A single back-pack laser radar based single tree volume and crown volume estimation method

Abstract

This invention discloses a method for estimating the volume of individual trees and canopy volume based on a single backpack LiDAR device, belonging to the field of forest resource survey technology. The method includes: acquiring LiDAR data using a backpack LiDAR device; preprocessing the single backpack LiDAR point cloud data; segmenting the point cloud data into individual trees using a density-based spatial clustering algorithm and a comparative shortest path algorithm; extracting the canopy point cloud based on the vertical distribution profile of the individual tree point cloud combined with the derivative extremum method; estimating the canopy volume using the voxel method and the AlphaShape algorithm; reconstructing the tree branch and trunk structure using a quantitative structural model (QSM) to estimate the volume of individual trees; and verifying the accuracy. This method constructs a three-dimensional quantitative structural model of individual trees using a single backpack LiDAR point cloud, enhancing the efficiency and accuracy of estimating the volume of individual trees and canopy volume.

Description

Technical Field

[0001] This invention belongs to the field of forest resource survey technology, specifically relating to a method for estimating the volume of individual timber and crown volume based on a single backpack lidar. Background Technology

[0002] Individual timber volume and canopy volume are crucial factors in forest resource surveys, significantly impacting forest management, individual timber competition assessment, and carbon sequestration. Furthermore, a precise understanding of individual timber structure, and a clearer picture of the influence of growing environment and management practices on individual timber growth, is essential for sustainable forest management and ecosystem solid-state sequestration. Conventional forest resource surveys often employ a combination of fieldwork and empirical models, neglecting the variability of tree growth. This approach is inefficient, often lacks precision, and is highly susceptible to systematic bias, making it difficult to implement and promote its application across large areas.

[0003] In recent years, methods for estimating individual tree volume based on LiDAR point cloud data have been gradually applied to individual tree surveys and carbon sequestration studies. Backpack-mounted LiDAR can obtain detailed three-dimensional structural information of trees (especially understory vegetation), handling scanning work in complex terrain with high efficiency. Simultaneously, it avoids the loss of point cloud data caused by mutual occlusion between tree trunks, ensuring data integrity and facilitating accurate estimation of individual tree volume and canopy volume. Extracting information from individual tree LiDAR point clouds helps achieve accurate estimation of individual tree volume and canopy volume.

[0004] In existing technologies, research on LiDAR-based estimation of single tree volume and canopy volume parameters includes: "Trunk Volume Calculation Based on 3D Laser Point Cloud and Cross-sectional Profile Curve," published in *Forestry Science*, Issue 11. This study estimated the volume of 183 tree trunk point clouds (1 m in length) from 7 tree species using a method based on 3D laser point clouds and cross-sectional profile curves. The results showed that the relative root mean square error of timber volume estimation based on ground-based 3D laser point clouds could be controlled within 64%. "Research on volume prediction of single tree canopy based on three-dimensional (3D) LiDAR and clustering segmentation," published in *International Journal of Remote Sensing*, Volume 42. This study, based on KD-tree single-tree segmentation, extracted canopy geometric features and constructed a regression model, achieving accurate extraction of canopy volume.

[0005] However, existing methods rely on multiple LiDAR data sources, making spatial registration and fusion of multi-source data difficult, and resulting in low sampling efficiency and stability. Furthermore, the lack of mobile, ground-based LiDAR technology hinders efficient scanning of tree trunks, making it difficult to avoid point cloud data loss due to occlusion between tree trunks, thus failing to obtain universal and accurate estimates of timber volume and canopy volume. Moreover, there is a lack of comprehensive and in-depth analysis of the horizontal and vertical distribution characteristics of individual tree point clouds, combined with tree morphological patterns and quantitative structural models to extract individual tree volume and canopy volume. Summary of the Invention

[0006] To address the problems mentioned in the background art, this invention proposes a method for estimating the volume of individual trees and the canopy volume based on a single backpack LiDAR. The backpack LiDAR can obtain detailed three-dimensional structural information of trees (especially understory vegetation), which can handle scanning work in complex terrain and has high scanning efficiency. At the same time, it can avoid the loss of point cloud data caused by mutual occlusion between tree trunks, ensuring the integrity of the data and helping to achieve accurate estimation of individual tree volume and canopy volume.

[0007] Technical Solution: To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for estimating the volume of a single timber and the crown volume of a tree based on a single backpack lidar includes the following steps:

[0009] Step 1: Collect LiDAR data using a backpack lidar device;

[0010] Step 2: Based on the single backpack LiDAR point cloud data, preprocess the single backpack LiDAR point cloud data.

[0011] Step 3: Perform single-tree segmentation on the point cloud data using density-based spatial clustering and shortest path comparison algorithms;

[0012] Step 4: Extract the canopy point cloud based on the vertical distribution profile of single tree point clouds combined with the derivative extremum method;

[0013] Step 5: Estimate the canopy volume using the voxel method and the AlphaShape algorithm;

[0014] Step 6: Reconstruct the tree branch and trunk structure using the quantitative structural model (QSM) to estimate the volume of individual timbers;

[0015] Step 7: Accuracy verification.

[0016] As a preferred option, the specific content of preprocessing the single backpack LiDAR point cloud data in step 2 is as follows:

[0017] First, an improved progressive encrypted triangular network filtering algorithm is used for ground point classification;

[0018] Secondly, a digital elevation model is established using the Kriging interpolation method;

[0019] Furthermore, based on the constructed digital elevation model, the point cloud data of individual tree topography is normalized to remove the influence of topography.

[0020] As a preferred option, the specific details of using density-based spatial clustering and shortest path comparison algorithms to perform single-tree segmentation of point cloud data in step 3 are as follows:

[0021] A density-based spatial clustering algorithm was used to separate the point cloud data of each individual tree in the sample plot, and the shortest path comparison algorithm was combined to segment the tree canopy;

[0022] When detecting tree canopies, point clouds with a height of 1.2m to 1.4m are used as input data. The minimum number of cluster points is set to 500, and the clustering threshold is set to 0.2m. The point clouds of the tree trunk within this range are extracted, and their centroids are calculated. The trunk radius is obtained by calculating the average distance from the centroid to each point on the trunk, and then the diameter at breast height (DBH) value is obtained.

[0023] As a preferred method, a density-based spatial clustering algorithm is used to separate the point cloud data of each individual tree in the sample plot, and the tree canopy is segmented using a shortest path comparison algorithm.

[0024] First, initial clusters are formed using DBSCAN based on point cloud density. The core is to define the neighborhood radius Eps and the minimum number of points MinPts to determine the core points and density-reachable points. The formula is as follows:

[0025] ;

[0026] ;

[0027] ;

[0028] in, Let Eps be the set of points in the neighborhood of point p. Represents a normalized canopy point cloud dataset, ( , , () represents the coordinates of point p; , , () represents the coordinates of point q; Indicates the Euclidean distance between two points;

[0029] Then, by comparing the shortest path algorithm to optimize the boundary, the centroid of the individual tree clusters is first calculated. The formula is:

[0030] ;

[0031] ;

[0032] ;

[0033] in, Indicates the number of point clouds within a cluster. Indicates the first in the cluster Coordinates of a point;

[0034] Starting from the center of mass again, using Calculate the shortest path length from each point to the centroid;

[0035] in, This represents the shortest path length from point p to the centroid; This represents the shortest path length from point q to the centroid; Represent the set of 6 to 10 nearest neighbors of point p; Indicates the Euclidean distance between two points;

[0036] Finally passed Divide the adhered areas;

[0037] in, This indicates that the boundary points are determined by 1.2 to 1.5 times the value of Eps. Indicates the distance between two adjacent points. Indicates with p j The shortest path length from the next adjacent point to the centroid. p j The shortest path length from a point to its centroid.

[0038] As a preferred option, the specific content of extracting forest canopy point clouds based on the vertical distribution profile of single-tree point clouds combined with the derivative extremum method in step 4 is as follows:

[0039] By calculating the maximum height h of a single log max Minimum height h min Obtain the maximum and minimum heights of the single tree point cloud; in h min ~h max Point cloud layers are sliced ​​vertically within the height range, with an interval height Δh between adjacent layers of 0.1m and a slice thickness t of 0.02m. The width w of each height layer is calculated, the curve equation of h with respect to w is fitted, and its derivative curve is obtained.

[0040] Starting from the bottom up, find the h value corresponding to the first oscillation point on the derivative curve, and denote this value as the height point at the bottom of the tree canopy. cbh Extracting height range in h cbh ~hmax The point cloud at that location is a canopy point cloud;

[0041] The method for determining the first oscillation point is as follows:

[0042] >10t;

[0043] Where h represents the height; w represents the width of each height layer; and t represents the slice thickness.

[0044] As a preferred option, the specific details of estimating the canopy volume using the voxel method and the AlphaShape algorithm in step 5 are as follows:

[0045] Before extracting canopy volume using the voxel method and the AlphaShape algorithm, sensitivity tests were performed on BLS data for voxel size and Alpha value. Canopy volumes were extracted for voxel sizes of 0.2, 0.3, 0.4, 0.5, and 0.6, and Alpha values ​​of 0.3, 0.5, 1, 2, and 5, respectively. The average error of different methods using different parameters for canopy volume extraction was then compared. The combination of voxel size and Alpha value with the smallest average error was selected for canopy volume extraction from BLS point clouds.

[0046] As a preferred option, the specific content of step 6, which involves reconstructing the tree branch and trunk structure using the quantitative structural model (QSM) to estimate the volume of individual timbers, is as follows:

[0047] The core of quantitative structural models lies in backbone segmentation and topological reconstruction:

[0048] First, the elevation data of the point cloud is used to detect the orientation of the tree trunks;

[0049] Then, the tree point cloud is layered by height to identify the main branch points and separate the trunk part;

[0050] Secondly, based on the distribution and connection method of the point cloud, the model of the tree trunk component is reconstructed;

[0051] Finally, the volumes of the cylindrical portions of the trunk represented in the reconstructed QSM model are added together to obtain the trunk volume.

[0052] Preferably, the specific content of the trunk volume is obtained by adding up the cylindrical volumes of the trunk portion represented in the reconstructed QSM model:

[0053] ;

[0054] Where V represents the trunk volume; n represents the total number of cylinders; i represents the cylinder number; r1 i h1 represents the radius of each cylinder;i This indicates the height of each cylinder.

[0055] As a preferred option, the specific content of the accuracy verification in step 7 is as follows:

[0056] The coefficient of determination R is used for estimating the volume of individual timbers and the canopy volume. 2 Accuracy is evaluated using Root Mean Square Error (RMSE), Root Mean Square Error (RMSE%), Bias, and Bias%; where RMSE% is the relative value of RMSE and Bias% is the relative value of Bias.

[0057] R 2 The formulas for calculating RMSE, RMSE%, Bias, and Bias% are as follows:

[0058] ;

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] in, This represents the true value of the volume of a single timber or the volume of the canopy. This represents an estimate of the volume of a single timber or the canopy volume. This represents the average volume of a single tree or the canopy volume, where m represents a single tree and Bias represents the deviation.

[0064] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0065] (1) Conventional forest resource surveys mostly adopt the method of combining field surveys with empirical models, which does not take into account the variability of tree growth, resulting in low efficiency and often low accuracy. They are also likely to have systematic biases, making it difficult to apply and promote them in a wide area. The backpack LiDAR of this invention can obtain detailed three-dimensional structural information of trees (especially understory vegetation), and can handle scanning work in complex terrain with high scanning efficiency. At the same time, it can avoid the loss of point cloud data caused by mutual occlusion between tree trunks, ensuring the integrity of the data and helping to achieve accurate estimation of individual tree volume and canopy volume.

[0066] (2) Previous methods were based on multiple LiDAR data sources, and spatial registration and fusion of multi-source data were difficult. However, the method of this invention constructs a three-dimensional quantitative structural model of a single tree using a single backpack LiDAR point cloud. Since this method uses a single point cloud and a quantitative structural model, it enhances the efficiency and accuracy of estimating the volume of a single tree and the canopy volume.

[0067] (3) Previous methods lacked mobile ground-based LiDAR technology, making it impossible to efficiently scan tree trunks and difficult to avoid point cloud data loss caused by mutual shading between tree trunks. Therefore, this method can fundamentally enhance the ability to estimate the volume of individual trees and canopy volume, thereby improving the accuracy of individual tree structural parameter estimation.

[0068] (4) This method considers the influence of lateral branches on the volume estimation of trees. The quantitative structural model method can effectively alleviate the problem of excessive trunk diameter caused by lateral branches, thereby improving the accuracy of single-tree volume. At the same time, this invention is not only conducive to the high-precision estimation of single-tree volume and canopy volume, but also easy to transfer (that is, it can be applied in different forest types in different regions). The verification results show that the overall accuracy of single-tree volume and canopy volume estimation by this invention is improved by more than 10% compared with other similar estimation methods. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of DEM construction and point cloud normalization according to the present invention;

[0070] Figure 2 This is a cross-validation diagram of canopy volume extracted based on the voxel method and the AlphaShape method of the present invention;

[0071] Figure 3 This is a verification diagram of the accuracy of the main body volume calculated by the method of this invention. Detailed Implementation

[0072] The present invention will be further illustrated below with reference to specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0073] Example 1

[0074] The method for estimating individual tree volume and canopy volume based on a single backpack LiDAR provided in this application can be applied in fields such as forest resource surveys, carbon sequestration, and landscape design and evaluation. It facilitates a precise understanding of individual tree structure and further clarifies the impact of growth environment and management practices on individual tree growth, which is of great significance for sustainable forest management and ecosystem solid-state sequestration. Conventional forest resource surveys mostly employ a combination of field surveys and empirical models, failing to consider the variability of tree growth, resulting in low efficiency and often low accuracy, potentially leading to systematic biases and hindering widespread practical application. Backpack LiDAR can obtain detailed three-dimensional structural information of trees (especially understory vegetation), handling scanning work in complex terrain with high efficiency. Simultaneously, it avoids the loss of point cloud data caused by mutual occlusion between tree trunks, ensuring data integrity and facilitating accurate estimation of individual tree volume and canopy volume.

[0075] This application presents a method for estimating the volume of individual trees and canopy volume based on a single backpack LiDAR. The method processes single backpack LiDAR point cloud data, including denoising and height normalization. Then, density-based spatial clustering and comparative shortest path algorithms are used to segment the point cloud data into individual trees. Canopy point clouds are extracted based on the vertical distribution profile of the individual tree point clouds combined with derivative extremum methods. The AlphaShape algorithm is then used to estimate the canopy volume. Finally, the tree branch and trunk structure is reconstructed using a quantitative structural model (QSM) to estimate the trunk volume. The specific implementation steps are as follows:

[0076] Step 1: Collect LiDAR data using a backpack lidar device;

[0077] The study area was selected as the Dongtai Yellow Sea Forest Park in Yancheng City, Jiangsu Province (120°07′~120°53′E, 32°33′~32°57′N). Poplar plantations within the study area were chosen as the research object, and data were collected from six square poplar plots with an area of ​​30 m × 30 m. LiDAR point cloud data was acquired using a LiBackpack DGC50H Backpack single-back-mounted laser scanning system, with a scanning frequency of 640,000 points / second. An "S"-shaped walking pattern was used for single-back-mounted LiDAR (BLS) point cloud data acquisition.

[0078] Step 2: Preprocess the single backpack LiDAR point cloud data;

[0079] First, an improved Progressive Encrypted Triangular Network Filtering (IPTD) algorithm is used for ground point classification.

[0080] First, extract the lowest elevation points (0.5%~1%) from the original point cloud as initial ground seed points to construct a Delaunay triangulation. Then, iteratively calculate the vertical distance from non-seed points to their respective triangulation faces. A dynamic threshold is used to determine ground points (the dynamic threshold adapts to different terrain complexities, avoiding classification errors from fixed thresholds). This process continues until the number of newly added ground points meets the termination condition. The determination formula is:

[0081]

[0082] in, This represents the perpendicular distance from a point to the triangular mesh surface; , , () represents the three-dimensional coordinates of the i-th non-seed point. It is the plane equation of the triangular mesh to which the projection of this point belongs. The three constants b and c represent the coordinates of the normal vector of the triangular mesh surface; This represents the distance from the triangular mesh to the origin of the coordinate system. The basic vertical distance threshold is (usually 0.05~0.1m), k is the adaptation coefficient (usually 0.01~0.02), and L is the average length of the triangulation mesh edges.

[0083] Secondly, a digital elevation model with a resolution of 0.2m was established using the Kriging interpolation method.

[0084] First, outliers at ground points are removed. The study area is then divided into grids of 0.2m × 0.2m. Next, a spatial variogram (semivariance function) is fitted to reflect the spatial correlation of elevations. Finally, the weights of neighboring ground points are determined by solving the Kriging equations. The optimal unbiased estimate of the elevation at the center of each grid is then calculated using the following interpolation formula:

[0085]

[0086] in, This indicates that the optimal unbiased estimate of the elevation of each grid center is performed; The center coordinates of a 0.2m resolution grid cell. For the first in the grid neighborhood Elevation values ​​of each ground point The weight corresponding to this ground point (obtained by solving the variogram) must satisfy the following conditions: Unbiasedness constraint.

[0087] Furthermore, based on the constructed digital elevation model (DEM), the point cloud data of individual trees is normalized to remove the influence of terrain.

[0088] First, project each point cloud data (x, y, z) onto the DEM grid, match the ground elevation of its corresponding grid cell, and then subtract the corresponding ground elevation from the original elevation to obtain the relative elevation. At the same time, remove noise points with negative elevation values ​​after normalization. The calculation formula is as follows:

[0089]

[0090] in, The normalized relative elevation (representing the actual height of the point cloud relative to the ground, i.e., the effective height of the canopy). This represents the original absolute elevation value of the point cloud. For this point Ground elevation values ​​of DEM grid cells corresponding to the coordinates.

[0091] Step 3: Perform single-tree segmentation on the point cloud data using density-based spatial clustering and shortest path comparison algorithms;

[0092] First, initial clusters are formed using DBSCAN based on point cloud density. The core is to define the neighborhood radius Eps and the minimum number of points MinPts to determine the core points and density-reachable points. The formula is as follows:

[0093]

[0094]

[0095]

[0096] in, For point p Neighborhood point set, For the normalized canopy point cloud dataset, ( , , () represents the coordinates of point p; , , () represents the coordinates of point q; The Euclidean distance between two points Take 0.5~2m, Take 5~20, the density can reach the point to form the initial single tree cluster.

[0097] Then, by comparing the shortest path algorithm to optimize the boundary, the centroid of the individual tree clusters is first calculated. The formula is:

[0098]

[0099]

[0100]

[0101] in, The number of point clouds within the cluster. For the first in the cluster The coordinates of the points.

[0102] Starting from the center of mass again, using Calculate the shortest path length from each point to the centroid.

[0103] in, This represents the shortest path length from point p to the centroid; This represents the shortest path length from point q to the centroid; Let p be the set of its 6 to 10 nearest neighbors; The distance between the two points is Euclidean.

[0104] Finally passed Divide the adhered areas.

[0105] in, for The boundary points are determined by 1.2 to 1.5 times the value of the boundary points. This represents the distance between two adjacent points. Indicates with p j The shortest path length from the next adjacent point to the centroid. p j The shortest path length from a point to its centroid.

[0106] When detecting tree canopies, point clouds with a height of 1.2m to 1.4m are used as input data. The minimum number of cluster points (MinPts) is set to 500, and the clustering threshold (Eps) is set to 0.2m. The trunk point clouds within this range are extracted, and their centroids are calculated. The trunk radius is calculated from the average distance from the centroid to each point on the trunk, and then the diameter at breast height (DBH) value is obtained.

[0107] Step 4: Extract the canopy point cloud based on the vertical distribution profile of single tree point clouds combined with the derivative extremum method;

[0108] This involves extracting the canopy layer of a single tree. This is done by calculating the maximum height (h) of a single tree. max Minimum height (h) min ), to obtain the maximum and minimum heights of a single tree point cloud.

[0109] We first need to use the independent point cloud cluster dataset obtained after segmenting individual trees. Extract the normalized relative elevation values ​​of all points to construct a height subset. Then, the maximum height of the single log can be directly calculated using the formula. and minimum height The formula is:

[0110]

[0111]

[0112] in, For the first Normalized relative elevation of each point This indicates the height of the top of the corresponding single piece of wood. This indicates the height of the base of the corresponding single log.

[0113] Within this altitude range (h) min ~h max Point cloud layers are sliced ​​vertically within the cloud, with an interval height (Δh) of 0.1m between adjacent layers and a slice thickness (t) of 0.02m. The width (w) of each height layer is calculated, and the curve equation of h with respect to w is fitted, and its derivative curve is obtained.

[0114] The value of h corresponding to the first oscillation point on the derivative curve from bottom to top is the height point at the bottom of the tree canopy, denoted as h. cbh Extracting height range in h cbh ~h max The point cloud at that location is a canopy point cloud. The method for determining the first oscillation point is as follows:

[0115] >10t

[0116] Where h represents the height; w represents the width of each height layer; and t represents the slice thickness.

[0117] Step 5: Extract canopy volume using voxel and AlphaShape methods;

[0118] The principle of the voxel method is to divide the three-dimensional model space into small cubic units (called voxels) and calculate the total volume of these voxels to estimate the overall volume of the model.

[0119] The AlphaShape algorithm is used to find the convex hull or non-convex shape of a point set, where the value of α determines the looseness of the geometry formed by the point set. For both the voxel method and the AlphaShape algorithm, the choice of parameters significantly affects the results. In the voxel method, smaller voxels can provide more accurate estimates, but missing point cloud data can increase the estimation error. Larger voxels may cause certain details to be ignored. For the AlphaShape algorithm, smaller parameters may underestimate the volume due to the void structure in the cloud data, while larger voxels may lead to less accurate predictions. Therefore, choosing appropriate parameters based on the characteristics of the model and the accuracy requirements is crucial.

[0120] In this embodiment, before extracting canopy volume using the voxel method and the AlphaShape algorithm, sensitivity tests were performed on the BLS data for the voxel size and α (shape control parameter of the AlphaShape algorithm). Specifically, the parameter sensitivity test was conducted using a combination of single-variable control and cross-comparison. Fixed gradients were set for the key parameters of the voxel method and the AlphaShape method, and canopy volume data under the corresponding parameters were extracted and the error was calculated. Specifically, the voxel method selected voxel size as the core sensitive parameter, setting five gradient values ​​of 0.2, 0.3, 0.4, 0.5, and 0.6; the AlphaShape method selected alpha value (α) as the core sensitive parameter, setting five gradient values ​​of 0.3, 0.5, 1, 2, and 5. Based on the above parameter gradient settings, the canopy volume of the same batch of BLS point cloud data was extracted using both the voxel method (different voxel sizes) and the AlphaShape method (different α values), obtaining the canopy volume extraction results of the two methods under various parameters. Subsequently, a cross-comparison matrix was constructed to compare the canopy volume extraction results corresponding to all parameter combinations, and the mean difference under each parameter combination was calculated as the core indicator for evaluating the merits of the parameter combinations.

[0121] By comprehensively cross-comparing the Mean Differences of all parameter combinations (25 groups in total) formed by the five voxel sizes of the voxel method and the five α values ​​of the AlphaShape method, the influence of different parameter configurations on the accuracy of canopy volume extraction was systematically analyzed. Finally, the voxel size and α parameter combination with the smallest Mean Differences was identified and used as the optimal parameter configuration for subsequent BLS point cloud canopy volume extraction, ensuring that the canopy volume extraction results have both high accuracy and stability.

[0122] Step 6: Reconstruct the tree branch and trunk structure based on the quantitative structural model (QSM) to estimate the volume of individual timbers;

[0123] The core of quantitative structural models lies in backbone segmentation and topological reconstruction:

[0124] First, the elevation data from the point cloud is used to detect the orientation of the tree trunks; then the root of the trunk is defined. ) and top ( To establish a local neighborhood, a series of candidate radii are predefined (r=0.05, 0.08, 0.10, 0.15). For each radius, the point... ∈ neighborhood The definition is as follows:

[0125]

[0126] in, Indicates a tree trunk dotted with clouds. This represents the point cloud of the i-th tree trunk; This represents the point cloud of the j-th tree trunk; denoted by , where r represents the Euclidean distance and r represents the radius.

[0127] Based on neighborhood relationships, the following weighted adjacency matrix was constructed:

[0128]

[0129] in, This represents a weighted adjacency matrix.

[0130] And symmetrically transformed into (in, This represents the weighted adjacency matrix after symmetry processing; Represents a weighted adjacency matrix; This represents the transpose of the weighted adjacency matrix, forming a neighborhood graph. A connectivity analysis is then performed, specifically:

[0131]

[0132] in, This represents the connected component to which the root of the tree trunk belongs; This represents the connected component to which the top of the tree trunk belongs; This represents the connected component to which node v belongs. The minimum radius that guarantees connectivity between the root node and the vertex nodes is selected.

[0133] Next, the tree point cloud is layered by height, the main branch points are identified, and the trunk is separated, which is the shortest path between the root node and the top node in the minimum spanning tree. Specifically:

[0134]

[0135] in, This represents all possible paths connecting two nodes in the minimum spanning tree. Let represent the weighted adjacency matrix after symmetry processing, and p represent the trunk point cloud.

[0136] Then, based on the distribution and connection method of the point cloud, the model of the trunk component is reconstructed;

[0137] Based on the local cylinder radius r1 of the tree trunk and the path segment height h1, a standard cylindrical mesh is generated along the default Z-axis. The cylinder cross-section mesh is as follows:

[0138]

[0139] in, This represents the cylindrical function; r1 is the local cylindrical radius of the tree trunk (calculated using the same method as the diameter-to-breast height fitting method in step 3), 16 is the number of segments in the cylindrical cross-section (determining the smoothness of the cylinder), and X and Y are the two-dimensional grid coordinates of the standard cylindrical cross-section. Stretch the cylinder height to the path segment length:

[0140]

[0141]

[0142] in, The default height grid (value 0~1) generated for the cylinder function, where h1 is the path segment length (i.e., cylinder height). This is the height grid for the stretched cylinder. Indicates the starting point of the path; This indicates the end point of the path; to rotate the standard cylinder (along the Z-axis) to align with the direction vector of the path segment and adapt it to the main direction, an orthogonal rotation of the standard cylinder is required:

[0143]

[0144] in, Y Z For flattened column vectors of a standard cylindrical mesh, This is the transpose of the rotation matrix (to achieve coordinate system rotation). This is the rotated cylindrical coordinate matrix (i.e., the reconstructed trunk component model).

[0145] Finally, the volume of the trunk is obtained by adding up the volumes of the cylindrical parts of the trunk represented in the reconstructed QSM model. The formula is as follows:

[0146] ;

[0147] Where V represents the trunk volume; n represents the total number of cylinders; i represents the cylinder number; r1 i h1 represents the radius of each cylinder; i This indicates the height of each cylinder.

[0148] Step 7: Accuracy verification.

[0149] For the estimated volume of individual timbers and canopy volume, the coefficient of determination R is mainly used. 2 Accuracy is evaluated using Root Mean Square Error (RMSE), Relative Root Mean Square Error (RMSE%), Bias, and Relative Bias% (Bias%). Here, RMSE% is the relative value of RMSE, Bias% is the relative value of Bias, and R0 is the relative value of Bias. 2The formulas for calculating RMSE, RMSE%, Bias, and Bias% are as follows:

[0150] ;

[0151] ;

[0152] ;

[0153] ;

[0154] ;

[0155] in, This represents the true value of the volume of a single timber or the volume of the canopy. This represents an estimate of the volume of a single timber or the canopy volume. This represents the average volume of a single tree or the canopy volume, where m represents the volume of a single tree; Bias represents the deviation.

[0156] The specific experimental procedure in this embodiment is as follows:

[0157] Poplar plantations were selected as the research object. Six square poplar plots with an area of ​​30 m × 30 m were investigated. Factors such as stand age, stand density and topography were taken into account. The selected plots were 8 years old (n=25), 12 years old (n=39), 17 years old (n=21), 20 years old (n=30), 27 years old (n=29) and 35 years old (n=11), covering different stand age stages of 0-10 years, 11-15 years, 16-20 years, 21-30 years and 31 years and above.

[0158] The ground survey used a total station to determine the coordinates of the four corner points of the sample plot, and measured the diameter at breast height (DBH), tree height (H), and branch height (CBH) of each individual tree within the plot. The coordinates of the corner points, the center coordinates, and the positional information of each individual tree were acquired using a real-time differential device (Fengjiang Intelligent Technology Co., Ltd.) FJDTrion RTK GNSS, which ensured a positioning accuracy of 3 cm within a 60° tilt angle. DBH measurements were taken using a steel tape measure in two opposite directions, and the average value was recorded. Tree height and branch height were measured using a Vertex V ultrasonic altimeter (Haglöf, Swiss), employing both ultrasonic and laser modes. Measurements were taken twice for each mode, and the average value was recorded. The mean values ​​and standard deviations of the DBH, H, and CBH data obtained from the sample plot survey are shown in Table 1.

[0159] Table 1. Statistical Table of Characteristic Parameters of Sample Plots

[0160]

[0161] LiDAR data was acquired using a backpack-mounted LiDAR device: The LiBackpack DGC50H Backpack laser scanning system (Beijing Digital Green Earth Technology Co., Ltd.) was used to acquire single backpack-mounted LiDAR data. The scanning frequency was set to 640,000 points / second, and the BLS point cloud data was acquired using an "S" shaped walking method.

[0162] Data preprocessing: After data collection, the improved Progressive Encryption Triangular Network Filtering (IPTD) algorithm was first used for ground point classification. Secondly, a 0.2m resolution digital elevation model (DEM) was established using Kriging interpolation. Based on the constructed DEM, the small-scale topographic point cloud data of individual trees was normalized to remove topographic influences. The results are as follows: Figure 1 As shown.

[0163] Individual tree segmentation: A density-based spatial clustering algorithm was used to separate the point cloud data of each individual tree in the sample plot, and the Comparative Shortest Path (CSP) algorithm was combined to segment the tree crown. When detecting the tree crown, the point cloud with a height of 1.2 m to 1.4 m was used as input data. The minimum number of cluster points (MinPts) was set to 500, and the clustering threshold (Eps) was set to 0.2 m. The trunk point cloud within this range was extracted, and its centroid was calculated. The trunk radius was obtained by calculating the average distance from the centroid to each point on the trunk, and then the diameter at breast height (DBH) value was obtained. The results are shown in Table 2, where r2 is the detection rate (%), F is the accuracy (%), and p1 is the overall precision (%).

[0164] Table 2 Statistical Results of Single Wood Segmentation Accuracy

[0165]

[0166] Single tree canopy extraction: by calculating the maximum height of a single tree (h) max Minimum height (h) min This retrieves the maximum and minimum heights of a single tree point cloud. Within this height range (h... min ~h max Point cloud slices are created vertically within the canopy, with an interval height (Δh) of 0.1m between adjacent slices and a slice thickness (t) of 0.02m. The width (w) of each height slice is calculated, and a curve equation of h with respect to w is fitted, with its derivative curve obtained. The h value corresponding to the first oscillation point on the derivative curve from bottom to top is the height point at the bottom of the canopy, denoted as h. cbh Extracting height range in h cbh ~h max The point cloud at that location is a canopy point cloud.

[0167] Voxel and AlphaShape Methods for Canopy Volume Extraction: Before using the voxel and AlphaShape algorithms to extract canopy volume, this embodiment performs sensitivity tests on BLS data for voxel size and α parameters. Specifically, canopy volumes are extracted when voxel size = 0.2, 0.3, 0.4, 0.5, 0.6 and α = 0.3, 0.5, 1, 2, 5. The mean differences between different methods using different parameters are then compared. A smaller mean difference indicates a better canopy volume extraction effect for that parameter combination. The voxel size and α parameter combination with the smallest mean difference is selected for canopy volume extraction from BLS point clouds.

[0168] Quantitative structural model estimation of single timber volume: First, the elevation data of point cloud is used to detect the direction of tree trunk; then, the tree point cloud is layered by height to identify the main branch points and separate the trunk part; then, the model of the trunk component is reconstructed according to the distribution and connection of the point cloud; finally, the volume of the trunk is obtained by adding the volumes of the cylindrical parts of the trunk represented in the reconstructed QSM model.

[0169] Accuracy verification: For the estimated individual timber volume and canopy volume, the coefficient of determination R is mainly used. 2 Accuracy is evaluated using Root Mean Square Error (RMSE), Relative Root Mean Square Error (RMSE%), Bias, and Relative Bias% (Bias%). Here, RMSE% is the relative value of RMSE, and Bias% is the relative value of Bias.

[0170] The results of estimating the volume of individual timbers and the canopy volume using the method proposed in this invention are shown below. Figure 2 and Figure 3 As shown.

[0171] Figure 2 The results show that the canopy volume extraction results of the point cloud of the backpack lidar using the voxel method and the AlphaShape algorithm are quite similar, and the cross-comparison proves that the canopy volume calculation accuracy is high.

[0172] Figure 3 This indicates that the single-tree trunk volume calculated by the method of this application for the point cloud of the backpack lidar is close to the result of the traditional binary volume table method, and has high accuracy.

[0173] Another existing method 1: "Trunk Volume Calculation Based on 3D Laser Point Cloud and Cross-sectional Profile Curve" published in the 11th issue of "Forestry Science" estimates the volume of 183 tree trunk point clouds with a length of 1 m from 7 tree species by using a trunk volume calculation method based on 3D laser point cloud and cross-sectional profile curve. The estimation results show that the relative root mean square error of the forest timber volume estimation based on the ground 3D laser point cloud can be controlled within 64%.

[0174] Another existing method 2: “Research on volume prediction of single tree canopy based on three-dimensional (3D) LiDAR and clustering segmentation” published in Volume 42 of the International Journal of Remote Sensing. This method extracts canopy geometric features and constructs a regression model based on KD-tree single tree segmentation, thus achieving accurate extraction of canopy volume.

[0175] Table 3 Comparison of the accuracy of single-log volume and crown volume estimation by other methods

[0176]

[0177] The proposed method for estimating individual timber volume and canopy volume based on a single backpack LiDAR provides a more flexible and convenient way to obtain detailed 3D structural information of trees, offering significant advantages in complex forest stands and terrain conditions. The flexible scanning method avoids the loss of point cloud data caused by mutual occlusion between tree trunks. Furthermore, the combination of voxel and AlphaShape methods ensures the accuracy of canopy volume calculation from both the internal composition and external morphological representation of the canopy point cloud. The quantitative structural model method based on trunk reconstruction can directly and accurately estimate individual timber volume from the perspectives of the 3D structure and spatial topology of the tree trunk. Verification results show that, compared with other similar estimation methods, the overall accuracy of estimating individual timber volume and canopy volume using this invention is improved by more than 10% (as shown in Table 3).

[0178] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for estimating the volume of a single timber and the crown volume of a tree based on a single backpack lidar, characterized in that, Includes the following steps: Step 1: Collect LiDAR data using a backpack lidar device; Step 2: Based on the single backpack LiDAR point cloud data, preprocess the single backpack LiDAR point cloud data. Step 3: Perform single-tree segmentation on the point cloud data using density-based spatial clustering and shortest path comparison algorithms; Step 4: Extract the canopy point cloud based on the vertical distribution profile of single tree point clouds combined with the derivative extremum method; Step 5: Estimate the canopy volume using the voxel method and the AlphaShape algorithm; Step 6: Reconstruct the tree branch and trunk structure using the quantitative structural model (QSM) to estimate the volume of individual timbers; Step 7: Accuracy verification.

2. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 1, characterized in that: The specific steps in step 2 for preprocessing the single backpack LiDAR point cloud data are as follows: First, an improved progressive encrypted triangular network filtering algorithm is used for ground point classification; Secondly, a digital elevation model is established using the Kriging interpolation method; Furthermore, based on the constructed digital elevation model, the point cloud data of individual tree topography is normalized to remove the influence of topography.

3. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 1, characterized in that: Step 3, which uses density-based spatial clustering and shortest path comparison algorithms to perform single-tree segmentation of the point cloud data, involves the following steps: A density-based spatial clustering algorithm was used to separate the point cloud data of each individual tree in the sample plot, and the shortest path comparison algorithm was combined to segment the tree canopy; When detecting tree canopies, point clouds with a height of 1.2m to 1.4m are used as input data. The minimum number of cluster points is set to 500, and the clustering threshold is set to 0.2m. The point clouds of the tree trunk within this range are extracted, and their centroids are calculated. The trunk radius is obtained by calculating the average distance from the centroid to each point on the trunk, and then the diameter at breast height (DBH) value is obtained.

4. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 3, characterized in that: The density-based spatial clustering algorithm was used to separate the point cloud data of each individual tree in the sample plot, and the shortest path comparison algorithm was combined to segment the tree canopy. The specific content is as follows: First, initial clusters are formed using DBSCAN based on point cloud density. The core is to define the neighborhood radius Eps and the minimum number of points MinPts to determine the core points and density-reachable points. The formula is as follows: ； ； ； in, Let Eps be the set of points in the neighborhood of point p. Represents a normalized canopy point cloud dataset, ( , , () represents the coordinates of point p; , , () represents the coordinates of point q; Indicates the Euclidean distance between two points; Then, by comparing the shortest path algorithm to optimize the boundary, the centroid of the individual tree clusters is first calculated. The formula is: ； ； ； in, Indicates the number of point clouds within a cluster. Indicates the first in the cluster Coordinates of a point; Starting from the center of mass again, using Calculate the shortest path length from each point to the centroid; in, This represents the shortest path length from point p to the centroid; This represents the shortest path length from point q to the centroid; Represent the set of 6 to 10 nearest neighbors of point p; Indicates the Euclidean distance between two points; Finally passed Divide the adhered areas; in, This indicates that the boundary points are determined by 1.2 to 1.5 times the value of Eps. Indicates the distance between two adjacent points. Indicates with p j The shortest path length from the next adjacent point to the centroid. p j The shortest path length from a point to its centroid.

5. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 1, characterized in that: The specific content of extracting forest canopy point clouds based on the vertical distribution profile of single-tree point clouds combined with the derivative extremum method in step 4 is as follows: By calculating the maximum height h of a single log max Minimum height h min Obtain the maximum and minimum heights of the single tree point cloud; in h min ~h max Point cloud layers are sliced ​​vertically within the height range, with an interval height Δh between adjacent layers of 0.1m and a slice thickness t of 0.02m. The width w of each height layer is calculated, the curve equation of h with respect to w is fitted, and its derivative curve is obtained. Starting from the bottom up, find the value of h corresponding to the first oscillation point on the derivative curve, and denote this value as the height point at the bottom of the tree canopy. cbh Extracting height range in h cbh ~h max The point cloud at that location is a canopy point cloud; The method for determining the first oscillation point is as follows: ＞10t； Where h represents the height; w represents the width of each height layer; and t represents the slice thickness.

6. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 1, characterized in that: The specific details of estimating the canopy volume using the voxel method and the AlphaShape algorithm in step 5 are as follows: Before extracting canopy volume using the voxel method and the AlphaShape algorithm, sensitivity tests were performed on BLS data for voxel size and Alpha value. Canopy volumes were extracted for voxel sizes of 0.2, 0.3, 0.4, 0.5, and 0.6, and Alpha values ​​of 0.3, 0.5, 1, 2, and 5, respectively. The average error of different methods using different parameters for canopy volume extraction was then compared. The combination of voxel size and Alpha value with the smallest average error was selected for canopy volume extraction from BLS point clouds.

7. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 1, characterized in that: Step 6, which involves reconstructing the tree branch and trunk structure using the quantitative structural model (QSM) to estimate the volume of individual timbers, includes the following: The core of quantitative structural models lies in backbone segmentation and topological reconstruction: First, the elevation data of the point cloud is used to detect the orientation of the tree trunks; Then, the tree point cloud is layered by height to identify the main branch points and separate the trunk part; Secondly, based on the distribution and connection method of the point cloud, the model of the tree trunk component is reconstructed; Finally, the volumes of the cylindrical portions of the trunk represented in the reconstructed QSM model are added together to obtain the trunk volume.

8. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 7, characterized in that: The specific content of the trunk volume is obtained by adding up the volumes of the cylindrical parts of the trunk represented in the reconstructed QSM model: ； Where V represents the trunk volume; n represents the total number of cylinders; i represents the cylinder number; r1 i h1 represents the radius of each cylinder; i This indicates the height of each cylinder.

9. The method for estimating the volume of a single timber and the crown volume based on a single backpack lidar according to claim 1, characterized in that: The specific content of the accuracy verification in step 7 is as follows: The coefficient of determination R is used for estimating the volume of individual timbers and the canopy volume. 2 Accuracy is evaluated using Root Mean Square Error (RMSE), Root Mean Square Error (RMSE%), Bias, and Bias%; where RMSE% is the relative value of RMSE and Bias% is the relative value of Bias. R 2 The formulas for calculating RMSE, RMSE%, Bias, and Bias% are as follows: ； ； ； ； ； in, This represents the true value of the volume of a single timber or the volume of the canopy. This represents an estimate of the volume of a single timber or the canopy volume. This represents the average volume of a single tree or the canopy volume, where m represents a single tree and Bias represents the deviation.

Images