1 Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou 730070, China
2 National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou 730070, China
3 Key Laboratory of Science and Technology in Surveying & Mapping, Gansu Province (Lanzhou Jiaotong University), Lanzhou 730070, China
Show less
文章历史+
Received
Published
Revised
2025-08-09
2026-06-08
2025-11-28
Online Date
2026-08-07
PDF (19894K)
摘要
单木分割是获取森林单木信息的关键技术,高精度分割森林单木是精细化计算森林资源的基础。在非均匀分布和高郁闭度的梯度郁闭林中,仅依靠经验设置参数阈值的聚类方法难以实现高精度单木分割,严重影响森林资源精细化管理。为此,提出了一种基于自适应半径DBSCAN(density-based spatial clustering of applications with noise)的梯度郁闭林单木点云分割方法。首先,对预处理后的点云数据作z轴切片,进行初步DBSCAN聚类,计算各子簇的水平尺度,为后续聚类参数自适应提供依据。其次,通过剔除子簇尺度极值并计算中间尺度值均值,自适应确定每一切片的聚类半径,以适应不同高度层级的结构差异。最后,结合自适应聚类半径、核心点识别与基于k维树(k-dimensional tree, KD-Tree)的邻域搜索策略,实现单木点云的准确分割。结果表明:所提出的方法应用在5个梯度郁闭林样地单木分割的平均召回率和精确率分别达到0.94和0.99,相较于现有方法具有明显优势。该方法在梯度郁闭度复杂森林环境下具有良好的通用性和鲁棒性,本研究对于提高森林资源调查的精度和效率具有重要意义。
Abstract
Accurate single-tree segmentation plays a crucial role in obtaining detailed individual tree information, which underpins the precise estimation and sustainable management of forest resources. In forests with gradient canopy closure and non-uniform tree distribution, clustering methods that rely solely on empirically set parameters often fail to achieve accurate single-tree segmentation, which significantly constrains refined forest resource management. To address this challenge, an adaptive-radius density-based spatial clustering of applications with noise (DBSCAN) algorithm was proposed for single-tree point cloud segmentation in forests with gradient canopy closure.First, the preprocessed point cloud data were sliced along the z-axis and initially clustered using DBSCAN. The horizontal scale of each sub-cluster was analyzed to guide the adaptive adjustment of clustering in the subsequent steps. Second, removing extreme scale values, the mean of the remaining scales was used to determine an adaptive clustering radius for each slice, accommodating structural variations across different canopy heights. Finally, by integrating the adaptive clustering radius, core point identification, and a k-dimensional tree (KD-Tree)-based neighborhood search strategy, the accurate segmentation of individual tree point clouds was achieved. Experimental results show that the proposed method achieves an average recall of 0.94 and precision of 0.99 across five forest plots, outperforming existing segmentation techniques. The proposed method exhibits strong generalization capability and robustness in complex gradient canopy environments, providing a valuable tool for improving the accuracy and efficiency of forest resource surveys.
SHANG Jia-hao, KOU Rui-xiong, SHI Zhuang, JIN Xue-rui, QIN Shi-zhuo.
Single-tree Point Cloud Segmentation Method for Gradient Canopy Closure Forests Based on Adaptive Radius DBSCAN[J].
Science Technology and Engineering, 2026, 26(16): 6712-6719 DOI:10.12404/j.issn.1671-1815.2505840
目前,单木分割方法主要包括实地测量方法、基于图像和点云的方法。其中,实地测量方法能准确获取单木信息,但存在人力成本过高和测量效率低下等问题,难以满足大范围森林资源调查的需求[5]。基于图像的单木分割方法是通过处理航片和卫片实现单木分割。相较之下,激光雷达(light detection and ranging, LiDAR)[6]凭借其精确的测距能力和良好的树冠穿透性[7-8],对森林形态的表达显著优于图像。为此,基于激光雷达点云的森林单木分割技术已经成为森林资源调查和生态监测中的关键技术[9]。
在基于点云的单木分割方面,针对不同林型和树种特征,不同学者提出了多种单木分割策略。郭继富等[10]融合核密度估计、利用点云生成的数字表面模型(digital surface model, DSM)与K-means聚类,实现了西北云杉林的单木分割。李阳等[11]提出一种改进的K-means聚类算法进行单木分割,构建了适用于人工针阔混交林的分割模型。Kaisen等[12]则采用层次聚类对中南林业科技大学芦头实验林场的7 块人工林,2 块天然林进行了有效的单木提取。Lino等[13]提出一种基于DBSCAN聚类和圆柱体体素化的单木分割方法,该方法在Baixa Limia-Serra do Xurés自然公园的树木分割实验中取得了约90%的分割率。郁闭度是森林中乔木树冠在阳光直射下在地面的总投影面积(冠幅)与此林地(林分)总面积的比值,它是反映林分密度的指标,上述聚类方法均针对单一郁闭度样地进行分割,使得这些方法在大范围梯度郁闭林中存在一定局限性:一是预设聚类数或聚类半径对分类结果影响较大,增加模型对初始参数的依赖性;二是在不同郁闭度条件下分割精度波动较大,难以稳定适应梯度郁闭度的复杂森林环境[14]。
为评估所提方法的有效性,将其与基于距离判别的树木分割方法(tree segmentation based on distance discrimination, TSDD)和基于固定聚类半径的DBSCAN方法(fixed-radius DBSCAN, FD)进行对比,基于距离判别的方法[19]通过归一化点云数据,选取局部极大值点作为种子点,并依据点与种子点的距离关系进行树木分割。固定聚类半径的DBSCAN方法[20]通过设定固定的聚类半径和最小邻域点数,利用密度聚类的思想将点云数据划分为不同的树木个体。
SiepielskiM A, MorrisseyB M, BuoroM, et al. Response to comment on precipitation drives global variation in natural selection[J]. Science, 2018, 359(6374): e5760.
WeiXiaobo, LiZhiqiang, ChenZhaoguang, et al. Study on single tree segmentation of mangrove trees of different species based on UAV LiDAR[J]. Science Technology and Engineering, 2024, 24(3): 963-969.
[4]
ZhaoK G, SuarezC J, GarciaM, et al. Utility of multitemporal lidar for forest and carbon monitoring: tree growth, biomass dynamics, and carbon flux[J]. Remote Sensing of Environment, 2018, 204: 883-897.
WangLiangsong, LiNing, WangCheng, et al. Effect of canopy morphological characteristics on individual tree biomass estimation based on airborne LiDAR[J]. Science Technology and Engineering, 2024, 24(31): 13304-13311.
ZhuNingning, YangBisheng, DongZhen. The explicit model of forest carbon storage based on remote sensing[J]. Acta Geodaetica et Cartographica Sinica, 2024, 53(1): 36-49.
WuXiaokang, WangHaoyu, FengBaokun, et al. Individual tree biomass estimation based on UAV airborne LiDAR[J]. Science Technology and Engineering, 2022, 22(34): 15028-15035.
LiuQing, ZhuNingning, YuWandong, et al. A method for automatic registration of unmanned aerial vehicle and terrestrial laser scanning point clouds in forest areas[J]. Geospatial Information, 2022, 20(5): 96-101.
[13]
DaiW X, YangB S, DongZ, et al. A new method for 3D individual tree extraction using multispectral airborne LiDAR point clouds[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 144: 400-411.
[14]
ChenX Y, YuK Y, YuS H, et al. Study on single-tree segmentation of Chinese fir plantations using coupled local maximum and height-weighted improved K-means algorithm[J]. Forests, 2023, 14(11): 2130.
GuoJifu, SunJianyu, HouJinliang, et al. Research on single-tree segmentation algorithm for Northwest spruce based on UAV-LiDAR point cloud data[J]. Remote Sensing Technology and Application, 2025, 40(1): 156-166.
LiYang, XuHongbo, LingChengxing, et al. Research on the method of individual tree segmentation and individual tree volume estimation using backpack LiDAR[J]. Remote Sensing Technology and Application, 2025, 40(3): 568-581.
[19]
KaisenM, ChenZ X, FuL Y, et al. Performance and sensitivity of individual tree segmentation methods for UAV-LiDAR in multiple forest types[J]. Remote Sensing, 2022, 14(2): 298.
[20]
LinoC, JoaquínM, HenriqueL, et al. Individual tree segmentation method based on mobile backpack LiDAR point clouds[J]. Sensors, 2021, 21(18): 6007.
ZhuBodong, LuoHongbin, JinJing, et al. Optimization of individual tree segmentation methods for high canopy density plantation based on UAV LiDAR[J]. Scientia Silvae Sinicae, 2022, 58(9): 48-59.
[23]
ZhengJ J, XiangM, ZhangT, et al. An improved adaptive grid-based progressive triangulated irregular network densification algorithm for filtering Airborne LiDAR Data[J]. Remote Sensing, 2024, 16(20): 3846.
[24]
YunT, JiangK, LiG C, et al. Individual tree crown segmentation from airborne LiDAR data using a novel Gaussian filter and energy function minimization-based approach[J]. Remote Sensing of Environment, 2021, 256: 112307.
[25]
XueJ K, ShenB. A novel swarm intelligence optimization approach: sparrow search algorithm[J]. Systems Science & Control Engineering, 2020, 8(1): 22-34.
[26]
YinH, ZhangW, LuC W, et al. Rapid characterization of high-steep slope rock mass integrity using fractal dimension of thermal infrared point cloud[J]. Engineering Geology, 2025, 354: 108161.
KongDan, PangYong, LiangXiaojun, et al. Individual tree segmentation of airborne LiDAR point cloud based on layer-stacking method[J]. Scientia Silvae Sinicae, 2024, 60(3): 87-99.
[29]
FuH P, LiH, DongY Q, et al. Segmenting individual tree from TLS point clouds using improved DBSCAN[J]. Forests, 2022, 13(4): 566.