摘要
【目的】针对传统机载激光雷达(Light Detection And Ranging,LiDAR)所生成的冠层高度模型分辨率较低,不利于高郁闭度人工针叶林中冠幅较小的树种单木分割的问题,基于大疆禅思L1激光雷达高密度点云,提出了一种基于层次化泛洪的单木分割算法。【方法】采用大疆禅思L1激光雷达设备,选择东北地区樟子松与兴安落叶松人工林作为实验样地。首先对原始的高密度激光雷达点云依次进行拼接、去噪、高程归一化处理,针对两块样地生成分辨率分别为0.1、0.2、0.4 m的高、中、低3种分辨率的冠层高度模型。对3种冠层高度模型分别进行先开后闭的形态学滤波处理,以降低冠层高度模型中单个树冠内部不同像素间高程差。然后采用反距离插值算法对3种冠层高度模型中的空洞像素进行平滑处理,并采用图像增强算法提高3种冠层高度模型中树冠内部像素与树冠间像素的对比度,以降低形态学滤波对林隙的模糊处理影响。最后利用局部最大值法分别在3种冠层高度模型中搜索树顶,基于搜索到的树顶位置,结合分层处理思想通过模拟泛洪算法实现单木分割,并基于一般树冠形态,对分割后树冠投影形状、面积进行约束,以优化分割后树冠形状。【结果】针对人工针叶林林分,提出的单木分割算法结合高分辨率冠层高度模型在两种树种样地下最高分割精度达到90%以上。其中,在冠幅较小的兴安落叶松样地中,基于高、中、低分辨率冠层高度模型的单木分割精度F值分别达到91.6%、85.9%、80.2%。而冠幅较大的樟子松样地中,基于高、中、低分辨率冠层高度模型的单木分割精度F值分别为86.2%、84.1%、75.9%。【结论】基于冠层高度模型的单木分割场景中,冠幅较大的树种对于一定范围内分辨率变化不敏感,高分辨率冠层高度模型可以提高人工针叶林单木分割精度,尤其是对于冠幅较小的树种分割精度提高较大。本研究提出的单木分割方法结合无人机高密度LiDAR点云在高郁闭度人工针叶林样地中可达到较高分割精度。
【Objective】Given the low resolution of the canopy height model(CHM) based on the conventional airborne LiDAR point cloud, which is detrimental to the single tree segmentation of tree species with small canopy, a single tree segmentation algorithm based on hierarchical flooding was proposed in this paper.【Method】In this study, the DJI L1 UAV LiDAR was used to observe Pinus sylvestris and Larix gmelinii plots. Firstly, the raw point cloud data were denoised, merged and normalized to generate three CHMs with high, medium and low resolution. Secondly, the morphological filter was used to smooth the CHM image to reduce height differences of the single canopy in the high-resolution CHM. Then, the IDW method was used to smooth the CHM, and the image enhancement method was applied to reduce the bad influence of morphological filtering on forest gaps. Subsequently, the tree tops were searched with enhanced CHM images. Finally, the single tree segmentation was realized based on the tree tops through the simulated flooding algorithm. In addition, based on the general canopy shape, the projected shape and area of the segmented canopy were constrained to optimize the segmented canopy shape.【Result】For coniferous plantations, the proposed single-tree segmentation algorithm combined with a high-resolution canopy height model achieved the highest segmentation accuracy of more than 90% in the plots of the two tree species. In the Larix gmelinii plots with a small canopy, the F score of individual tree segmentation was up to 91.6% based on highresolution CHM, 85.9% based on medium-resolution CHM and 80.2% based on low-resolution CHM. In the Pinus sylvestris plots, the F scores were 86.2% and 84.1% based on the high and medium-resolution CHM, respectively, which was 75.9% based on low-resolution CHM.【Conclusion】In the analysis of single tree segmentation based on CHM, trees with a big canopy are insensitive to resolution changes within a certain range. High-resolution CHM can improve the segmentation accuracy of small-canopy trees. The single tree segmentation method proposed in this study combined with the high-density LiDAR point cloud of UAV can achieve high segmentation accuracy in coniferous plantations with high canopy closure.
作者
王鑫运
黄杨
邢艳秋
李德江
赵晓伟
WANG Xinyun;HUANG Yang;XING Yanqiu;LI Dejiang;ZHAO Xiaowei(Center for Research Institute of Forest Operations and Environment,Northeast Forest University,Harbin 150040,Heilongjiang,China;Heilongjiang Provincial Research Institute of Surveying and Mapping,Harbin 150040,Heilongjiang,China)
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2022年第8期66-77,共12页
Journal of Central South University of Forestry & Technology
基金
国家重点研发计划项目(2021YFE0117700-6)。
关键词
机载激光雷达
高密度点云
图像增强
单木分割
层次泛洪
airborne LiDAR
high-density point cloud
image enhancement
individual tree segmentation
simulation of flooding layer by layer
