Method suitable for network reconstruction of tree branch structure

Through technologies such as foundation lidar and L1-Median algorithm, the problems of incomplete branch identification and insufficient noise resistance in tree branch structure network reconstruction are solved, and a more accurate and complete tree branch structure network reconstruction is achieved, supporting forest fines Structural research and three-dimensional model construction.

CN119048670APending Publication Date: 2024-11-29INST OF BOTANY CHINESE ACAD OF SCI +1
View PDF 0 Cites 8 Cited by

Patent Information

Application Number
CN202411081847.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-06-04
Filing Date
2024-08-08
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

In the reconstruction of tree branch structure network, the prior art problems of incomplete branch identification, insufficient noise resistance, and inaccurate thickness, thin, length and angle information, making it difficult to obtain the internal structure information of the tree canopy, resulting in model distortion and inaccurate analysis.

Method used

Point cloud data is collected by ground-based lidar, and the branch and leaf separation pre-processing is used to generate seed points and skeleton lines by denoising and branch separation and leaf separation. The L1-Median algorithm is used to connect the skeleton lines, mark the branch and trunk levels, filter and adjust branches, and use The sliding window smooths the branches, extracts unfitted skeleton lines and expands them into the tree skeleton, calculates the branch length, thickness and angles, and generates an optimized tree branch structure network.

Benefits of technology

It improves the identification accuracy and completeness of the tree branch structure network, enhances the noise resistance, ensures the detailed expressiveness of the branch structure, and can more accurately reconstruct the three-dimensional model of the tree.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048670A_ABST
    Figure CN119048670A_ABST
Patent Text Reader

Abstract

The invention discloses a method suitable for network reconstruction of a tree branch structure, and the method comprises the following operation steps: carrying out the denoising and branch and leaf separation preprocessing of tree three-dimensional point cloud data collected by a ground-based laser radar, and obtaining a branch point cloud; generating seed points according to tree structure features; an L1-Media algorithm is used to obtain a rough skeleton; marking the grade of each branch; filtering the branches by taking the branch angle and the branch length as criteria; cutting off the branches with the underfitting phenomenon according to the result of the mutation test; adjusting the positions of the branching points; smoothing the branches; extracting skeleton lines of the branch point clouds which are not fitted, and expanding the skeleton lines to the existing tree skeleton; after the optimized rough skeleton of the target branch is obtained, the length, the thickness and the angle of each branch are calculated; and finally realizing the reconstruction of the tree branch structure network. The method overcomes the problems that a traditional branch structure network reconstruction algorithm has topological errors, small branches are difficult to recognize, and child branches are thicker than parent branches.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of ecological remote sensing, and in particular to a method suitable for reconstructing a tree branch structure network. Background Art

[0002] The branch structure network is the thickness of tree branches and their arrangement in three-dimensional space. It is the result of trees adapting to the surrounding environment and reflects the ability of trees to obtain light and water resources and resist external interference. The tree branch structure network information includes branch-level structural parameters such as branch quantity, thickness, length, and angle. It is an important basis for studying tree growth and development, evaluating wood quality, and conducting forest breeding. In addition, the tree branch structure network is also an important data source for reconstructing tree three-dimensional models and building three-dimensional virtual scenes.

[0003] In the early days, tree branch structure networks were often extracted through multi-angle image data, converting multi-perspective two-dimensional photos into three-dimensional point cloud information, and then identifying branches to construct branch networks. This method is only applicable to short and simple trees, and the acquired data is easily affected by ambient light. Due to reasons such as the tree's own occlusion, this type of method can only obtain surface information, and it is difficult to obtain the internal structure information of the tree canopy, which leads to partial distortion of the model during the reconstruction process. In addition, the branch networks obtained by some methods are only close to real trees in morphology, but there is a large deviation between the thickness, length, angle and other information and the actual measurement values, which cannot be used for subsequent analysis and research, and can only be used for tree three-dimensional model reconstruction and virtual scene modeling.

[0004] Ground-based LiDAR has the characteristics of high point density and precision, and can efficiently and accurately extract the fine structural parameters of trees, providing an important data source for the construction of tree branch structure networks. Early tree branch structure network extraction algorithms based on ground-based LiDAR usually only extract the information of the main branches from point cloud data, and often ignore the small branches, or use other algorithms to add small branches to provide the required three-dimensional tree model for virtual scene modeling, such as the Dijkstra shortest path algorithm. These algorithms focus more on the effect of reconstructing tree models and ignore the accuracy of tree structure parameters.

[0005] With the rise of research on tree fine structure, higher requirements have been placed on tree branch structure network algorithms in terms of recognition, accuracy and completeness of twigs, and the ability to obtain information such as twig position, direction, radius and length.

[0006] In the past decade, the newly developed tree branch structure network extraction algorithms have all paid more attention to the integrity and accuracy of branch extraction, among which the representative algorithms are TreeQSM algorithm, L1-MST algorithm, ADtree algorithm and Laplacian-based algorithm. Although the above algorithms are widely used, there are still many problems in the actual application process, such as the omission of small branches due to parameter sensitivity; the misclassification of branches due to limited anti-noise ability; the lack of constraints on branch thickness, resulting in high-level branches being thicker than low-level branches; and the failure to consider the distribution rules of branches, resulting in some erroneous branch topology information.

[0007] Therefore, ground-based lidar data can realize the extraction of real tree branch structure network, and has good accuracy in the extraction of parameters such as branch length, branch thickness, and branch angle.

[0008] However, how to develop a new tree branch structure network reconstruction algorithm that takes into account the inherent structural characteristics of trees, has a certain degree of anti-noise ability and is suitable for slightly missing tree point clouds, and has a high recognition rate for fine branches, so as to better serve the research on forest fine structure, forest breeding, and tree three-dimensional model construction, is still a problem that needs to be solved urgently. Summary of the invention

[0009] The object of the present invention is to provide a method for reconstructing a tree trunk structure network, comprising the following steps:

[0010] The laser radar point cloud of the target tree is collected by ground-based laser radar equipment; the branch point cloud is obtained by denoising and pre-processing the current laser radar point cloud by separating branches and leaves;

[0011] According to the structural characteristics of the tree, seed points are generated from the current branch point cloud; the seed points are extended using the L1-Median algorithm to generate skeleton lines, and the adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree;

[0012] According to the rough skeleton of the current tree branches, mark the branch level; filter the branches based on the branch angle and branch length; cut off the branches with underfitting according to the results of the mutation test; adjust the position of the branch points; use the sliding window to smooth the branches according to the point-by-point angle changes on the branches; extract the skeleton lines of the unfitted branch point cloud and expand them to the existing tree skeleton to generate the optimized target branch rough skeleton;

[0013] Combined with the optimized rough skeleton of the target tree branches, the length, thickness and branch angle of the branches are calculated to produce the final tree branch structure network; the final tree branch structure network is input into the three-dimensional visualization software for reconstruction to generate a visualized three-dimensional model of the tree branches.

[0014] Preferably, the laser radar point cloud of the target tree is collected by a ground-based laser radar device; and the branch point cloud is obtained by denoising and pre-processing the current laser radar point cloud by separating branches and leaves, including the following steps:

[0015] Using ground-based LiDAR equipment, the LiDAR point cloud of the target tree is obtained through multi-station scanning; the point cloud containing only the target tree is cropped; and the noise points and leaf points in the point cloud of the target tree are removed by statistical denoising method and manual cropping to generate a branch point cloud.

[0016] Preferably, the L1-Median algorithm is used to extend the growth of the seed points to generate skeleton lines, and the adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree, including the following steps:

[0017] The branch point cloud is screened by the height of the branch point cloud; the branch point cloud is marked by using the shortest path algorithm to obtain the final selection and marking seed point;

[0018] The L1-Median algorithm is used to grow the seed points and extract the L1-Medial broken skeleton;

[0019] Connect the broken skeletons to generate a rough skeleton of the branches of the current tree.

[0020] Preferably, the selection and marking of seed points includes: sorting the branch point clouds from low to high to form a height ranking list, counting the number of all branch point clouds in the current height ranking list to obtain a total target value; selecting the branch point clouds with a total target value of 5% of the height ranking in the current height ranking list and fitting a cylinder, taking the center of the bottom surface of the cylinder as the root skeleton point; using the shortest path algorithm, calculating the path from each branch point cloud to the root skeleton point and marking it as a seed point; and marking all scanned branch point clouds with a distance less than r0 to the path as associated points, and recording them as scan points at the same time;

[0021] Extracting the L1-Medial broken skeleton includes: using the above seed points and using the L1-Median algorithm to generate a broken skeleton; the skeleton points of the L1-Medial broken skeleton are calculated using the following formula:

[0022]

[0023] Among them, x i is the three-dimensional coordinate of the skeleton point;

[0024] p j is the three-dimensional coordinate of the branch point cloud;

[0025] I and J are the serial number sets of skeleton points and branch points respectively, that is, I = {1, 2, ..., i, ..., M-1, M}, where M is the number of skeleton points;

[0026] R(X) is the penalty term calculated based on the coordinates of all skeleton points (X);

[0027] J = {1, 2, ..., j, ..., N-1, N}, where N is the number of branches; d() is the function for calculating the Euclidean distance, and θ() is the function for calculating the weight:

[0028]

[0029] Among them, h is the search range;

[0030] In the first iteration, r is the same as the search range used when generating seed points, that is, r = r0; R(X) is the penalty term calculated based on the coordinates of all skeleton points (X);

[0031] The specific formula of R(X) is as follows:

[0032]

[0033]

[0034]

[0035] In the formula, γ i is the balance coefficient; and are the three eigenvalues ​​of the matrix Ci, where is the maximum eigenvalue, is the second largest eigenvalue, is the minimum eigenvalue; σ i is the maximum eigenvalue proportion, x i To characterize the direction of the skeleton points;

[0036] The gradient descent method is used to extract the skeleton line through iteration:

[0037]

[0038]

[0039] Where, den j is the point cloud density around the j-th scanned branch point cloud, k is the number of iterations; den jBefore the iteration starts, the calculation is completed, and the search range used is r0 / 2; μ is the penalty coefficient; after each iteration, the L1-Median algorithm will compare the current search range (r) with the base radius (R) of the tree. If r < R, r will be updated to (1 + s)r and enter the next iteration; otherwise, the L1-Median algorithm ends; s is the iteration speed. After the above operations, the L1-Medial broken skeleton can be obtained.

[0040] Connect the broken skeletons to generate a rough skeleton of the branches of the current tree, including: connecting the endpoints of the skeleton lines with a distance less than d cmin and avoiding existing loops by extracting the minimum spanning tree of each connected component; the skeleton line containing the root node is regarded as the target, and the remaining skeleton lines that meet the angle and point cloud density are regarded as sources. According to the maximum connection distance d of the connected component cmin , a continuous rough skeleton is generated.

[0041] Preferably, according to the rough skeleton of the branches of the current tree, mark the branch levels; filter the branches based on the branching angle and branch length; truncate the branches with underfitting phenomena according to the results of mutation testing; adjust the positions of the branching points; smooth the branches using a sliding window according to the angular changes of the points on the branches; extract the skeleton lines of the non-fitted branch point clouds and extend them to the existing tree skeletons to generate an optimized rough skeleton of the target branches, including the following operation steps:

[0042] Mark the branch levels for the rough skeleton of the branches according to the branching of the root skeleton points and the root skeleton points;

[0043] Filter the branches of the rough skeleton of the branches to obtain updated branch levels;

[0044] Fit the skeleton line points at the ends of the branches, and prune the ends of the branches that do not meet the conditions;

[0045] Adjust the branching positions of the line segments of the skeleton lines of the pruned branches and the scanned branch point clouds;

[0046] Smooth the branching of the skeleton point parameters of the branches with adjusted branching positions;

[0047] Extract the skeleton lines of the non-fitted branch point clouds and use the minimum spanning tree algorithm to generate an optimized rough skeleton of the target branches.

[0048] Preferably, marking the branch levels includes: regarding the branch where the root skeleton point is located as the first-level branch, and the branches on the first-level branch as the second-level branches, and marking the branch levels according to the above rules;

[0049] The branch filtering process includes: removing unreasonable branches based on the branching angle and branch length and synchronously updating the branch levels;

[0050] Pruning, including: fitting branch thickness along the skeleton line points of the terminal branches, using the sliding t-test to identify the location where the branch thickness changes suddenly, and truncating the under-fitted terminal branches;

[0051] Adjust the branch position, including: determining the most suitable branch point position by comparing the total length of the line segments of the affected skeleton line before and after the current branch point is replaced by the remaining skeleton points and the average distance from the scanned branch point cloud to the line segment of the skeleton line. The specific judgment formula is as follows:

[0052]

[0053] Where l and l′ are the total lengths of the line segments of the affected skeleton line before and after the replacement, respectively; d and d′ are the average distances from the scanned branch point cloud to the line segments of the skeleton line before and after the replacement, respectively;

[0054] Smoothing branches, including: sorting all scanned branch point clouds associated with the current branch, and smoothing branches according to parameters such as skeleton point density, angle, average distance, etc. between any two adjacent skeleton lines;

[0055] Expanding branches, including: using DBSCAN clustering to divide the unfitted branch point cloud into several connected units, the search radius and the minimum number of points required to form a cluster are set to R and 5 respectively; performing k-means clustering with a cluster number of 1 to n on the scanned branch point cloud of each part; determining the local root point and the initial search direction; the local root point is the cluster center closest to the existing skeleton point, and the initial search direction is from the skeleton point closest to the local root point to the local root point; finding potential connections between cluster centers, using the minimum spanning tree algorithm, expanding it to the existing tree skeleton, and generating an optimized target branch rough skeleton.

[0056] Preferably, the method of combining the optimized rough skeleton of the target tree branches, calculating the branch length, thickness and branch angle, and producing the final tree branch structure network includes the following steps:

[0057] The length of the branch was calculated by the length of its skeleton line;

[0058] The branch thickness is calculated by the average distance from its associated scanned branch point cloud to the skeleton line;

[0059] The branch angle is calculated by the angle between the branched sub-branch and the skeleton line of the parent branch.

[0060] Preferably, the final tree branch structure network is input into three-dimensional visualization software for reconstruction to generate a visualized three-dimensional model of tree branches; wherein the three-dimensional visualization software includes a mayavi package.

[0061] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0062] By analyzing the specific scheme of the method for tree trunk structure network reconstruction provided by the present invention, it can be known that in specific application;

[0063] The laser radar point cloud of the target tree is collected by ground-based laser radar equipment; the branch point cloud is obtained by denoising and pre-processing the current laser radar point cloud by separating branches and leaves; seed points are generated from the current branch point cloud according to the structural characteristics of the tree; the L1-Median algorithm is used to extend the growth of the seed points to generate skeleton lines, and the adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree; the branch level is marked according to the rough skeleton of the branches of the current tree; branches are filtered based on the branch angle and branch length; branches with underfitting phenomena are truncated according to the results of the mutation test; the position of the branch point is adjusted; the branches are smoothed according to the point-by-point angle changes on the branches using a sliding window; the skeleton lines of the unfitted branch point cloud are extracted and extended to the existing tree skeleton to generate an optimized rough skeleton of the target branch;

[0064] Combined with the optimized rough skeleton of the target tree branches, the length, thickness and branch angle of the branches are calculated to produce the final tree branch structure network; the final tree branch structure network is input into the three-dimensional visualization software for reconstruction to generate a visualized three-dimensional model of the tree branches.

[0065] Specifically, ground-based lidar is used to collect point cloud data of vegetation; vegetation data refers to three-dimensional point cloud data of trees or three-dimensional point cloud data of seedlings; of course, the three-dimensional point cloud data of trees is obtained after collection and screening; and the influence of noise points needs to be eliminated to improve the accuracy and reliability of the data; at the same time, by separating branches and leaves, the branch structure and shape of the tree can be displayed more clearly, providing strong support for subsequent tree modeling, growth analysis and other work.

[0066] Furthermore, considering that the branches of trees are connected to the bottom of the trunk by a path, generating seed points along this path ensures the accuracy of skeleton generation; the seed points select branch point clouds within a certain range as associated point clouds, which can accurately capture the structural information of tree branches; the L1-Median algorithm can efficiently extend the growth of seed points and generate skeleton lines, avoiding complex calculations and iterative processes and improving processing efficiency; at the same time, the minimum spanning tree algorithm can quickly connect the skeleton lines and generate a rough skeleton of the tree branches, further improving the efficiency of the overall processing flow.

[0067] Furthermore, based on the multiple branches of the rough skeleton of the tree branches; mark the level of each branch; use the branch angle and branch length as the criteria to filter the branches, which can remove the branches that do not conform to the actual growth law and improve the accuracy of the tree structure; truncate the branches that are underfitting. In the branch point cloud data, the feature extraction of some branches may be inaccurate or not completely denoised. Therefore, underfitting may occur when fitting these branches, so truncating the branches with underfitting improves the fitting and enhances the anti-noise ability; adjust the position of the branch point; use the sliding window method to smooth the branches according to the point-by-point angle changes on the branches to make them closer to the true The real shape; adjusting the position of the branching point and smoothing the branches can further refine the branch structure, making it closer to the shape and characteristics of the real tree, and can effectively handle the slight angle changes on the branches, improve the detail expression of the model, and ensure that the fine branches have a high recognition rate; extract the skeleton line of the branch point cloud that has not been fitted, and expand it to the existing tree skeleton, supplement and improve the original skeleton structure, improve the integrity of the tree model, and obtain the optimized target branch rough skeleton (solving the problem of slightly missing tree point clouds). Calculate the length, thickness and angle of each branch, so as to finally realize the reconstruction of the tree branch structure network, and on this basis, realize the construction of the tree 3D model;

[0068] Therefore, the method for reconstructing the tree trunk structure network adopted in the embodiment of the present invention overcomes the technical defects of the traditional processing methods pointed out in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 is a flowchart of a method for reconstructing a tree branch structure network;

[0071] Figure 2 A schematic diagram of tree-based laser radar point cloud data and branch point cloud data obtained after preprocessing provided by an embodiment of the present invention;

[0072] Figure 3 A schematic diagram of a rough skeleton of branches of a current tree extracted from the branch point cloud data provided by an embodiment of the present invention;

[0073] Figure 4 A schematic diagram of a broken skeleton for extracting branch point cloud data provided by an embodiment of the present invention;

[0074] Figure 5 A schematic diagram of the optimization process of the rough skeleton provided in an embodiment of the present invention;

[0075] Figure 6 A schematic diagram of a three-dimensional model generated by branch network data provided by an embodiment of the present invention;

[0076] Figure 7 A model map suitable for the network reconstruction method of tree branch structure. DETAILED DESCRIPTION

[0077] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] The present invention is further described in detail below through specific implementation examples in conjunction with the accompanying drawings.

[0079] Combination Figure 1 As shown, the present invention provides a method suitable for tree branch structure network reconstruction, which mainly includes four processing steps:

[0080] S10: Collecting a laser radar point cloud of a target tree (the trees in the embodiment of the present application include trees and seedlings) by using a ground-based laser radar device; obtaining a branch point cloud by performing denoising and branch-leaf separation preprocessing on the current laser radar point cloud;

[0081] S20: Generate seed points from the current branch point cloud according to the structural characteristics of the tree (i.e., the tree); use the L1-Median algorithm to extend the growth of the seed points to generate skeleton lines, select adjacent skeleton line endpoints and combine the minimum spanning tree algorithm to connect the skeleton lines to generate a rough skeleton of the branches of the current tree;

[0082] S30: Mark the branch level according to the rough skeleton of the current tree branch; filter the branches based on the branch angle and branch length; cut off the branches with underfitting according to the result of the mutation test; adjust the position of the branch point; use the sliding window to smooth the branches according to the point-by-point angle change on the branch; extract the skeleton line of the unfitted branch point cloud, and expand it to the existing tree skeleton to generate the optimized target branch rough skeleton;

[0083] That is, through the above-mentioned adjustments, filtering and screening of branches, and smoothing operations, the final optimized target branch rough skeleton is optimized; finally, the optimized target branch rough skeleton is combined to calculate the length and thickness of each branch to produce the final tree trunk structure network; combined with the branch network information, the three-dimensional model of the tree trunk can be reconstructed.

[0084] S40: Calculate the length, thickness and angle of branches in combination with the optimized rough skeleton of the target tree branches to produce a final tree trunk structure network; input the final tree trunk structure network into a three-dimensional visualization software for reconstruction to generate a visualized three-dimensional model of tree trunks.

[0085] First, we need to use ground-based laser radar to collect vegetation data; the vegetation data mentioned here refers to the three-dimensional point cloud data of trees or the three-dimensional point cloud data of seedlings; of course, the three-dimensional point cloud data of trees is obtained after collection and screening; among them, the three-dimensional point cloud data of trees, after the pre-processing steps of denoising and branch and leaf separation, obtains the branch point cloud;

[0086] According to the structural characteristics of trees (such as branch shape, branch intersections, etc.), all branches of the tree are connected to the bottom of the trunk by a path; therefore, seed points are generated along the current path, and the branch point cloud within a certain range of the seed point is selected as the associated point cloud; the L1-Median algorithm is used to extend the growth of the seed point to generate a skeleton line, and the adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree (that is, the preliminary branch skeleton of the current tree is obtained);

[0087] Among them, the rough skeleton of the branches of the current tree includes multiple branches (branches); the level of each branch is marked; the branches are filtered based on the branch angle and branch length; the branches with underfitting phenomenon are truncated according to the results of the mutation test (the branches with underfitting phenomenon are truncated. In the branch point cloud data, the feature extraction of some branches may be inaccurate or not completely denoised. Therefore, underfitting phenomenon may occur when fitting these branches, so the branches with underfitting phenomenon are truncated to improve the fitting); the position of the branch point is adjusted; the branches are smoothed (that is, the sliding window method is used to smooth the branches according to the point-by-point angle changes on the branches to make them closer to the real shape); the skeleton lines of the unfitted branch point cloud are extracted, and extended to the existing tree skeleton; after obtaining the optimized target branch rough skeleton, the length, thickness and angle of each branch are calculated; finally, the reconstruction of the tree branch structure network is realized, and on this basis, the construction of the tree three-dimensional model is realized.

[0088] like Figure 2As shown, during the execution of S10, the laser radar point cloud of the target tree is collected by the ground-based laser radar equipment; the branch point cloud is obtained by performing denoising and branch-leaf separation preprocessing on the current laser radar point cloud, including the following operation steps:

[0089] Using ground-based LiDAR equipment, the LiDAR point cloud of the target tree is obtained through multi-station scanning; the point cloud containing only the target tree is cropped; and the noise points and leaf points in the point cloud of the target tree are removed by statistical denoising method and manual cropping to generate a branch point cloud.

[0090] It should be noted that in the process of generating point clouds, it is necessary to undergo preprocessing to remove noise points and leaf points;

[0091] Among them, the noise points are caused by the shaking of branches and multipath effects during the acquisition of ground-based lidar data. They can be processed by a statistical denoising method to calculate the spatial distance distribution between the overall data point and the remaining branch point clouds in its neighborhood. If the spatial distance between a branch point cloud and the remaining branch point clouds in its neighborhood is greater than the mean value + 3 times the standard deviation, it can be considered as a noise point and removed.

[0092] Branch and leaf separation is mainly to identify which points in the point cloud belong to branches or leaves. It can be performed using a threshold-based branch and leaf separation algorithm and manual editing: calculate the line features of the branch point cloud, remove some leaves according to the threshold, and then further remove the remaining leaves through manual editing, and manually add points to the broken parts of the branch point cloud to generate a branch point cloud.

[0093] During the execution of S20, the L1-Median algorithm is used to extend the growth of the seed points to generate skeleton lines. The adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree. The specific operation steps are as follows:

[0094] like Figure 3 As shown, the embodiment of the present invention uses the branch point cloud combined with the rough skeleton extraction module proposed in the present invention to realize the extraction of the rough skeleton. The specific process mainly includes generating seed points, extracting L1-Medial broken skeletons and connecting broken skeletons;

[0095] S21: filtering the branch point cloud by the height of the branch point cloud; marking the branch point cloud by using the shortest path algorithm to obtain the final selection and marking seed point;

[0096] S22: Use L1-Median algorithm to grow the seed points and extract L1-Medial broken skeleton;

[0097] S23: Connect the broken skeletons to generate a rough skeleton of the branches of the current tree.

[0098] The seed point is the premise for extracting the L1-Medial broken skeleton. The present invention selects the top 5% of the branch point cloud to fit the cylinder, and takes the center of its bottom surface as the root skeleton point; uses the shortest path algorithm to calculate the path from each point (branch point cloud) to the root skeleton point and marks it as a seed point; and marks all scan points (also branch point clouds) with a distance less than r0 to the path as associated points;

[0099] Use the L1-Median algorithm to generate a broken skeleton (that is, due to insufficient point cloud density or missing points, the skeleton directly extracted by the L1-Median algorithm is discontinuous and disconnected, so it is called a broken skeleton (that is, the branches of the tree may block each other, resulting in the laser radar being unable to scan some areas of some branches. The blocked areas will be missing when generating branch point clouds, which may cause these areas to be unable to generate continuous skeleton segments when extracting the skeleton, forming a broken skeleton)); select adjacent skeleton line endpoints (that is, select the endpoints of each skeleton line and select all adjacent skeleton line endpoints), and connect the broken skeletons by extracting the minimum spanning tree of each connected unit (the connected unit refers to the space between the two skeleton line endpoints, and the space contains scattered point clouds, and then the minimum spanning tree algorithm is used to connect the two skeleton line endpoints) to form a rough skeleton of the current tree branch (use the minimum spanning tree algorithm to connect all the selected skeleton line endpoints to form a complete connected tree branch skeleton (that is, connect the broken skeletons to form a rough skeleton representing the tree branch structure)), such as Figure 4 shown.

[0100] The selection and marking of seed points: sort the branch point clouds from low to high, select the branch point clouds with the top 5% height and fit a cylinder, and take the center of the bottom surface of the cylinder as the root skeleton point; use the shortest path algorithm to calculate the path from each point (branch point cloud) to the root skeleton point and mark it as a seed point; and mark all scan points (also branch point clouds) with a distance less than r0 to the path as associated points. r0 is the initial search radius, which is generally set to a multiple of the base radius (R).

[0101] Extract L1-Medial fragmented skeleton: Using the above seed points, the L1-Median algorithm (Huang et al., 2013) is used to generate the fragmented skeleton. The skeleton points of the L1-Medial fragmented skeleton are calculated using the following formula:

[0102]

[0103] Among them, x i is the three-dimensional coordinate (row vector) of the skeleton point;

[0104] pj is the three-dimensional coordinate (row vector) of the scanning point;

[0105] I and J are the serial number sets of skeleton points and branch points respectively, that is, I = {1, 2, ..., i, ..., M-1, M}, where M is the number of skeleton points;

[0106] J = {1, 2, ..., j, ..., N-1, N}, where N is the number of branches; d() is the function for calculating the Euclidean distance, and θ() is the function for calculating the weight: Where h is the search range.

[0107] In the first iteration, r is the same as the search range used when generating the seed point, that is, r = r0; R(X) is the penalty term calculated based on the coordinates of all skeleton points (X). The specific formula of R(X) is as follows:

[0108]

[0109]

[0110]

[0111] In the formula, γ i is the balance coefficient, which ensures that the penalty coefficient μ is a constant for all skeleton points; and are the three eigenvalues ​​of the matrix Ci, where is the maximum eigenvalue, is the second largest eigenvalue, is the minimum eigenvalue; σ i is the maximum eigenvalue ratio, which can characterize the skeleton point x i The degree of directionality at x i to characterize the direction of the skeleton points).

[0112] The gradient descent method is used (that is, the density unevenness of the point cloud needs to be considered at the same time) to extract the skeleton line through multiple iterations:

[0113]

[0114] den j =1+∑ j∈J\{j} d(p j ,p j' ); Formula 6

[0115] Where, den j is the point cloud density around the jth scanning point, and k is the number of iterations. jBefore the iteration starts, the calculation is completed, and the search range used is r0 / 2. μ is the penalty coefficient, and it is recommended to be set to 0.35. After each iteration, the L1-Median algorithm will compare the current search range (r) with the base radius (R) of the tree. If r < R, r will be updated to (1 + s)r and enter the next iteration; otherwise, the L1-Median algorithm ends; s is the iteration speed, that is, the multiple by which the search radius increases after each iteration, and it is recommended to be set to 0.4; after the above operations, the L1-Medial broken skeleton can be obtained.

[0116] Connecting the broken skeleton: Since the L1-Medial broken skeleton generated in the previous step is usually broken, the above method for connecting the broken skeleton adopted in the embodiment of the present invention establishes the principle of connecting the broken skeleton according to the characteristics of the tree. Connect the endpoints of the skeleton lines with a distance less than d cmin and avoid possible loops by extracting the minimum spanning tree of each connected component (that is, when connecting the skeleton points in the connected component using other algorithms, it may cause the formation of a closed loop. As shown in Figure 4 , such loops do not appear in the branches of the tree, so the minimum spanning tree algorithm is used to avoid loop closure); the skeleton line containing the root node is regarded as the target (the skeleton line containing the root node is regarded as the target, mainly to set the starting point of the connection. The skeleton line containing the root node is the main trunk of the tree, and it is convenient and reasonable to find the broken skeleton connected to it starting from it; therefore, the other skeleton lines with angles and point densities are regarded as sources), and the remaining skeleton lines that meet the angle and point cloud density are regarded as sources, and try to connect the connected components that are farther apart (the maximum connection distance is set to d cmax ), generating a continuous rough skeleton (that is, for example: the included angle between the skeleton line to be connected and the connected skeleton should be greater than 10 degrees and less than 120 degrees; in terms of density, it is to control the point cloud density on the connection line, which cannot be less than 1 / 3 of the point cloud density of the connected component).

[0117] That is, select the branch point cloud in the top 5% in height and fit a cylinder, and take the center of the bottom surface of the cylinder as the root skeleton point; use the shortest path algorithm to calculate the path from each point to the root skeleton point and mark it as a seed point; use the above seed points and generate a broken skeleton using the L1-Median algorithm; select the adjacent skeleton line endpoints and connect the broken skeletons by extracting the minimum spanning tree of each connected component to form a rough skeleton;

[0118] In the process of step S30, the optimization process of the rough skeleton of the branches of the current tree is involved;

[0119] As shown in Figure 5 , in the embodiment of the present invention, the rough skeleton of the branches of the tree is combined with the skeleton optimization module proposed in the present invention for skeleton line optimization.

[0120] That is, the step of optimizing the rough skeleton can be divided into the following steps:

[0121] S31: branch level marking, branch filtering, branch pruning, branch position adjustment, branch smoothing, branch expansion;

[0122] In the above, the rough skeleton of the branch is marked with branch levels according to the root skeleton points and the branches of the root skeleton points;

[0123] Filter branches from the rough skeleton of the branch to obtain an updated branch grade;

[0124] Fit the skeleton line points at the ends of the branches and prune the ends of the branches that do not meet the conditions;

[0125] Adjust the branch position based on the line segments of the skeleton line of the pruned branches and the scanned branch point cloud;

[0126] Smooth branching is performed on the skeleton point parameters of the branches that adjust the branch positions;

[0127] The skeleton lines of the unfitted branch point cloud are extracted, and the minimum spanning tree algorithm is used to generate the optimized rough skeleton of the target branch.

[0128] S32: Branch level marking: The branch where the root skeleton point is located is the first-level branch, and the branches on the first-level branch are second-level branches. The branch level is marked according to the above rules;

[0129] S33: Filter branches: remove unreasonable branches based on branch angle and branch length, and update the branch level synchronously;

[0130] S34: Pruning: Due to the low density of scan points around the terminal branches, the rough skeleton line may be underfitted. The skeleton line points along the terminal branches (i.e., the skeleton line points are extracted through the branch point cloud, and the skeleton line points include bifurcation points (points connecting different skeleton branches) and endpoints (end points of skeleton branches)) are fitted to the branch thickness, and the sliding t-test is used to identify the location where the branch thickness mutation occurs, and the underfitting terminal branches are truncated;

[0131] S35: Adjusting the branch position: The most suitable branch point position is determined by comparing the total length of the line segments of the affected skeleton line before and after the current branch point is replaced by the remaining skeleton points and the average distance from the scanning point to the line segment of the skeleton line. The specific judgment formula is as follows:

[0132]

[0133] Where l and l' are the total lengths of the affected skeleton line segments before and after the replacement, respectively; d and d' are the average distances from the scan points to the skeleton line segments before and after the replacement, respectively. During the adjustment process, the angle and density constraints of the skeleton line are still considered;

[0134] S36: Smooth branches: Avoid too large an angle between two adjacent skeleton lines. Sort all scanned branch point clouds associated with the current branch, and smooth the branches according to the skeleton point density, angle, average distance and other parameters between any two adjacent skeleton lines. For example, the angle between three adjacent skeleton points is less than 45°, and the skeleton points on the skeleton line are ensured to have a point every 0.03m;

[0135] S37: Expand branches: Use DBSCAN clustering to divide the unfitted branch point cloud into several connected units, and the search radius and the minimum number of points required to form a cluster are set to R and 5 respectively; perform k-means clustering with a cluster number of 1 to n on the scanned branch point cloud of each part; determine the local root point and the initial search direction; the local root point is the cluster center closest to the existing skeleton point, and the initial search direction is from the skeleton point closest to the local root point to the local root point; find potential connections between cluster centers, use the minimum spanning tree algorithm, expand them to the existing tree skeleton, and generate an optimized target branch rough skeleton.

[0136] It should be noted that, according to the rough skeleton of the branches of the current tree, the branch where the root skeleton point is located is the first-level branch, and the branches on the first-level branch are the second-level branches. The branch levels are marked according to the above rules; according to the angle between the child branch and the parent branch (that is, the child branch and the parent branch refer to the branch of the tree branch) and the length of the branch, the child branches whose angles and lengths do not meet the threshold are removed (the branches that do not meet the conditions are removed according to the branch angle and branch length); the sliding t-test is used to identify the location where the branch thickness mutation occurs (that is, if in the t-test of a certain window, it is found that the difference between the data inside the window and outside the window is significant (that is, the t value exceeds the critical value corresponding to the set significance level), it is considered that the branch thickness mutation occurs at the end position of the window; all detected mutation point locations are recorded (usually the index of the skeleton line point or the corresponding physical location); according to the results of the mutation test, it can be decided to truncate the terminal branches that are considered to be underfitting; for example: if after the 90th mutation point The branch part that is considered abnormal or underfitting, and the branches after this point can be truncated), and the underfitting terminal branches are truncated (that is, due to the low density of scanning points around the terminal branches, the rough skeleton line may be underfitting (that is, the underfitting phenomenon is: the branch structure is incomplete, the branches are ignored or lost during the extraction process; the branch shape is inaccurate, and there is a significant difference between the branch shape and the branch shape of the actual tree; the branch point position is inaccurate; the bifurcation and bending of twigs and twigs fail to fully retain these detailed information; all of which will lead to underfitting); therefore, a sliding t-test is used to identify the location where the branch thickness mutation occurs, and these underfitting terminal branches are truncated); by comparing the total length of the affected skeleton line segments before and after replacing the current branch point with other skeleton points and the average distance from the scanning point to these segments, the most suitable branch point position can be determined, which can effectively optimize the branch structure and make it more consistent with the shape of the actual tree;

[0137] Finally, all the scan points associated with the current branch are sorted, and the branches are smoothed according to parameters such as point density, angle, and average distance. This can eliminate mutations in the skeleton line and make the skeleton line smoother and more continuous. The point cloud of the unfitted branches is extracted and divided into several connected units. Each connected unit is clustered to find potential connections between cluster centers. The minimum spanning tree algorithm is used to expand it to the existing tree skeleton to generate an optimized rough skeleton of the target branch.

[0138] From the analysis of the above steps, we can see that: according to the rough skeleton of the branches of the current tree, the branch where the root skeleton point is located is the first-level branch, and the branches on the first-level branch are the second-level branches. The branch levels are marked according to the above rules; according to the angle between the child branch and the parent branch and the branch length, the child branches whose angle and length do not meet the threshold are removed; the sliding t-test is used to identify the location where the branch thickness changes suddenly, and the terminal branches with underfitting are truncated;

[0139] The most suitable branch point position is determined by comparing the total length of the affected skeleton line segments before and after the current branch point is replaced by other skeleton points and the average distance from the scanning point to these segments;

[0140] Sort all scan points associated with the current branch, and smooth the branches according to parameters such as point density, angle, and average distance;

[0141] Extract the unfitted branch point cloud and divide it into several connected units. Cluster each connected unit to find the potential connection between cluster centers. Use the minimum spanning tree algorithm to expand it to the existing tree skeleton and generate an optimized rough skeleton of the target branch.

[0142] like Figure 6 As shown, this example combines the rough skeleton with the skeleton optimization module proposed in the present invention to achieve skeleton optimization.

[0143] Finally, if Figure 7 As shown in the figure, this example uses the optimized skeleton combined with point cloud to calculate the length, angle and thickness of each branch, generates the final branch structure network and constructs a three-dimensional model. The average distance from the branch-related point cloud to the skeleton line is used as the branch thickness, the length of its skeleton line is used as the branch length, and the angle between it and the parent branch is used as the branch angle. Finally, Mayavi software is used to map and display the model.

[0144] Analysis of the above steps shows that: further, the optimized skeleton is subjected to parameter extraction steps, including calculation of branch length, thickness and angle. The branch length is calculated by the length of its skeleton line; the branch thickness is calculated by the average distance from its associated scanning point to the skeleton line; the branch angle is calculated by the angle between the child branch and the parent branch skeleton line. By inputting the obtained branch network structure into a three-dimensional visualization software, such as the mayavi package, a visualized three-dimensional model of the tree can be obtained.

[0145] To sum up, the example of the present invention proposes a method suitable for reconstructing the tree branch structure network, by using a ground-based lidar to collect vegetation data; the vegetation data mentioned herein refers to the three-dimensional point cloud data of trees or the three-dimensional point cloud data of seedlings; of course, the three-dimensional point cloud data of trees is obtained after main collection and screening; and it is also necessary to eliminate the influence of noise points to improve the accuracy and reliability of the data; at the same time, by separating branches and leaves, the branch structure and shape of the tree can be displayed more clearly, providing strong support for subsequent tree modeling, growth analysis and other work.

[0146] Furthermore, considering that the branches of trees are connected to the bottom of the trunk by a path, generating seed points along this path ensures the accuracy of skeleton generation; the seed points select branch point clouds within a certain range as associated point clouds, which can accurately capture the structural information of tree branches; the L1-Median algorithm can efficiently extend the growth of seed points and generate skeleton lines, avoiding complex calculations and iterative processes and improving processing efficiency; at the same time, the minimum spanning tree algorithm can quickly connect the skeleton lines and generate a rough skeleton of the tree branches, further improving the efficiency of the overall processing flow.

[0147] Furthermore, according to the multiple branches of the rough skeleton of the tree branches; mark the level of each branch; use the branch angle and branch length as the criterion to filter the branches, which can remove the branches that do not conform to the actual growth law and improve the accuracy of the tree structure; truncate the branches that are underfitting. In the branch point cloud data, the feature extraction of some branches may be inaccurate or not completely denoised. Therefore, underfitting may occur when fitting these branches, so the branches with underfitting are truncated to improve the fitting; adjust the position of the branch point; use the sliding window method to adjust the angle change point by point on the branch. Smoothing branches to make them closer to the real shape; adjusting the position of branch points and smoothing branches can further refine the branch structure, making it closer to the shape and characteristics of real trees, and can effectively handle subtle angle changes on branches and improve the detail expression of the model; extracting the skeleton lines of the unfitted branch point cloud, and extending them to the existing tree skeleton, supplementing and improving the original skeleton structure, and improving the integrity of the tree model; after obtaining the optimized target branch rough skeleton, calculate the length, thickness and angle of each branch; finally, the tree branch structure network is reconstructed, and on this basis, the three-dimensional model of the tree is constructed.

Claims

1. A method for reconstructing a tree branch structure network, characterized in that: The following steps are included: The laser radar point cloud of the target tree is collected by ground-based laser radar equipment; the branch point cloud is obtained by denoising and pre-processing the current laser radar point cloud by separating branches and leaves; According to the structural characteristics of the tree, seed points are generated from the current branch point cloud; the seed points are extended using the L1-Median algorithm to generate skeleton lines, and the adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree; According to the rough skeleton of the current tree branches, mark the branch level; filter the branches based on the branch angle and branch length; cut off the branches with underfitting according to the results of the mutation test; adjust the position of the branch points; use the sliding window to smooth the branches according to the point-by-point angle changes on the branches; extract the skeleton lines of the unfitted branch point cloud and expand them to the existing tree skeleton to generate the optimized target branch rough skeleton; Combined with the optimized rough skeleton of the target branches, the branch length, thickness and branch angle are calculated to produce the final tree branch structure network; The final tree trunk structure network is input into a three-dimensional visualization software for reconstruction to generate a visualized three-dimensional tree trunk model.

2. A method for tree trunk structure network reconstruction as claimed in claim 1, characterized in that: The laser radar point cloud of the target tree is collected by ground-based laser radar equipment; the branch point cloud is obtained by denoising and pre-processing the current laser radar point cloud by separating branches and leaves, including the following steps: Using ground-based LiDAR equipment, the LiDAR point cloud of the target tree is obtained through multi-station scanning; the point cloud containing only the target tree is cropped; and the noise points and leaf points in the point cloud of the target tree are removed by statistical denoising method and manual cropping to generate a branch point cloud.

3. A method for reconstructing a tree trunk structure network as claimed in claim 2, characterized in that: The L1-Median algorithm is used to extend the growth of the seed points to generate skeleton lines. The adjacent skeleton line endpoints are selected and connected with the minimum spanning tree algorithm to generate a rough skeleton of the branches of the current tree, including the following steps: The branch point cloud is screened by the height of the branch point cloud; the branch point cloud is marked by using the shortest path algorithm to obtain the final selection and marking seed point; The L1-Median algorithm is used to grow the seed points and extract the L1-Medial broken skeleton; Connect the broken skeletons to generate a rough skeleton of the branches of the current tree.

4. A method for reconstructing a tree trunk structure network as claimed in claim 3, characterized in that ; The selection and marking of seed points includes: sorting the branch point clouds from low to high to form a height ranking list, counting the number of all branch point clouds in the current height ranking list to obtain a total target value; selecting the branch point clouds with a total target value of 5% of the height ranking in the current height ranking list and fitting a cylinder, taking the center of the bottom surface of the cylinder as the root skeleton point; using the shortest path algorithm, calculating the path from each branch point cloud to the root skeleton point and marking it as a seed point; and marking all scanned branch point clouds with a distance less than r0 to the path as associated points, and recording them as scan points at the same time; Extracting the L1-Medial broken skeleton includes: using the above seed points and using the L1-Median algorithm to generate a broken skeleton; the skeleton points of the L1-Medial broken skeleton are calculated using the following formula: Among them, x i is the three-dimensional coordinate of the skeleton point; p j is the three-dimensional coordinate of the branch point cloud; I and J are the serial number sets of skeleton points and branch points respectively, that is, I = {1, 2, ..., i, ..., M-1, M}, where M is the number of skeleton points; R(X) is the penalty term calculated based on the coordinates of all skeleton points (X); J = {1, 2, ..., j, ..., N-1, N}, where N is the number of branches; d() is the function for calculating the Euclidean distance, and θ() is the function for calculating the weight: Among them, h is the search range; In the first iteration, r is the same as the search range used when generating seed points, that is, r = r0; R(X) is the penalty term calculated based on the coordinates of all skeleton points (X); The specific formula of R(R) is as follows: C i =∑ i'∈I{i} θ(d(x i ,x i' ))(x i -x i' ) T (x i -x i' ); In the formula, γ i is the balance coefficient; and are the three eigenvalues ​​of the matrix Ci, where is the maximum eigenvalue, is the second largest eigenvalue, is the minimum eigenvalue; σ i is the maximum eigenvalue proportion, x i To characterize the direction of the skeleton points; The gradient descent method is used to extract the skeleton line through iteration: day j =1+∑ j∈j\{j} d(p j ,p j' ); where, den j is the point cloud density around the j-th scanned branch point cloud, and k is the number of iterations; den j is calculated before the start of the iteration, and the search range used is r0 / 2; μ is the penalty coefficient; after each iteration, the L1-Median algorithm will compare the current search range (r) with the base radius (R) of the tree. If r < R, r will be updated to (1 + s)r and enter the next iteration; otherwise, the L1-Median algorithm ends; s is the iteration speed. After the above operations, the L1-Medial broken skeleton can be obtained; Connect the broken skeletons to generate a rough skeleton of the branches of the current tree, including: the connection distance is less than d cmin The skeleton line endpoints are extracted to avoid the existing loops by extracting the minimum spanning tree of each connected unit; the skeleton line containing the root node is regarded as the target, and the remaining skeleton lines that meet the angle and point cloud density are regarded as the source. cmin , generating a continuous rough skeleton.

5. A method for tree trunk structure network reconstruction as claimed in claim 4, characterized in that: According to the rough skeleton of the current tree branches, mark the branch level; filter the branches based on the branch angle and branch length; cut off the branches with underfitting according to the results of the mutation test; adjust the position of the branch point; use the sliding window to smooth the branches according to the point-by-point angle changes on the branches; extract the skeleton line of the unfitted branch point cloud and expand it to the existing tree skeleton to generate the optimized target branch rough skeleton, including the following steps: Marking the branch level of the rough skeleton of the branch according to the root skeleton point and the branch of the root skeleton point; Filter branches from the rough skeleton of the branch to obtain an updated branch grade; Fit the skeleton line points at the ends of the branches and prune the ends of the branches that do not meet the conditions; Adjust the branch position based on the line segments of the skeleton line of the pruned branches and the scanned branch point cloud; Smooth branching is performed on the skeleton point parameters of the branches that adjust the branch positions; The skeleton lines of the unfitted branch point cloud are extracted, and the minimum spanning tree algorithm is used to generate the optimized rough skeleton of the target branch.

6. A method for reconstructing a tree trunk structure network as claimed in claim 5, characterized in that ; Marking the level of branches and trunks, including: taking the branch where the root skeleton point is located as the first-level branch, and the branches on the first-level branch as the second-level branch, and marking the level of branches and trunks according to the above rules; Filter branch processing, including: removing unreasonable branches based on branch angle and branch length, and updating the branch level synchronously; Pruning, including: fitting branch thickness along the skeleton line points of the terminal branches, using the sliding t-test to identify the location where the branch thickness changes suddenly, and truncating the under-fitted terminal branches; Adjust the branch position, including: determining the most suitable branch point position by comparing the total length of the line segments of the affected skeleton line before and after the current branch point is replaced by the remaining skeleton points and the average distance from the scanned branch point cloud to the line segment of the skeleton line. The specific judgment formula is as follows: Where l and l′ are the total lengths of the line segments of the affected skeleton line before and after the replacement, respectively; d and d′ are the average distances from the scanned branch point cloud to the line segments of the skeleton line before and after the replacement, respectively; Smoothing branches, including: sorting all scanned branch point clouds associated with the current branch, and smoothing branches according to parameters such as skeleton point density, angle, average distance, etc. between any two adjacent skeleton lines; Expanding branches, including: using DBSCAN clustering to divide the unfitted branch point cloud into several connected units, the search radius and the minimum number of points required to form a cluster are set to R and 5 respectively; performing k-means clustering with a cluster number of 1 to n on the scanned branch point cloud of each part; determining the local root point and the initial search direction; the local root point is the cluster center closest to the existing skeleton point, and the initial search direction is from the skeleton point closest to the local root point to the local root point; finding potential connections between cluster centers, using the minimum spanning tree algorithm, expanding it to the existing tree skeleton, and generating an optimized target branch rough skeleton.

7. A method for tree trunk structure network reconstruction as claimed in claim 6, characterized in that: The method of combining the optimized rough skeleton of the target tree branches, calculating the branch length, thickness and branch angle, and producing the final tree branch structure network includes the following steps: The length of the branch was calculated by the length of its skeleton line; The branch thickness is calculated by the average distance from its associated scanned branch point cloud to the skeleton line; The branch angle is calculated by the angle between the branched sub-branch and the skeleton line of the parent branch.

8. A method for tree trunk structure network reconstruction as claimed in claim 6, characterized in that: Inputting the final tree branch structure network into a three-dimensional visualization software for reconstruction to generate a visualized three-dimensional tree branch model; The 3D visualization software includes the mayavi package.

Citation Information

Cited By

  • Method and system for constructing single-tree point cloud model of power transmission line of power network

    CN119558101A

  • Tree biomass lossless estimation method and system suitable for branch level

    CN120219969A

  • Underground live-action roaming method and system based on laser point cloud

    CN122223240A

  • Fruit tree root system multi-dimensional phenotypic character extraction method based on three-dimensional point cloud

    CN122531004A

  • A tree skeleton skeleton refinement method, system, terminal and storage medium based on load and growth constraints

    CN122695117A