摘要
为了准确高效提取人工林结构参数,以晋西黄土区蔡家川流域人工油松林为研究对象,利用30块样地的激光雷达点云数据和样地实测数据,通过改变点云距离判别聚类算法的格网值和调整分水岭算法的冠层高度分辨率的方法,对比分析关键参数对单木分割的敏感性,探求点云距离判别聚类算法和分水岭算法对树高提取精度的最优参量。结果表明:(1)点云距离判别聚类算法单木提取的召回率为87.3%、准确率为86.0%、调和值为86.7%,优于分水岭算法(召回率为83.0%、准确率为83.8%、调和值为83.4%)。(2)点云距离判别聚类算法分割单木的敏感性,采用最小冠幅1/5的格网值,其召回率为87.3%、准确率为86.0%、调和值为86.7%,分割精度最高。分水岭算法分割单木的最优关键参量随林分密度不同而变化,当林分密度≤3600株/hm^(2),采用冠层高度分辨率0.3 m时,分割效果最优,其召回率为78.9%、准确率为85.2%、调和值为81.9%;当林分密度≥3700株/hm^(2),采用冠层高度分辨率0.2 m时,分割效果最优,召回率为87.2%、准确率为82.5%、调和值为84.8%。(3)分水岭算法提取树高精度(决定系数为0.88,均方根误差为0.93 m)优于点云距离判别聚类算法。
In order to accurately and efficiently extract the structural parameters of plantations,the Pinus tabulaeformis plantations in the Caijiachuan River Basin of the loess region of western Shanxi Province was chosen as the research object,and LiDAR point cloud data and field measurement data from 30 sample plots were used.By changing the grid value of the point cloud distance discriminative clustering algorithm and adjusting the canopy height resolution of the watershed algorithm,the sensitivity of key parameters to single tree segmentation was compared and analyzed.The optimal parameters of the point cloud distance discriminative clustering algorithm and the watershed algorithm for tree height extraction accuracy were explored.The results showed that:(1)The recall rate,precision,and F1-score of single tree extraction using the point cloud distance discriminative clustering algorithm are 87.3%,86.0%,and 86.7%,respectively,which is superior to the watershed algorithm(recall rate was 83.0%,precision was 83.8%,F1-score was 83.4%).(2)The sensitivity of single tree segmentation using the point cloud distance discriminative clustering algorithm is highest when using a grid value of 1/5 of the minimum canopy diameter,with recall rate,precision,and F1-score of 87.3%,86.0%,and 86.7%,respectively.The optimal key parameter for single tree segmentation using the watershed algorithm varies with forest stand density.When the forest stand density is≤3600 trees/hm^(2),the optimal canopy height resolution is 0.3 m,with a recall rate of 78.9%,precision of 85.2%,and F1-score of 81.9%.When the forest stand density is≥3700 trees/hm^(2),the optimal canopy height resolution is 0.2 m,with a recall rate of 87.2%,precision of 82.5%,and F1-score of 84.8%.(3)The watershed algorithm achieves higher tree height extraction accuracy(coefficient of determination was 0.88,root mean square error was 0.93 m)compared to the point cloud distance discriminative clustering algorithm.
作者
张燕妮
张学霞
张建军
程家琪
胡亚伟
赵炯昌
李阳
杨锐
Zhang Yanni;Zhang Xuexia;Zhang Jianjun;Cheng Jiaqi;Hu Yawei;Zhao Jiongchang;Li Yang;Yang Rui(Beijing Forestry University,Beijing 100083,P.R.China)
出处
《东北林业大学学报》
CAS
CSCD
北大核心
2024年第7期36-43,共8页
Journal of Northeast Forestry University
基金
国家重点研发计划项目(SQ2021YFE012026)。
关键词
机载激光雷达
单木分割
结构参数提取
CHM分辨率
格网值
Airborne LiDAR
Single tree segmentation
Structural parameter extraction
CHM resolution ratio
Grid value
