摘要
【目的】提出一种基于分层叠加的单木分割算法,以充分利用高密度激光雷达点云信息,提高林分中下层单木分割精度。【方法】区别于传统将冠顶点作为聚类种子点的单木分割算法,基于分层叠加的单木分割算法以点云水平切片后各层的局部最大值为种子点进行分层聚类,并通过分层叠加与迭代优化,减少枝杈等因素导致的过分割现象,在保证上层树单木分割精度的同时提高对中下层单木的提取能力。【结果】基于分层叠加的单木分割算法在不同密度落叶松林分均有较高单木分割精度,提取单木与实测单木总体匹配成功率最高达94%,在中高密度林分匹配成功率最高达92%,相较其他算法,对中下层单木的匹配率可提高20%~40%;在单木树高提取精度方面,单木提取树高与实测树高相关系数为0.8,相对均方根误差为8.45%,提取冠幅与实测冠幅相关系数最高为0.83,相对均方根误差为16.5%。【结论】通过分层聚类、聚类种子点优化选取,充分利用林分各层次点云信息,可提高单木分割精度,为森林经营管理提供高精度数据支持。
【Objective】This paper proposed an individual tree segmentation algorithm utilizing hierarchical layer stacking approach to optimize the use of high-density LiDAR point cloud data,thereby improving the accuracy of individual tree segmentation in the understory of forest stands.【Method】Diverging from traditional algorithms which utilize canopy vertices as cluster seeds,this hierarchical overlay-based segmentation algorithm selects local maxima of each layer after horizontal slicing of point clouds for tiered clustering.It diminishes the over-segmentation due to branches through layered overlay and iterative refinement,securing segmentation precision of canopy trees and boosting extraction of understory trees.【Result】The tree segmentation algorithm based on layer stacking exhibits high precision in larch stands of various stem densities,with a maximum matching success rate of 94%between extracted and observed trees,and up to 92%in medium to high density stands.Compared to other algorithms,the matching rate for mid and lower-layer trees can be improved by 20%to 40%.In terms of individual tree height extraction precision,the correlation coefficient between extracted and observed tree heights is 0.8,with a relative root mean square error of 8.45%.The highest correlation coefficient between extracted and observed crown widths is 0.83,with a relative root mean square error of 16.5%.【Conclusion】By stacking hierarchical clustering and optimizing seed point selection,the comprehensive use of point cloud data across forest layers enhances individual tree segmentation accuracy,providing valuable data support for forest management and operations.
作者
孔丹
庞勇
梁晓军
杜黎明
白羽
Kong Dan;Pang Yong;Liang Xiaojun;Du Liming;Bai Yu(Institute of Forest Resource Information Techniques,Chinese Academy of Forestry Key Laboratory of Forestry Remote Sensing and Information System,National Forestry and Grassland Administration Beijing 100091)
出处
《林业科学》
EI
CAS
CSCD
北大核心
2024年第3期87-99,共13页
Scientia Silvae Sinicae
基金
“十四五”和“十三五”国家重点研发项(2023YFD2200804,2017YFD0600404)。