流程。从30个样地中抽取10个样地,对比样木实测值与单木分割的估测值,结果表明:层堆叠种子点算法分割效果最优,F评分在64.61%~85.29%之间;点云分割算法居中,F评分在56.00%~80.60%之间;分水岭算法分割效果最差,F评分在45.57%~69.45%之间。同一种方法中,不同样地分割效果也存在差异,这可能与样地地形、树木结构形态等因素有关;样地中林木分布存在树木遮挡或树木分布结构不规则时,一定程度会降低单木分割精度。因此,根据不同林分情况,建立适用性较强的无人机激光雷达森林参数反演模型是未来努力的方向。The Chinese fir(Cunninghamia lanceolata)was taken as the research object.Unmanned aerial vehicle LiDAR(UAV-LiDAR)point cloud data was combined with field survey data which was used to explore and innovate the existing forest resource survey model,improve the efficiency of forest resource field survey,update the resource survey data,ensure the current status of the data.Considering that the study area is mountainous with undulating terrain,an improved triangulated irregular network densification filtering algorithm was selected for filtering and classification.The methods for individual tree segmentation including watershed segmentation,point cloud segmentation,and layer stacking seed point segmentation algorithm were compared,and the extraction of tree parameters including individual tree position,tree height,and crown width in the study area was completed,and the technical process of UAV-LiDAR on the inversion of forest parameters was optimized.Ten plots were selected from 30 plots,The measured value of sample wood was compared with the estimated value of individual tree segmentation of these plots.The results were as follows:The layer stacking seed point algorithm had the best segmentation effect;the F-score of the individual tree segmentation algorithm ranged from 64.61%to 85.29%,the point cloud segmentation algorithm was moderate,its F-score ranged from 56.00%to 80.60%,and the watershed segmentation algorithm which F-score ranged from 45.57%to 69.45%had a slightly poorer segmentation effect.Within the same method,there were differences in segmentation effects in different plots,which might be related to plot terrain and tree structure morphology.When there were tree occlusions or irregular tree distribution structures in the plot,the individual tree segmentation accuracy would be reduced to a certain extent.Therefore,establishing a forest parameter inversion model suitable for different forest conditions based on UAV-LiDAR data is the direction for future efforts.

关 键 词：激光雷达 单木分割方法 滤波 杉木 森林结构参数 

分 类 号：S771[农业科学—森林工程]