Branch and leaf separation method for point cloud of single tree

Abstract

The invention discloses a branch and leaf separation method for a point cloud of a single tree, relates to the technical field of forestry ecological remote sensing, and aims to solve the problem that the point cloud extraction precision and efficiency of a wood structure of a tree are not high. The method comprises the following steps: performing individual tree segmentation and denoising on original point cloud data to obtain individual tree point cloud; carrying out multi-angle rotation on the single-tree point cloud, analyzing the single-tree point cloud by using a LeWoS algorithm, and carrying out linearity constraint denoising to obtain an initial branch point cloud; dividing the initial branch point cloud into a plurality of branch structures through a clustering algorithm, and extracting end points of the branch structures through a PCA algorithm; marking unmarked point clouds in the neighborhood range of the determined branch point cloud path as endpoints of missing branches through a connectivity analysis algorithm and a judgment constraint condition, and topologically repairing the point clouds of the missing branches; and introducing a path completion mechanism based on path analysis, and communicating the endpoint of the missing branch with multiple branches with the endpoint of the initial branch point cloud to obtain a complete branch point cloud.

Description

Technical Field

[0001] The present invention relates to the technical field of forestry ecological remote sensing, and in particular to a method for separating branches and leaves of a single tree point cloud. Background Art

[0002] The structural characteristics of trees—the three-dimensional spatial distribution of branches, leaves, and inflorescences—determine their morphological structure and play a central role in their function by regulating key physiological and ecological processes, such as light interception, water and nutrient transport, and mechanical stability. Therefore, a deeper understanding of tree structure is crucial for advancing ecosystem research and achieving sustainable forest management.

[0003] In the 3D modeling process, extracting point clouds of tree wood structures (i.e., branches) is a crucial preprocessing step for model reconstruction. Extraction accuracy directly impacts the geometric realism and structural integrity of the 3D model. Over the past decade, researchers have proposed a variety of automatic segmentation algorithms based on TLS point clouds, which can be categorized into three main categories: radiometric, geometric, and graph-structure methods. Radiometric methods utilize the reflectance differences between leaves and branches in the near-infrared or short-wave infrared bands for classification, demonstrating some separation capabilities. However, these methods are significantly affected by factors such as laser incident angle, tree species, and scanning conditions, limiting their versatility. Geometric methods extract spatial distribution features of point clouds for classification, achieving high separation accuracy. However, they are sensitive to parameter settings and prone to misclassification or omission when dealing with complex branch structures. Graph-structure methods construct a connectivity graph of the point cloud and extract skeleton information, such as the shortest path. These methods offer better structural recognition capabilities, but they require high point cloud density and can easily result in fragmented or erroneous branch point clouds in areas with occlusion or missing points.

[0004] While these methods have been widely adopted, they still face challenges in addressing complex stand structures, diverse tree species, and the 3D modeling requirements they demand, including incomplete recognition, low classification accuracy, and poor structural continuity. These issues make them difficult to meet the practical demands of high-quality 3D tree modeling. Current modeling practices still rely on manual annotation to ensure the complete and accurate extraction of branch and trunk structures, resulting in high workload and low efficiency.

[0005] In summary, how to develop a branch and leaf separation algorithm with high automation, high classification accuracy, and the ability to fully extract the tree trunk structure to meet the needs of three-dimensional tree modeling based on TLS data is still an urgent problem that needs to be solved. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for separating branches and leaves from a single tree point cloud, so as to improve the extraction accuracy and efficiency of the tree wood structure point cloud.

[0007] In order to achieve the above object, the present invention provides the following technical solutions: A method for separating branches and leaves from a point cloud of a single tree, comprising: Perform single-tree segmentation and denoising on the original point cloud data of the target tree to obtain a single-tree point cloud; The single tree point cloud is rotated at multiple angles, analyzed using the LeWoS algorithm, and subjected to linear constraint denoising to obtain an initial branch point cloud; The initial branch point cloud is divided into multiple branch structures by a clustering algorithm, and the endpoints of the branch structures are extracted by a PCA algorithm; unmarked point clouds within the neighborhood of the determined branch point cloud path are marked as endpoints of missing branches by a connectivity analysis algorithm and judgment constraints, and the point clouds of the missing branches are topologically repaired; A path completion mechanism based on path analysis is introduced to connect the endpoints of the missing branches with multiple branches with the endpoints of the initial branch point cloud to obtain a complete branch point cloud.

[0008] A further technical solution is to rotate the single wood point cloud at multiple angles and analyze the single wood point cloud using the LeWoS algorithm, specifically including: The original point cloud is rotated 90° around the X-axis, Y-axis, and Z-axis respectively. The LeWoS algorithm is executed after each rotation to extract branch points, and then the branch points extracted from different angles are merged.

[0009] A further technical solution is that the linearity constraint specifically includes: The linearity evaluation index is obtained by calculating the covariance matrix of the local neighborhood of the point cloud of the target tree and performing eigenvalue decomposition. The linearity calculation formula is: ; Where: λ1 and λ2 are the maximum and second largest eigenvalues ​​of the local neighborhood covariance matrix respectively; when the linearity value is greater than the preset threshold of 0.7, it is determined to be a branch point.

[0010] A further technical solution is that the initial branch point cloud is divided into multiple branch structures by a clustering algorithm, including using the following formula: ; Where: Meanshift h ( p ) represents the mean shift vector of the current point; P Indicates the three-dimensional coordinates of the current point; P i Represents the coordinates of the neighborhood points; h The bandwidth parameter to control the clustering scale; G (·) is the distance-based Gaussian kernel function.

[0011] A further technical solution is to extract the endpoints of the branch structure using a PCA algorithm, specifically including: The PCA algorithm is used to perform principal component analysis on the point cloud cluster, extract the main direction axis, search for boundary points along the positive and negative directions of the main direction axis, and take the two points farthest from the geometric center as endpoints.

[0012] A further technical solution is that the determination constraint conditions specifically include: Point existence constraint: scan point clouds must exist between potential endpoint pairs; Path constraint: Dijkstra algorithm is used to calculate the shortest path between potential endpoint pairs. The shortest path between endpoint pairs cannot pass through existing branch point cloud clusters. Smoothness constraint: The path smoothness is evaluated by calculating the angle between the vectors of three adjacent points on the shortest path, and paths with multiple angles less than 150° are excluded.

[0013] A further technical solution is that the path completion mechanism based on path analysis specifically includes: The branch point cloud and unclassified point cloud were voxelized at a resolution of 2 cm. The shortest path was found from each voxel point, with the tree base as the root node. Each voxel point was classified into two categories: reachable points and unreachable points, depending on whether there was a shortest path from a certain point to the tree base. For the shortest path to a reachable point, if the missing condition is met, the path is determined to be a potential missing branch, and its endpoint is a missing endpoint; For unreachable points, their original resolution is restored, and then the branch structure is extracted through cluster analysis and linearity screening. Valid path segments are extracted through repeated path analysis, and missing endpoints are extracted. Connectivity analysis is performed on the missing endpoints and the existing branch points to obtain the complete leaf separation results.

[0014] A further technical solution is that the missing condition includes: The two endpoints of the shortest path are branch points; There is at least one intermediate point on the shortest path, and all of them are unmarked points; The length of the shortest path exceeds 5 cm.

[0015] Compared with the existing technology, the branch and leaf separation method of a single tree point cloud provided by the present invention has the following technical advantages: 1. The proposed automatic segmentation strategy, which combines point cloud geometric features, topological structure, and spatial distribution patterns, can achieve high-precision separation of branches and leaves in complex tree structures. Based on the initial branch points extracted by the LeWoS algorithm, this strategy integrates local spatial geometry and path analysis to construct discriminative point cloud categories without relying on artificial rules or prior scene knowledge. This strategy offers excellent versatility and robustness, making it suitable for data processing requirements across different tree species and scanning accuracies.

[0016] 2. The branch structure extraction and connection mechanism designed by the present invention, by constructing a path search method based on point existence constraints, path constraints and path smoothness constraints, can effectively identify and connect broken branch structures in scenarios such as occlusion and incomplete scanning, thereby ensuring the spatial continuity and geometric integrity of the branch point cloud.

[0017] 3. The present invention introduces a method combining mean shift clustering with linearity constraints, and combines it with an automatic extraction and connection method of broken branches based on connectivity analysis to extract and connect small branches of different scales, making the overall structure restoration more complete and significantly improving the ability to identify and extract small branches.

[0018] In summary, the automatic branch and leaf separation method proposed in the present invention not only has significant advantages in recognition accuracy, topological continuity maintenance, and detail integrity, but its multi-level path analysis mechanism effectively improves the ability to complete missing branch point clouds, and also provides a reliable data foundation and technical support for applications such as three-dimensional tree modeling, forestry resource assessment, and ecological monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 A flow chart of a method provided in an embodiment of the present application; Figure 2 A single tree point cloud data map provided by an embodiment of the present invention; Figure 3 A schematic diagram of identifying and connecting missing branches according to an embodiment of the present invention; Figure 4 A schematic diagram of extracting a small branch structure provided by an embodiment of the present invention; Figure 5 A schematic diagram of the branch point cloud extraction process provided by an embodiment of the present invention shows the gradual optimization process from initial branch point cloud extraction to the final complete branch point cloud; Figure 6 Schematic diagram of the results of the branch-leaf separation algorithm provided in an embodiment of the present invention; Figure 7 A schematic diagram of the analysis results of the branch point cloud provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To facilitate a clear description of the technical solutions of the embodiments of the present invention, the words "first" and "second" are used in the embodiments of the present invention to distinguish between identical or similar items with substantially the same functions and effects. For example, the first threshold and the second threshold are merely used to distinguish between different thresholds and do not limit their order. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0021] It should be noted that "at least one" refers to one or more, and "more than one" refers to two or more. In the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0022] This paper proposes a method for separating branches and leaves from a single tree point cloud. The method consists of three main stages: extracting and connecting branches from the initial trunk point cloud, extracting and connecting small branch point clouds, and optimizing the point cloud for missing branches. By integrating multiple strategies, including multi-view structure extraction, connectivity analysis, density-based clustering, and path analysis, the method automatically extracts structurally complete and topologically continuous tree trunk point clouds, ultimately achieving precise separation of branches and leaves.

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0024] The embodiment of the present invention provides a method for separating branches and leaves from a single tree point cloud. The main process of the method is described as follows. Figure 1 : Step S1: Perform single-tree segmentation and denoising on the original point cloud data of the target tree to obtain a single-tree point cloud; perform multi-angle rotation on the single-tree point cloud, analyze the single-tree point cloud using the LeWoS algorithm, and obtain an initial branch point cloud after denoising using linear constraints.

[0025] The original point cloud data of the target trees are obtained by scanning with a ground-based LiDAR device. Single tree segmentation is an existing technology that aims to accurately extract the three-dimensional structural parameters of a single tree from LiDAR point cloud or image data, such as Figure 2 shown.

[0026] The LeWoS algorithm is an improved method based on the k-means clustering algorithm. It uses the LeWos model to perform preliminary branch and leaf separation on the point cloud. A path-tracing detection algorithm then performs fine separation on the mixed branch and leaf point cloud. As the path length increases from 10 to 100, the number of branch points increases while the number of leaf points decreases. Ultimately, the optimal path length for the tree is selected for path-tracing measurement. To address the issues of branch breakage and directional sensitivity inherent in the LeWoS algorithm, this embodiment employs a multi-angle fusion extraction method. Specifically, the single tree point cloud is rotated 90° around the X, Y, and Z axes. The LeWoS algorithm is executed after each rotation to extract branch points. Branch points extracted from different angles are then merged.

[0027] Because the LeWoS algorithm tends to misclassify leaf points with a local planar distribution as wood points, resulting in a large number of misclassified points in the extraction results, this embodiment introduces a linearity constraint to further filter the merged branch point cloud. This linearity constraint is based on the local neighborhood structure characteristics of each point in the point cloud, and uses geometric features to determine whether it conforms to the linear form of the branch. Specifically, let the neighboring point set of a point in the point cloud be { P 1, P 2,…, P i}, then its covariance matrix C can be calculated as follows: ; Where: is the centroid of the neighborhood points.

[0028] Perform eigenvalue decomposition on the covariance matrix C and obtain three eigenvalues λ 1, λ 2, λ 3. Arrange in descending order λ 1≥ λ 2≥ λ 3. This embodiment uses the following linearity metrics to quantify the local structure of the point: ; Where: λ 1. λ2 are the maximum and second-largest eigenvalues ​​of the local neighborhood covariance matrix, respectively. When the linearity value is greater than a preset threshold, the point and its neighborhood are considered to exhibit a clear one-dimensional linear extension trend, consistent with the geometric characteristics of the branch, and the point is determined to be a branch point. Otherwise, the point is considered to likely belong to a local planar structure and is not retained. In this embodiment, the preset threshold for linearity is 0.7.

[0029] Reflects the distribution intensity of the point cluster along the main direction; Reflects the distribution intensity of the point cluster in the secondary direction; the advantage of calculating by linear threshold is that the physical meaning is clear: when the point cluster is distributed in a linear manner λ 1≫ λ 2, Linearity ( v ) approaches 1; when the point clusters are distributed in a planar manner λ 1≈ λ 2, Linearity ( v ) approaches 0. And the scale effect is eliminated by the ratio of the eigenvalues, which has scale invariance; at the same time, the square root operation has the effect of suppressing the interference of outliers and can enhance the robustness to noise.

[0030] Through the above linearity constraint processing, we can effectively remove the misjudged leaf points in the initial extraction results of the LeWoS algorithm, retaining the point cloud that is more consistent with the characteristics of the branch structure, thereby providing a more accurate basic input for subsequent structural connection and optimization. Step S2: The initial branch point cloud is divided into multiple branch structures through a clustering algorithm, the endpoints of the branch structure are extracted through the PCA algorithm, and the unmarked point clouds in the neighborhood of the determined branch point cloud path are marked as the endpoints of the missing branches through the connectivity analysis algorithm and the judgment constraint conditions, and the point clouds of the missing branches are topologically repaired.

[0031] Particularly, dividing the initial branch point cloud into a plurality of branch structures by using a clustering algorithm includes: dividing the initial branch point cloud into a plurality of branch structures by using a DBSCAN clustering algorithm.

[0032] The DBSCAN clustering algorithm is a density-based clustering algorithm that can identify clusters of any shape and handle noise points and outliers. Its core idea is to identify core points and add all points within the neighborhood of each core point to the same cluster. This process continues until no more points can be added. If a point is not within the neighborhood of any core point, it is considered a noise point.

[0033] For the branch structure: The PCA algorithm is used to extract the two end points of each branch structure as the endpoints of the structure. Specifically, the PCA algorithm is used to perform principal component analysis on the point cloud cluster, extract the main direction axis, search for boundary points along the positive and negative directions of the main direction axis, and use the two points farthest from the geometric center as endpoints to identify the growth ends of the branches and provide topological nodes for subsequent connectivity analysis.

[0034] The judgment constraints are the spatial relationship between the structural endpoints and the unmarked points, specifically including the following three points: ① Point existence constraint: There must be a scanned point cloud between the potential endpoint pairs; ② Path constraint: The Dijkstra algorithm is used to calculate the shortest path between potential endpoint pairs. The shortest path between endpoint pairs cannot pass through the existing branch point cloud clusters; ③ Smoothness constraint: The path smoothness is evaluated by calculating the vector angle formed by three adjacent points on the shortest path, and paths with multiple angles less than 150° are excluded.

[0035] The smoothness constraint can be expressed as follows: ; Where: θ i is the angle formed by three consecutive points on the path; Q i For the path i The three-dimensional coordinate vector of a point; Q i+1 and Q i+2 The three-dimensional coordinates of two subsequent consecutive points.

[0036] In step S2, the unmarked point clouds in the neighborhood of the determined branch point cloud path are marked as the endpoints of the missing branches through the connectivity analysis algorithm and the judgment constraint conditions. Specifically, for the effective shortest path connecting two extracted branch point cloud paths or endpoint pairs, the unmarked points around the path are further identified and classified to determine whether they belong to the branch structure, such as Figure 3 shown.

[0037] Taking a path with starting point A and ending point B as an example, first project all points in the vicinity of point A (including branches and unmarked points) onto a plane perpendicular to the path direction (vector AB). Then, estimate the branch radii of endpoints A and B using the following iterative method: ; in, d ai Indicates the first i The distance from point A to point A on the projection plane is k is the number of points used in each radius estimation.r i+1 -r i < 0.02 m , the iteration ends, and the current radius ri is the radius of the branch to which the endpoint A belongs. The same operation is performed on the end point B of the path, and the result is: ; Final search radius r avg Obtained by weighted average of the two end radii: ; in, a r is the adjustment coefficient. r avg The determined neighborhood range is when the unmarked points around the path are in r avg Within the determined cylindrical space, after screening through the constraint conditions, it is classified into branch points, and finally the broken branch connection of the branch structure and the topological repair of the branch structure are achieved.

[0038] In step S2, the method of dividing the initial branch point cloud into a plurality of branch structures by a clustering algorithm further includes: dividing the unclassified point cloud in the initial branch point cloud into a plurality of small branch structures by a Mean Shift clustering algorithm, so as to identify and extract other small branches except the main trunk and main branches, such as Figure 4 shown.

[0039] Mean Shift clustering algorithm is a non-parametric density-based clustering algorithm that does not require a preset number of categories and can adapt to changes in local point density. Therefore, it is particularly suitable for complex and irregular tree crown structures. h Clustering is performed to extract branch structures of different scales. The cluster center of each point is continuously moved to a higher density point until convergence: ; Where: p Indicates the three-dimensional coordinates of the current point; Meanshift h ( p ) represents the movement vector of the current point; pi represents the coordinates of the neighboring points; h The bandwidth parameter to control the clustering scale; G (·) is the distance-based Gaussian kernel function.

[0040] For small branch structures: After clustering the small branches using the Mean Shift clustering algorithm, the subsequent analysis process is the same as the branch structure analysis process described above and will not be repeated here. Finally, the small branch structure is extracted and connected to the initial branch point cloud to improve the integrity of the initial branch point cloud.

[0041] Step S3: Introduce a path completion mechanism based on path analysis to connect the endpoints of the missing branches with multiple branches with the endpoints of the initial branch point cloud to obtain a complete branch point cloud. Figure 5 shown.

[0042] Considering that endpoint extraction based on PCA can only obtain endpoints at both ends of each cluster, it is often impossible to extract all endpoints for structures with multiple branches. Therefore, an endpoint completion mechanism based on path analysis is introduced, which specifically includes the following steps: Step S31: Perform voxel processing on the branch point cloud and the unclassified point cloud with a resolution of 2 cm. Taking the tree base as the root node, find the shortest path from each voxel point. Based on whether there is a shortest path from a certain point to the tree base, each voxel point is divided into two categories: reachable point and unreachable point.

[0043] Step S32: For the shortest path to a reachable point, if there is a path segment that meets the missing condition, the path is determined to be a potential missing branch, and its endpoints are missing endpoints; wherein the missing conditions include: ① the two endpoints of the shortest path are branch points; ② there is at least one intermediate point in the shortest path and both are unmarked points; ③ the length of the shortest path exceeds 5 cm.

[0044] Step S33: For unreachable points, restore their original resolution, then extract branchlet structures through cluster analysis and linearity screening, repeat path analysis to extract valid path segments, and extract missing endpoints.

[0045] Step S34: Perform connectivity analysis on the missing endpoints and the existing branch points to obtain a complete leaf separation result, such as Figure 6 and Figure 7 shown.

[0046] By introducing voxelization processing and global path analysis strategies, the potential missing areas of the branch point cloud can be effectively identified, and the endpoint information under the multi-branch structure can be restored through path completion, further improving the extraction ability of complex structures.

[0047] Compared to existing technologies, this proposed method utilizes an automated segmentation strategy that combines the geometric characteristics, topological structure, and spatial distribution patterns of point clouds to achieve high-precision separation of branches and leaves in complex tree structures. Based on initial branch points extracted by the LeWoS algorithm, this strategy integrates local spatial geometry and path analysis to construct discriminative point cloud categories without relying on artificial rules or prior scene knowledge. This method offers excellent versatility and robustness, resulting in high recognition efficiency and effectiveness, making it suitable for data processing requirements across diverse tree species and at varying scanning accuracies.

[0048] Although the present invention has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the invention. It will be apparent that various modifications and variations may be made to the present invention by those skilled in the art without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such modifications and variations as fall within the scope of the claims of the present invention and their equivalents.

Claims

1. A method for separating branches and leaves from a single tree point cloud, characterized in that: include: Perform single-tree segmentation and denoising on the original point cloud data of the target tree to obtain a single-tree point cloud; The single tree point cloud is rotated at multiple angles, analyzed using the LeWoS algorithm, and subjected to linear constraint denoising to obtain an initial branch point cloud; The initial branch point cloud is divided into multiple branch structures by a clustering algorithm, and the endpoints of the branch structures are extracted by a PCA algorithm; Through the connectivity analysis algorithm and the judgment constraint conditions, the unmarked point clouds in the neighborhood of the determined branch point cloud path are marked as the endpoints of the missing branches, and the point clouds of the missing branches are topologically repaired; A path completion mechanism based on path analysis is introduced to connect the endpoints of the missing branches with multiple branches with the endpoints of the initial branch point cloud to obtain a complete branch point cloud.

2. The method for separating branches and leaves of a single tree point cloud according to claim 1, characterized in that: The single wood point cloud is rotated at multiple angles and analyzed using the LeWoS algorithm, specifically including: The single tree point cloud is rotated 90° around the X-axis, Y-axis, and Z-axis respectively. The LeWoS algorithm is executed after each rotation to extract branch points, and then the branch points extracted from different angles are merged.

3. The method for separating branches and leaves from a single tree point cloud according to claim 1, characterized in that: The linearity constraints specifically include: The linearity evaluation index is obtained by calculating the covariance matrix of the local neighborhood of the point cloud of the target tree and performing eigenvalue decomposition. The linearity calculation formula is: ； Where: λ1 and λ2 are the maximum and second largest eigenvalues ​​of the local neighborhood covariance matrix respectively; when the linearity value is greater than the preset threshold of 0.7, it is determined to be a branch point.

4. The method for separating branches and leaves of a single tree point cloud according to claim 1, characterized in that: The clustering algorithm is used to divide the initial branch point cloud into multiple branch structures, including using the following formula: ； Where: Meanshift h ( p ) represents the mean shift vector of the current point; P Indicates the three-dimensional coordinates of the current point; P i Represents the coordinates of the neighborhood points; h The bandwidth parameter to control the clustering scale; G (·) is the distance-based Gaussian kernel function.

5. The method for separating branches and leaves of a single tree point cloud according to claim 1, characterized in that: The endpoints of the branch structure are extracted by the PCA algorithm, specifically including: The PCA algorithm is used to perform principal component analysis on the point cloud cluster, extract the main direction axis, search for boundary points along the positive and negative directions of the main direction axis, and take the two points farthest from the geometric center as endpoints.

6. The method for separating branches and leaves of a single tree point cloud according to claim 1, characterized in that: The determination constraints specifically include: Point existence constraint: scan point clouds must exist between potential endpoint pairs; Path constraint: Dijkstra algorithm is used to calculate the shortest path between potential endpoint pairs. The shortest path between endpoint pairs cannot pass through existing branch point cloud clusters. Smoothness constraint: The path smoothness is evaluated by calculating the angle between the vectors of three adjacent points on the shortest path, and paths with multiple angles less than 150° are excluded.

7. The method for separating branches and leaves from a single tree point cloud according to claim 1, characterized in that: The path completion mechanism based on path analysis specifically includes: The branch point cloud and unclassified point cloud were voxelized at a resolution of 2 cm. The shortest path was found from each voxel point, with the tree base as the root node. Each voxel point was classified into two categories: reachable points and unreachable points, depending on whether there was a shortest path from a certain point to the tree base. For the shortest path to a reachable point, if the missing condition is met, the path is determined to be a potential missing branch, and its endpoint is a missing endpoint; For unreachable points, their original resolution is restored, and then the branch structure is extracted through cluster analysis and linearity screening. Valid path segments are extracted through repeated path analysis, and missing endpoints are extracted. Connectivity analysis is performed on the missing endpoints and the existing branch points to obtain the complete leaf separation results.

8. The method for separating branches and leaves of a single tree point cloud according to claim 7, characterized in that: The missing conditions include: The two endpoints of the shortest path are branch points; There is at least one intermediate point on the shortest path, and all of them are unmarked points; The length of the shortest path exceeds 5 cm.

Images