CN109961470B - An accurate estimation method of living tree leaf attributes based on laser point cloud - Google Patents
An accurate estimation method of living tree leaf attributes based on laser point cloudDownload PDFInfo
- Publication number
- CN109961470B CN109961470BCN201910130528.3ACN201910130528ACN109961470BCN 109961470 BCN109961470 BCN 109961470BCN 201910130528 ACN201910130528 ACN 201910130528ACN 109961470 BCN109961470 BCN 109961470B
- Authority
- CN
- China
- Prior art keywords
- leaf
- point
- point cloud
- points
- tree
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Software Systems (AREA)
- Length Measuring Devices By Optical Means (AREA)
- Image Analysis (AREA)
Abstract
Description
Translated fromChinese技术领域Technical Field
本发明属于活立木叶属性分析技术领域，具体涉及一种基于激光点云的活立木叶属性精准估测方法。The present invention belongs to the technical field of standing tree leaf attribute analysis, and in particular relates to a method for accurately estimating standing tree leaf attributes based on laser point cloud.
背景技术Background Art
现有的叶片属性包括叶面积计算，叶倾角和方位角，叶面积计算，叶倾角和方位角的估测方法之前主要采用人工现场采集数据的方法。此方法工作量大，耗费时间长，并且不能精确的得到树冠每个单片叶子的叶片属性。尽管基于地面激光扫描数据对树叶属性估计进行了许多研究，但是相关技术还是不成熟，因此仍需设计一种基于计算机图形的算法，以准确地从大量的点云中提取单个叶子的叶片属性，进而来替代人工现场采集数据方法。Existing leaf attributes include leaf area calculation, leaf inclination and azimuth. The leaf area calculation, leaf inclination and azimuth estimation methods mainly use the method of manually collecting data on site. This method is labor-intensive, time-consuming, and cannot accurately obtain the leaf attributes of each single leaf in the crown. Although many studies have been conducted on leaf attribute estimation based on terrestrial laser scanning data, the relevant technology is still immature. Therefore, it is still necessary to design an algorithm based on computer graphics to accurately extract the leaf attributes of a single leaf from a large number of point clouds, thereby replacing the manual on-site data collection method.
发明内容Summary of the invention
本发明所要解决的技术问题是针对上述现有技术的不足提供一种基于激光点云的活立木叶属性精准估测方法，本基于激光点云的活立木叶属性精准估测方法可以准确得到树冠每个单片叶子的叶片属性，减少人工劳动工作量，效率高。The technical problem to be solved by the present invention is to provide a method for accurately estimating the leaf attributes of standing trees based on laser point clouds in response to the deficiencies of the above-mentioned prior art. The method for accurately estimating the leaf attributes of standing trees based on laser point clouds can accurately obtain the leaf attributes of each single leaf in the crown, reduce the workload of manual labor, and has high efficiency.
为实现上述技术目的，本发明采取的技术方案为：In order to achieve the above technical objectives, the technical solution adopted by the present invention is:
一种基于激光点云的活立木叶属性精准估测方法，包括以下步骤：A method for accurately estimating leaf attributes of standing trees based on laser point cloud comprises the following steps:
(1)获取树木点云数据，对树木点云数据进行处理，实现枝叶分离；(1) Obtaining tree point cloud data, processing the tree point cloud data, and realizing branch and leaf separation;
(2)通过具有自适应半径的球体邻域模型和四个辅助准则提取叶子的中心区域点；(2) Extracting the central area points of the leaves through a spherical neighborhood model with an adaptive radius and four auxiliary criteria;
(3)通过基于密度的空间聚类算法聚类叶子的中心区域点，获得每个叶子表面的中心点；(3) Clustering the central area points of the leaves through a density-based spatial clustering algorithm to obtain the central point of each leaf surface;
(4)通过每个叶子表面的中心点和三维分水岭算法获取每个叶子表面的非中心区域点，从而实现单片叶子点云的分割；(4) The non-central area points of each leaf surface are obtained through the center point of each leaf surface and the three-dimensional watershed algorithm, thereby realizing the segmentation of a single leaf point cloud;
(5)采用Delaunay三角剖分面向单片叶子点云推导出每片叶子的面积；(5) Delaunay triangulation is used to derive the area of each leaf from the point cloud of a single leaf;
(6)对单片叶子点云，通过用最小二乘法拟合叶片平面从而获得叶子表面的法向矢量，计算叶子表面的法向矢量与天顶角之间的夹角，该夹角为对应叶子的叶片倾角；通过该步骤的方法计算出每个叶子对应的叶片倾角；(6) For a single leaf point cloud, the leaf plane is fitted using the least squares method to obtain the normal vector of the leaf surface, and the angle between the normal vector of the leaf surface and the zenith angle is calculated. The angle is the leaf inclination angle of the corresponding leaf. The leaf inclination angle corresponding to each leaf is calculated by the method of this step;
(7)计算水平面的北方向与叶子主轴线在水平面上的投影之间的顺时针角度，该顺时针角度即为对应叶子的叶片方位角，其中叶子主轴线为对应叶子的单片叶子点云中最远两点之间的线；通过该步骤的方法计算出每个叶子对应的叶片方位角。(7) Calculate the clockwise angle between the north direction of the horizontal plane and the projection of the main axis of the leaf on the horizontal plane. The clockwise angle is the leaf azimuth of the corresponding leaf, where the main axis of the leaf is the line between the two farthest points in the single leaf point cloud of the corresponding leaf. Calculate the leaf azimuth corresponding to each leaf through the method of this step.
作为本发明进一步改进的技术方案，所述的步骤2具体为：As a further improved technical solution of the present invention, the step 2 is specifically as follows:
(1)枝叶分离后得到的叶子点云集合P中的任意点云pi的坐标为pi(xi，yi，zi)，将点云pi的邻域点pi，j的坐标定义为pi，j(xi，j，yi，j，zi，j)(j＝1，2，3，...，n3)，且满足||pi，j-pi||≤r1，其中n3表示点云pi的邻域点pi，j的数量，r1为球面邻域模型的半径，球面邻域模型的中心为点云pi，r1＝Wtree/4，Wtree表示当前树木整个树冠的平均叶宽；(1) The coordinates of any point cloud pi in the leaf point cloud set P obtained after branch and leaf separation are pi (xi ,yi , zi ), and the coordinates of the neighborhood point pi,j of point cloud pi are defined as pi,j (xi,j ,yi,j ,zi,j ) (j=1,2,3,...,n3 ), and satisfy ||pi,j -pi ||≤r1 , where n3 represents the number of neighborhood points pi,j of point cloud p i, r1 is the radius of the spherical neighborhood model, the center of the spherical neighborhood model is point cloud pi , r1 =Wtree /4, Wtree represents the average leaf width of the entire crown of the current tree;
(2)辅助准则一：(2) Auxiliary criterion 1:
寻找满足公式(1)的点云pi，公式(1)表示当前点云pi和点
其中
(3)辅助准则二：(3) Auxiliary criterion 2:
寻找满足公式(2)的点云pi，公式(2)表示当前点云pi与拟合平面
其中拟合平面
(4)辅助准则三：(4) Auxiliary criterion three:
寻找满足公式(3)的点云pi，公式(3)表示当前点云pi的所有邻域点pi，j与拟合平面
其中n3表示点云pi的邻域点pi，j的数量；Where n3 represents the number of neighboring points pi,j of point cloud pi ;
(5)辅助准则四：(5) Auxiliary criterion 4:
将拟合平面
其中
(6)获取同时满足辅助准则一中的公式(1)、辅助准则二中的公式(2)、辅助准则三中的公式(3)以及辅助准则四中的公式(4)的点云pi的数据集合
作为本发明进一步改进的技术方案，所述的步骤3具体为：As a further improved technical solution of the present invention, the step 3 is specifically as follows:
(1)设置树木的参数MinPts的值，其中参数MinPts为基于密度的空间聚类算法所需的最小点数，采用公式(5)和公式(6)计算出参数MaxR的值：(1) Set the value of the tree parameter MinPts, where MinPts is the minimum number of points required by the density-based spatial clustering algorithm. Use formulas (5) and (6) to calculate the value of the parameter MaxR:
其中n4表示树冠中叶片的中心区域点的点数量，n的值为3，表示点的维数，Γ是伽马函数，T是由点云形成的实验空间的体积，
(2)将参数MinPts和参数MaxR作为基于密度的空间聚类算法的两个输入参数；然后使用基于密度的空间聚类算法对中心区域点
作为本发明进一步改进的技术方案，所述的步骤4中每个叶子表面的非中心区域点的获取步骤为：As a further improved technical solution of the present invention, the step of obtaining the non-central area points on the surface of each leaf in step 4 is as follows:
定义叶子的非中心区域点的数据集为
其中
本发明的有益效果为：本基于激光点云的活立木叶属性精准估测方法通过处理树冠的整个点云来自动检测单叶，以获得单个叶片尺度信息，包括叶面积和叶角分布，与人工现场采集数据的方法相比，效率更高，减少人工劳动成本，且可以精确的得到树冠每个单片叶子的叶片属性。The beneficial effects of the present invention are as follows: this method for accurately estimating the leaf attributes of standing trees based on laser point clouds automatically detects single leaves by processing the entire point cloud of the crown to obtain the scale information of a single leaf, including leaf area and leaf angle distribution. Compared with the method of manually collecting data on site, it is more efficient, reduces labor costs, and can accurately obtain the leaf attributes of each single leaf in the crown.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本实施例选择的目标树木的扫描数据收集示意图。FIG. 1 is a schematic diagram of scanning data collection of target trees selected in this embodiment.
图2为本实施例显示三种目标树木的真实叶倾角和叶方位角分布的箱形图。FIG. 2 is a box plot showing the distribution of the actual leaf inclination angles and leaf azimuth angles of three target trees in this embodiment.
图3为本实施例枝叶分离后的三种目标树木的枝干点云和树冠叶子点云示意图。FIG3 is a schematic diagram of the branch point cloud and the crown leaf point cloud of the three target trees after the branches and leaves are separated in this embodiment.
图4为本实施例所提出的单个叶子分割算法的流程图。FIG4 is a flow chart of a single leaf segmentation algorithm proposed in this embodiment.
图5为本实施例叶子的中心点的判别准则示意图。FIG. 5 is a schematic diagram of the criteria for determining the center point of a leaf in this embodiment.
图6为本实施例公式(7)在单个叶片分割中的作用示意图。FIG6 is a schematic diagram showing the effect of formula (7) in the segmentation of a single leaf in this embodiment.
图7为本实施例自动分割叶子获得的点云冠层数据的散点图。FIG. 7 is a scatter plot of the point cloud canopy data obtained by automatically segmenting leaves in this embodiment.
图8为本实施例的方法和三种目标树的手动方法之间所估计的叶属性比较示意图。FIG8 is a schematic diagram showing a comparison of leaf attributes estimated by the method of this embodiment and three manual methods of target trees.
具体实施方式DETAILED DESCRIPTION
下面根据图1至图8对本发明的具体实施方式作出进一步说明：The specific implementation of the present invention is further described below with reference to FIGS. 1 to 8 :
本实施例提供一种基于激光点云的活立木叶属性精准估测方法，包括：This embodiment provides a method for accurately estimating leaf attributes of standing trees based on laser point clouds, including:
步骤(一)、树木点云数据的采集；Step (1), collection of tree point cloud data;
步骤(二)、对树木点云数据进行处理，实现枝叶分离；Step (ii), processing the tree point cloud data to separate branches and leaves;
步骤(三)、单一叶片的分割，包括：Step (iii), segmentation of a single leaf, comprises:
(3.1)、提取叶子的中心区域点：通过具有自适应半径的球体邻域模型和四个辅助准则提取叶子的中心区域点；(3.1) Extract the central area point of the leaf: extract the central area point of the leaf through a spherical neighborhood model with an adaptive radius and four auxiliary criteria;
(3.2)、叶子中心区域点的聚类：通过基于密度的空间聚类算法聚类叶子的中心区域点，获得每个叶子表面的中心点；(3.2) Clustering of leaf center area points: Cluster the leaf center area points using a density-based spatial clustering algorithm to obtain the center point of each leaf surface;
(3.3)、单片叶子的分离：通过每个叶子表面的中心点和三维分水岭算法获取每个叶子表面的非中心区域点，从而实现单片叶子点云的分割；(3.3) Separation of single leaves: The non-central area points of each leaf surface are obtained through the center point of each leaf surface and the three-dimensional watershed algorithm, thereby realizing the segmentation of the single leaf point cloud;
步骤(四)、基于分类叶点的叶属性计算，包括：Step (iv), leaf attribute calculation based on classified leaf points, includes:
(4.1)、采用Delaunay三角剖分面向单片叶子点云推导出每片叶子的面积；(4.1) Delaunay triangulation is used to derive the area of each leaf from the point cloud of a single leaf;
(4.2)、对单片叶子点云，通过用最小二乘法拟合叶片平面从而获得叶子表面的法向矢量，计算叶子表面的法向矢量与天顶角之间的夹角，该夹角为对应叶子的叶片倾角；通过该步骤的方法计算出每个叶子对应的叶片倾角；(4.2) For a single leaf point cloud, the leaf plane is fitted using the least squares method to obtain the normal vector of the leaf surface, and the angle between the normal vector of the leaf surface and the zenith angle is calculated. The angle is the leaf inclination angle of the corresponding leaf. The leaf inclination angle corresponding to each leaf is calculated by the method of this step.
(4.3)、计算水平面的北方向与叶子主轴线在水平面上的投影之间的顺时针角度，该顺时针角度即为对应叶子的叶片方位角，其中叶子主轴线为对应叶子的单片叶子点云中最远两点之间的线；通过该步骤的方法计算出每个叶子对应的叶片方位角。(4.3) Calculate the clockwise angle between the north direction of the horizontal plane and the projection of the main axis of the leaf on the horizontal plane. The clockwise angle is the leaf azimuth of the corresponding leaf, where the main axis of the leaf is the line between the two farthest points in the single leaf point cloud of the corresponding leaf. Calculate the leaf azimuth corresponding to each leaf through the method of this step.
下面对上述的每个步骤进行详细阐述。Each of the above steps is explained in detail below.
步骤(一)、树木点云数据的采集Step 1: Collection of tree point cloud data
本实施例使用地面激光扫描仪(Leica C10)仪器扫描具有不同宽叶和不同叶密度的树木的树冠，并且使用单站扫描来获得树木点云数据。Leica C10仪器是一台532nm相位扫描仪，具有360°×270°的向上视场，激光速率为50000点/秒。距离测量精度为±1.5mm，距离为3m。扫描仪出口处的圆形激光束直径和光束发散分别为3mm和0.22mrad，在离仪器3m的距离处产生约0.4mm的连续光束之间的最小距离。This embodiment uses a terrestrial laser scanner (Leica C10) instrument to scan the crowns of trees with leaves of different widths and different leaf densities, and uses single-station scanning to obtain tree point cloud data. The Leica C10 instrument is a 532nm phase scanner with a 360°×270° upward field of view and a laser rate of 50,000 points/second. The distance measurement accuracy is ±1.5mm at a distance of 3m. The circular laser beam diameter and beam divergence at the scanner exit are 3mm and 0.22mrad, respectively, resulting in a minimum distance between continuous beams of about 0.4mm at a distance of 3m from the instrument.
本实施例的基于激光点云的活立木叶属性精准估测方法适用于不同树种的树木，为了方便表述，本实施例选取三种树木作为目标树木，即一棵典型的粗糠树，一棵典型的紫薇和一棵典型的八角金盘。它们由地面激光扫描仪(Leica C10)扫描得到的点云数据如图1所示。这些树木的结构特征包括树高、树冠高度和冠层投影面积，如图1所示。其中激光雷达点密度分别为16946.3点(points)pts·m-2，16188.76pts·m-2和51483.96pts·m-2。树木的高度分别为3.5米，3.42米和1.6米。树冠的底部高度分别为0.97m，1.26m和0.01m。冠部的投影面积分别为1.84m2,3.136m2和4.592m2。每棵树的胸径(DBH)统计数据也列在图1中，其中DBH定义为胸部每个分支直径的总和。这些比较表明，对于该三种目标树木，三种目标树木之间的基本结构参数各异。由于选定的目标树木中没有出现极端天气(台风，寒害，酸雨等)，因此没有叶子掉落与扫描噪声点增加的情况。这些选定的目标树木适合于研究叶属性估测。The method for accurately estimating leaf attributes of standing trees based on laser point cloud in this embodiment is applicable to trees of different tree species. For the convenience of description, this embodiment selects three trees as target trees, namely a typical chaff tree, a typical crape myrtle and a typical golden plate. The point cloud data obtained by scanning them by a ground laser scanner (Leica C10) are shown in Figure 1. The structural characteristics of these trees include tree height, crown height and canopy projection area, as shown in Figure 1. The laser radar point density is 16946.3 points (points) pts·m-2 , 16188.76pts·m-2 and 51483.96pts·m-2 respectively. The heights of the trees are 3.5 meters, 3.42 meters and 1.6 meters respectively. The bottom heights of the crowns are 0.97m, 1.26m and 0.01m respectively. The projection areas of the crowns are1.84m2 ,3.136m2 and4.592m2 respectively. The DBH statistics of each tree are also listed in Figure 1, where DBH is defined as the sum of each branch diameter at breast height. These comparisons show that for the three target trees, the basic structural parameters are different between the three target trees. Since no extreme weather (typhoon, cold damage, acid rain, etc.) occurred in the selected target trees, there was no leaf drop and increase in scanning noise points. These selected target trees are suitable for studying leaf attribute estimation.
图1为树木选择和扫描数据收集示意图：使用地面激光扫描仪扫描三种目标树木，三种目标树木包括(a)粗糠树、(b)紫薇和(c)八角金盘；图1中(d)、(e)和(f)显示了扫描点云并罗列了三种目标树木的一系列结构特征。Figure 1 is a schematic diagram of tree selection and scanning data collection: a terrestrial laser scanner was used to scan three target trees, including (a) chaff tree, (b) crape myrtle, and (c) golden tamarisk; Figure 1 (d), (e), and (f) show the scanned point clouds and list a series of structural features of the three target trees.
本实施例为了获得真叶面积和叶角分布并证明本实施例的活立木叶属性精准估测方法的有效性，可以使用LI-3000C便携式面积计和角度测量装置测量所有目标树木的每个单叶面积和叶片倾角和叶片方位角。为了验证最终结果，本实施例对具有不同叶子数的三棵目标树木(粗糠树、紫薇和八角金盘)进行了采样，并使用角度测量装置手动测量了所有叶子的叶片倾角和叶片方位角。叶片角度分布箱形图(见图2，图2显示三棵目标树木的真实叶倾角和叶方位角分布的箱形图)表明，随着采样叶子数量的增加，叶片倾角和叶片方位角越稳定；图2表明，分析数据集的数量占树冠叶片总数的40％，叶片倾角和叶片方位角值趋于相对稳定。In order to obtain the true leaf area and leaf angle distribution and prove the effectiveness of the method for accurately estimating the leaf attributes of standing trees in this embodiment, the LI-3000C portable area meter and angle measuring device can be used to measure the area of each single leaf and the leaf inclination and leaf azimuth of all target trees. In order to verify the final results, this embodiment sampled three target trees (chaff tree, crape myrtle and golden plate) with different numbers of leaves, and manually measured the leaf inclination and leaf azimuth of all leaves using an angle measuring device. The box plot of leaf angle distribution (see Figure 2, Figure 2 shows the box plot of the true leaf inclination and leaf azimuth distribution of the three target trees) shows that as the number of sampled leaves increases, the leaf inclination and leaf azimuth become more stable; Figure 2 shows that the number of analyzed data sets accounts for 40% of the total number of crown leaves, and the leaf inclination and leaf azimuth values tend to be relatively stable.
步骤(二)、对树木点云数据进行处理，实现枝叶分离Step (2): Process the tree point cloud data to separate branches and leaves
枝叶分离旨在将地面激光扫描仪扫描得到的点云数据分类为枝干和叶子，是实现叶片分离和获得个体树叶特征的必要先决条件。未知程度的枝干成分可能导致叶面积估算误差。考虑到枝叶分离结果的重要性，在本实施例中，使用每个点的法向量，结构张量和点法向量特征来分离树木的叶子。图3显示了枝叶分离的结果，给出了三种目标树木中每一个的枝干点云和树冠叶子点云。图3具体为计算每个点的一系列特征包括法向量，结构张量和点法向量来实现枝叶分离结果。粗糠树为第一行的(a1)、(a2)、(a3)，紫薇为第二行的(b1)、(b2)、(b3)，八角金盘为第三行的(C1)、(C2)、(C3),其中(a1)、(b1)、(c1)为三种树木的初始点云；(a2)、(b2)、(c2)为枝叶分离后三种树木的枝干点云；(a3)、(b3)、(c3)为枝叶分离后三种树木的树冠叶子点云。Branch and leaf separation aims to classify the point cloud data obtained by scanning with a terrestrial laser scanner into branches and leaves, and is a necessary prerequisite for achieving leaf separation and obtaining individual leaf features. The unknown degree of branch components may lead to errors in leaf area estimation. Considering the importance of the branch and leaf separation results, in this embodiment, the normal vector, structure tensor and point normal vector features of each point are used to separate the leaves of the trees. Figure 3 shows the results of branch and leaf separation, giving the branch point cloud and crown leaf point cloud of each of the three target trees. Figure 3 specifically calculates a series of features for each point including the normal vector, structure tensor and point normal vector to achieve the branch and leaf separation results. The chaff trees are (a1), (a2), and (a3) in the first row, the crape myrtle are (b1), (b2), and (b3) in the second row, and the golden plate is (C1), (C2), and (C3) in the third row, among which (a1), (b1), and (c1) are the initial point clouds of the three trees; (a2), (b2), and (c2) are the branch point clouds of the three trees after the branches and leaves are separated; (a3), (b3), and (c3) are the crown and leaf point clouds of the three trees after the branches and leaves are separated.
步骤(三)、单一叶片的分割Step (3): Segmentation of a single leaf
叶片是树冠的主要部分，主要呈现绿色扁平椭圆状，表面积大，有利于气体交换和光能的吸收，这对研究叶片结构参数具有重要意义。叶子的大小和形状随树种的不同而不同。然而，大小形态在同一植株中相对稳定，这一特性可以用作鉴定植物物种和个体叶子分割。同一植物的叶片宽度具有相对稳定的数值范围。本实施例的三种树木的叶片具有平面特征，因此每片叶片的相应激光扫描点也具有平面特征。The leaf is the main part of the tree crown, which is mainly green, flat and oval, with a large surface area, which is conducive to gas exchange and absorption of light energy, which is of great significance for studying leaf structure parameters. The size and shape of the leaf vary with the tree species. However, the size and shape are relatively stable in the same plant, and this characteristic can be used to identify plant species and segment individual leaves. The leaf width of the same plant has a relatively stable numerical range. The leaves of the three trees in this embodiment have planar characteristics, so the corresponding laser scanning points of each leaf also have planar characteristics.
基于上述特征，本实施例所提出的单个叶子分割算法由三个主要阶段步骤组成，如图4所示。第一阶段步骤为：(1)、通过具有自适应半径的球体邻域模型和四个辅助准则提取叶子的中心区域点；即利用具有自适应半径的球体邻域模型提取叶子中心的点云；并定义了四个辅助准则以确保叶子的提取中心区域点的准确性。第一阶段步骤为：(2)、通过基于密度的空间聚类算法(DBSCAN)聚类叶子的中心区域点，从而获得对应于每个叶子表面的中心点；第三阶段为(3)、通过每个叶子表面的中心点和三维分水岭算法获取每个叶子表面的非中心区域点，从而实现单片叶子点云的分割。图4为所提出的单个叶子分割算法的流程图。整个分割方法包括三个阶段。(a)第一阶段是使用具有自适应半径和四个辅助准则的球体邻域模型来提取每个叶子的中心区域点。从第一阶段提取的深黑点为每个叶子的中心区域点。(b-d)在第二阶段，使用DBSCAN算法聚类每个叶子的中心区域点，并且获得每个叶子的中心点，如图(d)五角星标记。(d-e)在第三阶段，使用基于每个提取的叶表面的中心点的空间分水岭算法和两个形态学相关参数来实现单独的叶分离。Based on the above characteristics, the single leaf segmentation algorithm proposed in this embodiment consists of three main stages, as shown in Figure 4. The first stage steps are: (1) extracting the central area points of the leaves through a spherical neighborhood model with an adaptive radius and four auxiliary criteria; that is, using a spherical neighborhood model with an adaptive radius to extract the point cloud of the leaf center; and defining four auxiliary criteria to ensure the accuracy of the extracted central area points of the leaves. The first stage steps are: (2) clustering the central area points of the leaves through a density-based spatial clustering algorithm (DBSCAN) to obtain the central point corresponding to the surface of each leaf; the third stage is (3) obtaining the non-central area points of each leaf surface through the central point of each leaf surface and the three-dimensional watershed algorithm, thereby realizing the segmentation of a single leaf point cloud. Figure 4 is a flowchart of the proposed single leaf segmentation algorithm. The entire segmentation method includes three stages. (a) The first stage is to use a spherical neighborhood model with an adaptive radius and four auxiliary criteria to extract the central area points of each leaf. The dark black points extracted from the first stage are the central area points of each leaf. (b-d) In the second stage, the DBSCAN algorithm is used to cluster the central area points of each leaf, and the central point of each leaf is obtained, as marked by the five-pointed star in Figure (d). (d-e) In the third stage, the spatial watershed algorithm based on the central point of each extracted leaf surface and two morphological related parameters are used to achieve individual leaf separation.
(3.1)、提取叶子的中心区域点(第一阶段)(3.1) Extract the central area points of the leaves (first stage)
对于具有不同形态结构和复杂空间分布的叶子，确定每个叶片所在的空间平面尤为重要。通过分析，三棵目标树木的叶子可以近似地看作平面处理，并且每个叶子具有自己的中心点，通常位于叶子的中心。然而，每片叶子具有不同的天顶角和方位角，因此难以直接定位每片叶片的平面。然而，中心点必须能够在整个叶片的某个空间方向上延伸以形成平面，该平面总是可以在3D空间中找到，即叶片的点云所在的平面。在本实施例提出的自适应半径的球面邻域模型中，当叶片点云在整个球体模型中均匀地填充球体的某个大圆平面时，球体的中心即是叶子的中心区域点。随机分布的扫描点云位于每个叶子表面上的3D空间中的平面上。本实施例采用三元方程Sζ(x，y，z)＝0定义该平面，并且叶子上的任何点pi(xi，yi，zi)，(i＝1，2，3，...，n1)满足pi(xi，yi，zi)∈Sζ，(ζ＝1，2，3，...，n2)。n1表示每个树冠中扫描点的总数。n2表示每个树冠中的叶子总数，并且n2的值是未知的。Sζ表示3D空间中的每个叶子表面。然而，每个叶子表面是不规则的并且不能通过数学方程容易地表达。为了呈现每个叶子的表面方程，Sζ可以由每个叶子表面的中心区域点表示。每个叶子的中心区域中的点的提取也是接下来单独叶子分割的基础。因此，本实施例主要使用空间球面邻域模型来提取叶子的中心区域点。For leaves with different morphological structures and complex spatial distributions, it is particularly important to determine the spatial plane where each leaf is located. Through analysis, the leaves of the three target trees can be approximately regarded as planes, and each leaf has its own center point, which is usually located at the center of the leaf. However, each leaf has a different zenith angle and azimuth, so it is difficult to directly locate the plane of each leaf. However, the center point must be able to extend in a certain spatial direction of the entire leaf to form a plane, which can always be found in 3D space, that is, the plane where the point cloud of the leaf is located. In the spherical neighborhood model with adaptive radius proposed in this embodiment, when the leaf point cloud uniformly fills a large circular plane of the sphere in the entire sphere model, the center of the sphere is the central area point of the leaf. The randomly distributed scanned point cloud is located on a plane in 3D space on the surface of each leaf. This embodiment uses the ternary equation Sζ (x, y, z) = 0 to define the plane, and any point pi (xi,yi , zi ), (i = 1, 2, 3, ..., n1 ) on the leaf satisfies pi (xi ,yi , zi ) ∈S ζ , (ζ = 1, 2, 3, ..., n2 ).n 1 represents the total number of scan points in each crown. n2 represents the total number of leaves in each crown, and the value of n2 is unknown. Sζ represents the surface of each leaf in 3D space. However, each leaf surface is irregular and cannot be easily expressed by mathematical equations. In order to present the surface equation of each leaf, Sζ can be represented by the central area points of each leaf surface. The extraction of points in the central area of each leaf is also the basis for the subsequent individual leaf segmentation. Therefore, this embodiment mainly uses the spatial spherical neighborhood model to extract the central area points of the leaf.
根据上述分析，每个叶片大致处于平面分布中，并且对于每个叶片的中心区域点，邻域点必须均匀地分布在球体邻域模型的某个大圆上。本实施例定义了四个辅助准则，满足这些辅助准则的点pi(xi，yi，zi)被判断为叶子中心区域点。According to the above analysis, each leaf is roughly in a plane distribution, and for each leaf center area point, the neighborhood points must be evenly distributed on a certain great circle of the spherical neighborhood model. This embodiment defines four auxiliary criteria, and the points pi (xi,yi , zi ) that meet these auxiliary criteria are judged as leaf center area points.
在本实施例中，通过球邻域模型逐点提取叶子的中心区域点是个体叶片分割的关键步骤。首先，必须定义以下变量：每个球体邻域模型的半径为r1，每个球体的中心为pi，如图5中的(a4)所示。球面邻域模型半径r1的确定是逐点分类中的关键步骤，因为该变量会影响分类精度。在本实施例中，基于本实施例的灵敏度分析，将搜索球邻域的半径设置为r1＝Wtree/4，以平衡分类精度和计算效率.Wtree表示当前树种整个树冠的平均叶宽。对于叶子点云集合
提取叶子的中心区域点的具体步骤为(3.1.1)至(3.1.6)：The specific steps for extracting the center area points of the leaves are (3.1.1) to (3.1.6):
(3.1.1)定义变量：(3.1.1) Define variables:
枝叶分离后得到的叶子点云集合P中的任意点云pi的坐标为pi(xi，yi，zi)，将点云pi的邻域点pi，j的坐标定义为pi，j(xi，j，yi，j，zi，j)(j＝1，2，3，...，n3)，且满足||pi，j-pi||≤r1，其中n3表示点云pi的邻域点pi，j的数量，r1为球面邻域模型的半径，球面邻域模型的中心为点云pi，r1＝Wtree/4，Wtree表示当前树木整个树冠的平均叶宽。The coordinates of any point cloud pi in the leaf point cloud set P obtained after branch and leaf separation are pi (xi ,yi , zi ), and the coordinates of the neighborhood point pi,j of point cloud pi are defined as pi,j (xi,j ,yi,j ,zi,j ) (j=1,2,3,...,n3 ), and satisfy ||pi,j -pi ||≤r1 , where n3 represents the number of neighborhood points pi,j of point cloud p i, r1 is the radius of the spherical neighborhood model, the center of the spherical neighborhood model is the point cloud pi , r1 =Wtree /4, Wtree represents the average leaf width of the entire crown of the current tree.
(3.1.2)辅助准则一：(3.1.2) Auxiliary criterion 1:
寻找满足公式(1)的点云pi，公式(1)表示当前点云pi和点
其中
(3.1.3)辅助准则二：(3.1.3) Auxiliary criterion 2:
寻找满足公式(2)的点云pi，公式(2)表示当前点云pi与拟合平面
其中拟合平面
(3.1.4)辅助准则三：(3.1.4) Auxiliary criterion three:
保证由当前点pi及其邻域点组成的分布具有空间平面特征。即寻找满足公式(3)的点云pi，公式(3)表示当前点云pi的所有邻域点pi，j与拟合平面
其中n3表示点云pi的邻域点pi，j的数量。Where n3 represents the number of neighborhood pointspi,j of point cloudpi .
(3.1.5)辅助准则四：(3.1.5) Auxiliary criterion 4:
将拟合平面
寻找满足公式(4)的点云pi：Find the point cloud pi that satisfies formula (4):
(3.1.6)获取同时满足辅助准则一中的公式(1)、辅助准则二中的公式(2)、辅助准则三中的公式(3)以及辅助准则四中的公式(4)的点云pi的数据集合
本实施例的图5展示了叶子的中心区域点的判别准则，图5中的(a1-a4)表明球体的半径h值是叶子宽度的四分之一，深黑点表示从每片叶子中提取的中心区域点。(a2)(a3)和(a4)表示满足辅助准则二和辅助准则三，即点pi及其邻域点pi，j到拟合平面
(3.2)、叶子中心区域点的聚类(第二阶段)(3.2) Clustering of leaf center points (second stage)
通过第一阶段处理得到的每个树冠中的叶子的中心区域点有很多。因此，本实施例采用聚类算法(DBSCAN)，确定每片叶子分割的中心点。DBSCAN需要两个输入参数，其中包含形成聚类所需的最小点数(MinPts)和核心点附近的最大半径(此处为类间的最大距离)(MaxR)。确定两个输入参数是叶子中心区域点聚类过程中的关键步骤，因为这些参数会影响聚类结果的准确性。本实施例根据扫描点密度和每个叶面积的面积，将三颗目标树木(粗糠树，紫薇和八角金盘)的参数MinPts分别设置为15,10和30,然后，可以使用公式(5)和公式(6)从MinPts和点云的大小计算MaxR。There are many points in the central area of the leaves in each crown obtained through the first stage of processing. Therefore, this embodiment adopts a clustering algorithm (DBSCAN) to determine the central point of each leaf segmentation. DBSCAN requires two input parameters, including the minimum number of points required to form a cluster (MinPts) and the maximum radius near the core point (here is the maximum distance between classes) (MaxR). Determining the two input parameters is a key step in the clustering process of leaf central area points because these parameters will affect the accuracy of the clustering results. In this embodiment, the parameter MinPts of the three target trees (chaff tree, crape myrtle and golden plate) is set to 15, 10 and 30 respectively according to the scanning point density and the area of each leaf area. Then, MaxR can be calculated from MinPts and the size of the point cloud using formulas (5) and (6).
叶子中心区域点的聚类的具体步骤为：The specific steps of clustering the leaf center area points are:
(3.2.1)设置树木的参数MinPts的值，其中参数MinPts为基于密度的空间聚类算法所需的最小点数，采用公式(5)和公式(6)计算出参数MaxR的值：(3.2.1) Set the value of the tree parameter MinPts, where MinPts is the minimum number of points required by the density-based spatial clustering algorithm. Use formulas (5) and (6) to calculate the value of the parameter MaxR:
其中n4表示树冠中叶片的中心区域点的点数量，n的值为3，表示点的维数，Γ是伽马函数，T是由点云形成的实验空间的体积，
(3.2.2)将参数MinPts和参数MaxR作为基于密度的空间聚类算法的两个输入参数；然后使用基于密度的空间聚类算法对中心区域点
本实施例在使用DBSCAN算法进行聚类分析之后，每个叶子的中心区域点被分割成不同的聚类，如图4中的(c)所示。在中心区域点
(3.3)、单片叶子的分离(第三阶段)(3.3) Separation of single leaves (third stage)
三维分水岭算法用于分割树冠的剩余叶点云，即非中心区域点
本阶段的每个叶子表面的非中心区域点的获取步骤具体为：The specific steps for obtaining the non-central area points on the surface of each leaf in this stage are:
定义叶子的非中心区域点的数据集为
其中
步骤(四)、基于分类叶点的叶属性计算Step (IV): Calculation of leaf attributes based on classified leaf points
通过本实施例的方法进行单叶分割后，采用Delaunay三角剖分面向单片叶子点云推导出每片叶子的面积。对于分割后的单叶扫描数据点云，通过用最小二乘法拟合叶片平面来获得每个叶子表面的法向矢量，采用叶面法向矢量与天顶角之间的夹角计算叶片倾角。在本实施例中，每个叶片的主轴被定义为单叶点云中最远两点之间的线，因此，叶片方位角可以通过计算水平面上的北方向和叶子主轴线的投影之间的顺时针角度来获得。After the single leaf segmentation is performed by the method of this embodiment, the area of each leaf is derived from the single leaf point cloud using Delaunay triangulation. For the segmented single leaf scan data point cloud, the normal vector of each leaf surface is obtained by fitting the leaf plane using the least squares method, and the leaf inclination angle is calculated using the angle between the leaf surface normal vector and the zenith angle. In this embodiment, the main axis of each leaf is defined as the line between the farthest two points in the single leaf point cloud. Therefore, the leaf azimuth can be obtained by calculating the clockwise angle between the north direction on the horizontal plane and the projection of the main axis of the leaf.
本实施例的验证结果如下。The verification results of this embodiment are as follows.
(1)枝叶分离结果：(1) Branch and leaf separation results:
在将本实施例的枝叶分离方法应用于目标树木的初步扫描数据集之后，从点云获得的结果和实际现场测量之间获得了很好的一致性。这些良好的结果很大程度上归因于正确选择特征，包括法向量，结构张量和点法向量的分布。表1中列出了每种目标树木的枝叶分类准确度。总体精度是本实施例中枝叶分离方法产生的结果与使用人工划分枝干点云得出的结果的比率。下表1为不同树种枝叶分离的总体准确度评估。After applying the branch and leaf separation method of this embodiment to a preliminary scanned data set of target trees, good consistency was obtained between the results obtained from the point cloud and the actual field measurements. These good results are largely attributed to the correct selection of features, including the normal vector, the structure tensor, and the distribution of the point normal vector. The accuracy of branch and leaf classification for each target tree is listed in Table 1. The overall accuracy is the ratio of the results produced by the branch and leaf separation method of this embodiment to the results obtained using manually divided branch and trunk point clouds. Table 1 below is an overall accuracy evaluation of branch and leaf separation for different tree species.
(2)叶子分割结果：(2) Leaf segmentation results:
本实施例的方法分割了三个目标树木。详细结果显示在图7中，图7为使用本实施例提出的方法自动分割和随机着色叶子获得的点云冠层数据的散点图。获得的个体叶面点用于叶参数估计。图7中(a1)-(a4)为粗糠树，(b1)-(b4)为紫薇，(c1)-(c4)为八角金盘。图7中(a1)、(b1)和(c1)中的深黑点表示由球邻域模型提取的每个叶的中心区域点，浅灰点表示每个叶的整个叶点。图7中(a2)、(b2)和(c2)使用基于密度的空间聚类对中心区域点进行聚类，并使用随机颜色进行可视化。图7中(a3)、(b3)和(c3)表示叶子分割结果，且分别在(a4)、(b4)和(c4)中呈现部分放大图像。图7中的(a4)、(b4)和(c4)中的n为叶片数。The method of this embodiment segmented three target trees. The detailed results are shown in Figure 7, which is a scatter plot of the point cloud canopy data obtained by automatically segmenting and randomly coloring leaves using the method proposed in this embodiment. The individual leaf surface points obtained are used for leaf parameter estimation. In Figure 7, (a1)-(a4) are chaff trees, (b1)-(b4) are crape myrtles, and (c1)-(c4) are golden plates. The dark black dots in (a1), (b1) and (c1) in Figure 7 represent the central area points of each leaf extracted by the ball neighborhood model, and the light gray dots represent the entire leaf points of each leaf. In Figure 7, (a2), (b2) and (c2) use density-based spatial clustering to cluster the central area points and use random colors for visualization. In Figure 7, (a3), (b3) and (c3) represent the leaf segmentation results, and partially enlarged images are presented in (a4), (b4) and (c4), respectively. In Figure 7, (a4), (b4) and (c4) n is the number of leaves.
结果表明，三种目标树冠均表现出良好的叶片分割效果。通过本实施例的方法精确地提取目标树木的每个叶的中心区域点，并且提取结果在图7中的(a1)、图7中的(b1)和图7中的(c1)中显示为深黑点。然后，使用DBSCAN算法对目标树木的提取的中心区域点进行聚类，并且将各个叶的中心区域点分割成不同的类，如图7中的(a2)、图7中的(b2)和图7中的(c2)中的点所示。对于粗糠树获得了最佳的分割效果，其在整个冠层中具有相对平坦的叶子并且遮挡效应不明显。对于紫薇来说，获得了较好的结果，因为该树具有叶片尺寸校和叶密度搞；这种树型也实现了良好的分割效果。最后，叶片形状结构最复杂的八角金盘也具有良好的分割效果。The results show that all three target tree crowns exhibit good leaf segmentation effects. The central area points of each leaf of the target tree are accurately extracted by the method of this embodiment, and the extraction results are shown as dark black dots in (a1) in Figure 7, (b1) in Figure 7, and (c1) in Figure 7. Then, the extracted central area points of the target trees are clustered using the DBSCAN algorithm, and the central area points of each leaf are segmented into different classes, as shown by the points in (a2) in Figure 7, (b2) in Figure 7, and (c2) in Figure 7. The best segmentation effect was obtained for the chaff tree, which has relatively flat leaves throughout the canopy and the occlusion effect is not obvious. For the crape myrtle, better results were obtained because the tree has a moderate leaf size and high leaf density; this tree type also achieved a good segmentation effect. Finally, the golden twig with the most complex leaf shape structure also has a good segmentation effect.
(3)叶子属性计算结果：(3) Leaf attribute calculation results:
通过本实施例的方法分割后，将粗糠树的垂直分为上，中，下三个不同的水平层，紫薇分为上下两个不同的水平层，而八角金盘未分层。将冠中的分割叶点用作初步数据并随机选择作为分析数据集。对于每个目标树木，分析数据集中的叶子均匀分布在东，西，北，南四个方向上，分析数据集中叶子的数量占总树冠叶子数量的40％。在本实施例中，采用Delaunay三角剖分估计每个指定叶片的叶面积，并计算矢量角度以获得叶片方位角和叶片倾角。使用本实施例的方法获得的不同树种的具体叶参数(例如，叶长，叶宽，叶面积，叶方位角和叶倾角)列于表2中。手动测量和使用本实施例的方法比较结果在表2中罗列，这表明本实施例的方法获得了令人满意的树叶数量的估计。表2显示由三角剖分算法确定的叶面积与通过手动测量获得的实际叶面积非常接近。此外，直接在点云上计算的叶片角度分布值与在实际样地中获得的人工角度的测量结果高度相关。After segmentation by the method of this embodiment, the vertical of the chaff tree is divided into three different horizontal layers: upper, middle and lower, the crape myrtle is divided into two different horizontal layers, and the golden plate of the eight-angle plate is not layered. The segmented leaf points in the crown are used as preliminary data and randomly selected as the analysis data set. For each target tree, the leaves in the analysis data set are evenly distributed in the four directions of east, west, north and south, and the number of leaves in the analysis data set accounts for 40% of the total number of crown leaves. In this embodiment, Delaunay triangulation is used to estimate the leaf area of each specified leaf, and the vector angle is calculated to obtain the leaf azimuth and leaf inclination. The specific leaf parameters (e.g., leaf length, leaf width, leaf area, leaf azimuth and leaf inclination) of different tree species obtained using the method of this embodiment are listed in Table 2. The results of manual measurement and comparison using the method of this embodiment are listed in Table 2, which shows that the method of this embodiment obtains a satisfactory estimate of the number of leaves. Table 2 shows that the leaf area determined by the triangulation algorithm is very close to the actual leaf area obtained by manual measurement. In addition, the leaf angle distribution values calculated directly on the point cloud are highly correlated with the measurement results of the artificial angles obtained in the actual sample plot.
表2为使用本实施例的方法和手动测量的三个目标树木的叶属性估计的统计：Table 2 shows the statistics of leaf attribute estimation of three target trees using the method of this embodiment and manual measurement:
其中：NP：点数；PD：点密度；AS：分割的准确性；NLUM：使用本实施例的方法的叶子数量；NLMM：手动测量的叶子数量；RELN：叶子数量的误差比率；RELL：叶子平均长度的误差比率；RELW：叶子平均宽度的误差比率；RELA：叶子平均面积的误差比率；LIAUM：使用本实施例的方法计算出的叶片倾角；LIAMM：手动测量的叶片倾角；LAAUM：使用本实施例的方法计算出的叶片方位角；LAAMM：手动测量的叶片方位角。Wherein: NP: number of points; PD: point density; AS: segmentation accuracy; NLUM: number of leaves using the method of this embodiment; NLMM: number of leaves measured manually; RELN: error ratio of the number of leaves; RELL: error ratio of the average length of leaves; RELW: error ratio of the average width of leaves; RELA: error ratio of the average area of leaves; LIAUM: blade inclination angle calculated using the method of this embodiment; LIAMM: blade inclination angle measured manually; LAAUM: blade azimuth angle calculated using the method of this embodiment; LAAMM: blade azimuth angle measured manually.
图8描绘了使用本实施例的方法和手动测量获得的测量值在单个叶属性值估计上，包括叶面积，叶片方位角和叶片倾角。图8为本实施例的算法和三种目标树的手动方法之间所估计的叶属性比较：(a1)(b1)(c1)为使用LI-300C与LiDAR估计的单个树叶叶面积的比对散点图分布；(a2)(b2)(c2)为通过手工测量获得的和使用本实施例的方法获得的单片树叶片倾角的比对散点图分布；(a3)(b3)(c3)为通过手动测量获得的和使用本实施例的方法获得的单个树叶方位角的比对散点图分布。使用本实施例的方法对正确检测到的树的叶属性估计与人工实测值接近于1：1线(图8)。粗糠树的置信区间宽度相对较窄，表明粗糠树的叶片分割和属性估计结果具有较高的质量。一般而言，叶面积，叶片倾角和叶片方位角的人工测量值和方法估算值之间没有显着的平均差异(见表3)。该结果表明，单个叶面积估计与从实际测量获得的值非常一致，并且上部树冠中的单个叶面积大于中间和下部树冠中的单个叶面积(图8中(a1)、(b1))。叶片倾角估计与三个目标树木的实际测量值得到结果一致(R2＝0.908，RMSE＝6.806°，R2＝0.901，RMSE＝8.365°，R2＝0.849，RMSE＝6.158°；图8中的(a2)，(b2)和(c2))。对于粗糠树和紫薇树而言，上层树冠中的大倾角的叶片的概率高于中间树冠和下层树冠(图8中的(a2)、(b2))。对此结果的一种解释可能是，上层树冠叶子比下层树冠叶子更垂直生长，以增加穿透植物冠层的太阳辐射量，最终增加光穿透的可能性并有利于下部叶片元素的生长。类似地，下层树冠叶片生长倾向于比上部叶片更水平地生长，以优化吸收的太阳辐射。估计的叶片方位角与从粗糠树，紫薇和八角金盘的三个目标树的实际测量值非常一致(R2＝0.981，RMSE＝7.680°，R2＝0.938，RMSE＝7.573°，R2＝0.947，RMSE＝3.946°；图8中的(c1)、(c2)和(c3))。叶片方位角分布随树高的变化非常小。在粗糠树中，叶片方位角在整个冠部上随机分布在3.3°至353.6°之间(图8中的(a3))。紫薇树从上层到下层的方位角分布也非常相似(图8中的(b3))。此外，八角金盘方位角概率范围为50.9°至266.3°，这可能归因于植被叶片的光源定向性质或单侧TLS的不完整数据收集(图8中的(c3))。FIG8 depicts the comparison of the measured values obtained by the method of the present embodiment and the manual measurement on the estimation of single leaf attribute values, including leaf area, leaf azimuth and leaf inclination. FIG8 is a comparison of the leaf attributes estimated between the algorithm of the present embodiment and the manual method of three target trees: (a1)(b1)(c1) is a comparison scatter plot distribution of the leaf area of a single leaf estimated using LI-300C and LiDAR; (a2)(b2)(c2) is a comparison scatter plot distribution of the leaf inclination of a single leaf obtained by manual measurement and using the method of the present embodiment; (a3)(b3)(c3) is a comparison scatter plot distribution of the azimuth of a single leaf obtained by manual measurement and using the method of the present embodiment. The leaf attribute estimation of the correctly detected tree using the method of the present embodiment is close to the 1:1 line with the manual measured value (FIG8). The confidence interval width of the rough chaff tree is relatively narrow, indicating that the leaf segmentation and attribute estimation results of the rough chaff tree are of high quality. In general, there were no significant mean differences between the manual measurements and the method estimates of leaf area, leaf inclination, and leaf azimuth (see Table 3). The results showed that the individual leaf area estimates were in good agreement with the values obtained from actual measurements, and that the individual leaf areas in the upper crown were larger than those in the middle and lower crowns (Fig. 8 (a1), (b1)). The leaf inclination estimates were consistent with the actual measurements of the three target trees (R2 =0.908, RMSE=6.806°, R2 =0.901, RMSE=8.365°, R2 =0.849, RMSE=6.158°; Fig. 8 (a2), (b2), and (c2)). For the chaff tree and the crape myrtle tree, the probability of leaves with large inclination angles was higher in the upper crown than in the middle and lower crowns (Fig. 8 (a2), (b2)). One explanation for this result could be that upper crown leaves grow more vertically than lower crown leaves to increase the amount of solar radiation penetrating the plant canopy, ultimately increasing the likelihood of light penetration and favoring the growth of lower leaf elements. Similarly, lower crown leaf growth tends to grow more horizontally than upper leaves to optimize absorbed solar radiation. The estimated leaf azimuths were in good agreement with actual measurements from three target trees: Psoralea corylifolia, Crape Myrtle, and Aglaonema fasciata (R2 =0.981, RMSE=7.680°, R2 =0.938, RMSE=7.573°, R2 =0.947, RMSE=3.946°; (c1), (c2), and (c3) in Figure 8 ). The leaf azimuth distribution varied very little with tree height. In Psoralea corylifolia, leaf azimuths were randomly distributed across the crown between 3.3° and 353.6° ((a3) in Figure 8 ). The azimuth distribution of the crape myrtle trees from the upper layer to the lower layer was also very similar ((b3) in Figure 8). In addition, the azimuth probability of the golden disk ranged from 50.9° to 266.3°, which may be attributed to the light source directional nature of the vegetation leaves or the incomplete data collection of the single-sided TLS ((c3) in Figure 8).
表3为叶形态特征(叶面积，叶片方位角和叶片倾角)的总体结果的统计：Table 3 is the statistics of the overall results of leaf morphological characteristics (leaf area, leaf azimuth and leaf inclination):
其中，LA：叶面积；LIA：叶倾角；LAA：叶方位角。R2表示的是方程的拟合度，RMSE是测量值和真实值的误差。Where, LA: leaf area; LIA: leaf inclination; LAA: leaf azimuth. R2 represents the goodness of fit of the equation, and RMSE is the error between the measured value and the true value.
本实施例考虑到具有不同形态结构和复杂空间分布的叶片，提出一种基于计算机图形的3D点云分割方法，通过处理树冠的整个点云来自动检测单叶，以获得单个叶片尺度信息，包括叶面积和叶角分布。本实施例提供的活立木叶属性精准估测方法与人工现场采集数据的方法相比，可以快速并准确得到树冠每个单片叶子的叶片属性，减少人工劳动工作量，效率高。This embodiment takes into account leaves with different morphological structures and complex spatial distributions, and proposes a 3D point cloud segmentation method based on computer graphics. By processing the entire point cloud of the crown, a single leaf is automatically detected to obtain the scale information of a single leaf, including leaf area and leaf angle distribution. Compared with the method of manually collecting data on site, the method for accurately estimating the leaf attributes of standing trees provided by this embodiment can quickly and accurately obtain the leaf attributes of each single leaf of the crown, reduce the workload of manual labor, and have high efficiency.
本发明的保护范围包括但不限于以上实施方式，本发明的保护范围以权利要求书为准，任何对本技术做出的本领域的技术人员容易想到的替换、变形、改进均落入本发明的保护范围。The protection scope of the present invention includes but is not limited to the above embodiments. The protection scope of the present invention shall be based on the claims. Any replacement, deformation, and improvement of the technology that can be easily thought of by technicians in this field shall fall within the protection scope of the present invention.
Claims (3)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910130528.3ACN109961470B (en) | 2019-02-21 | 2019-02-21 | An accurate estimation method of living tree leaf attributes based on laser point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910130528.3ACN109961470B (en) | 2019-02-21 | 2019-02-21 | An accurate estimation method of living tree leaf attributes based on laser point cloud |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109961470A CN109961470A (en) | 2019-07-02 |
| CN109961470Btrue CN109961470B (en) | 2023-04-25 |
Family
ID=67023587
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910130528.3AActiveCN109961470B (en) | 2019-02-21 | 2019-02-21 | An accurate estimation method of living tree leaf attributes based on laser point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109961470B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110262287A (en)* | 2019-07-14 | 2019-09-20 | 南京林业大学 | Canopy height on-line water flushing method for the highly automated control of the harvest machinery ceding of Taiwan |
| CN110837849B (en)* | 2019-10-16 | 2022-04-08 | 中国农业科学院农业信息研究所 | How to get the veins of plant leaves |
| CN111507194B (en)* | 2020-03-20 | 2022-07-22 | 东华理工大学 | Ground-based LiDAR branch and leaf point cloud separation method based on fractal dimension supervised learning |
| CN111667529A (en)* | 2020-05-25 | 2020-09-15 | 东华大学 | Plant point cloud blade segmentation and phenotype characteristic measurement method |
| CN111811655B (en)* | 2020-06-03 | 2021-06-15 | 浙江大学 | Method for evaluating polarization state of light, application, device, electronic equipment and medium |
| CN112348802B (en)* | 2020-11-13 | 2025-01-10 | 黑龙江八一农垦大学 | Plant leaf inclination angle measurement method and device |
| CN114066966B (en)* | 2021-11-16 | 2025-05-02 | 南京林业大学 | A method for estimating tree crown porosity based on porous media theory and computer graphics |
| CN114119719A (en)* | 2021-12-03 | 2022-03-01 | 北京林业大学 | A tree diameter correction measurement method based on 3D point cloud |
| CN119337162B (en)* | 2024-09-13 | 2025-09-16 | 西安电子科技大学 | A radar unknown interference clustering method based on adaptive DBSCAN algorithm |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389538A (en)* | 2015-10-09 | 2016-03-09 | 南京大学 | Method for estimating forest leaf-area index based on point cloud hemisphere slice |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8572019B2 (en)* | 2010-04-21 | 2013-10-29 | Global Market Insite, Inc. | Reducing the dissimilarity between a first multivariate data set and a second multivariate data set |
- 2019
- 2019-02-21CNCN201910130528.3Apatent/CN109961470B/enactiveActive
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389538A (en)* | 2015-10-09 | 2016-03-09 | 南京大学 | Method for estimating forest leaf-area index based on point cloud hemisphere slice |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109961470A (en) | 2019-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109961470B (en) | An accurate estimation method of living tree leaf attributes based on laser point cloud | |
| CN110378909B (en) | Single wood segmentation method for laser point cloud based on Faster R-CNN | |
| CN106199557B (en) | A kind of airborne laser radar data vegetation extracting method | |
| WO2024109419A1 (en) | Elevation information-fused processing method for remote sensing images | |
| CN105389538B (en) | A method of based on a cloud hemisphere slice estimation Forest Leaf Area Index | |
| CN105740798B (en) | A kind of point cloud object scene recognition methods based on structural analysis | |
| CN109461207A (en) | A kind of point cloud data building singulation method and device | |
| CN112347894B (en) | A single vegetation extraction method based on transfer learning and Gaussian mixture model separation | |
| CN110348478B (en) | Method for extracting trees in outdoor point cloud scene based on shape classification and combination | |
| CN112907520B (en) | Single tree crown detection method based on end-to-end deep learning method | |
| CN111814666B (en) | A method, system, medium and equipment for single-tree parameter extraction under complex forest stand | |
| CN113269825B (en) | Forest breast diameter value extraction method based on foundation laser radar technology | |
| CN117765006A (en) | Multi-level dense crown segmentation method based on unmanned aerial vehicle image and laser point cloud | |
| CN116503727B (en) | Vegetation aggregation index estimation method based on Poisson distribution and point cloud clustering | |
| CN114913359A (en) | Method for extracting high-flux leaf included angle and leaf inclination angle distribution of field corn | |
| CN114898118A (en) | Automatic statistical method and system for power transmission line house removal amount based on multi-source point cloud | |
| CN111060922A (en) | Tree point cloud extraction method based on airborne laser radar point cloud spatial distribution characteristics | |
| Li et al. | Coarse-to-fine segmentation of individual street trees from side-view point clouds | |
| CN116402879A (en) | Corn ear position measuring method, device, system and storage medium | |
| Wang et al. | Quantification of canopy heterogeneity and light interception difference within greenhouse cucumbers based on terrestrial laser scanning | |
| CN117908048B (en) | Method for measuring corn biomass in high flux based on unmanned aerial vehicle laser radar data | |
| CN117475151B (en) | Airborne point cloud segmentation method combining tree top trunk detection and forest model | |
| CN118628740A (en) | Bottom-up single tree segmentation method based on UAV LiDAR point cloud in rubber forest | |
| Zhu et al. | Stem-leaf segmentation and phenotypic trait extraction of maize shoots from three-dimensional point cloud | |
| CN112381029B (en) | A building extraction method from airborne LiDAR data based on Euclidean distance |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | Effective date of registration:20230921 Address after:210000 No.8 Huayuan Road, Xuanwu District, Nanjing, Jiangsu Province Patentee after:Nanjing Maoting Information Technology Co.,Ltd. Address before:Nanjing City, Jiangsu province 210037 Longpan Road No. 159 Patentee before:NANJING FORESTRY University | |
| TR01 | Transfer of patent right |