Sample plot individual tree parameter extraction method based on three-dimensional modeling, medium and equipment

Abstract

The invention relates to the technical field of forest identification, in particular to a sample plot individual tree parameter extraction method based on three-dimensional modeling, a medium and equipment, and the method comprises the following steps: obtaining radar point cloud data of a forest in a sample plot; obtaining complete sample point cloud data; obtaining seed points and diameter at breast height of the single tree in the sample plot, obtaining structural points of the sample plot, combining the structural points and the seed points with the diameter at breast height to establish a comparative shortest path, obtaining the point of the single tree structure and the shortest path from the point of the single tree structure to the seed point of the corresponding single tree according to the comparative shortest path, and counting the times of passing through each structural point; according to the shortest path passing times of the structure points of each single tree, obtaining a branch skeleton of the single tree in a path backtracking and frequency statistics mode; performing three-dimensional modeling on the branch skeleton, and then extracting parameters. Different from a traditional manual actual measurement mode, the method can replace the work of manually measuring the individual tree parameters of the sample plot, greatly reduces the field workload, and saves the cost.

Description

Technical Field

[0001] The invention relates to the technical field of forest tree identification, and in particular to a method, medium and equipment for extracting parameters of a single tree in a sample plot based on three-dimensional modeling. Background Art

[0002] Forest ecosystems are the main body of terrestrial ecosystems and an important part of the biosphere. They cover about 30% of the earth's land surface and can store more carbon than other terrestrial systems. They play an important role in regulating the global carbon balance and maintaining the global climate. Since the beginning of the 21st century, the reduction in global forest coverage has had a great impact on the global climate. Problems such as global warming and frequent extreme weather need to be improved. Therefore, scientific and effective management of forests is a way to solve these problems. Forest surveys can provide an important basis for the formulation of forest management and operation plans, and can help grasp the dynamic structure of forests and information such as the quantity and quality of forest resources in forest areas.

[0003] The content of forest survey is mainly to obtain forest spatial structure parameters, including canopy height, leaf area index, breast diameter, tree height, stock volume, biomass, etc. The former canopy height, leaf area index, breast diameter and tree height can be obtained more accurately through relevant instruments, but it is more difficult to obtain stock volume and biomass. Traditionally, the accurate acquisition of stock volume and biomass requires the destruction of plants. Although this method can obtain high-precision stock volume or biomass, it causes a waste of resources and is not suitable for long-term observation. Therefore, in the prior art, the estimation of stock volume and biomass depends on the forest structure parameters obtained by forest survey, and the forest stock volume or biomass of a certain area is estimated by combining parameters such as breast diameter and tree height with relevant models of specific regions and tree species. Although the artificial field forest survey based on sample plots can obtain relatively accurate forest structure parameters, it requires a lot of manpower, material resources and time.

[0004] Therefore, efficient and rapid acquisition of forest structure parameters is the fundamental driving force for the iterative development of forest survey methods and observation technologies. Summary of the invention

[0005] A first aspect of the present invention provides a method for extracting parameters of a single tree in a sample plot based on three-dimensional modeling, comprising the following steps:

[0006] Determine the location of the sample plot and obtain the LiDAR point cloud data of the trees in the sample plot;

[0007] Preprocess the acquired data to obtain complete sample point cloud data;

[0008] Obtain the seed points and DBH of individual trees in the sample plot, simplify the cloud data of the sample plot to obtain the structural points of the sample plot, combine the structural points with the seed points and the DBH to establish the shortest path, obtain the single-tree structural points and the shortest path from the single-tree structural points to the corresponding single-tree seed points based on the shortest path, and count the number of times each structural point is passed;

[0009] The branch skeleton of each tree is obtained by path backtracking and frequency statistics according to the number of times the shortest path passes through the structural point of each tree.

[0010] The three-dimensional model of a single tree in the sample plot is obtained by fitting the branch skeleton, and then the relevant parameters are extracted from the model.

[0011] Preferably, the radar point cloud data of the trees in the sampling site includes point cloud data including the sample site obtained by the drone according to the route map and point cloud data under the forest of the sample site obtained by the mobile laser radar equipment.

[0012] First, the laser radar point cloud data of the trees in the sample plot is obtained through the UAV laser radar equipment and the ground mobile laser radar equipment, and then the usable laser point cloud data is output through the relevant software.

[0013] Preferably, preprocessing the acquired data includes resampling, denoising, extracting ground points, obtaining a digital elevation model, and using the digital elevation model to normalize and align the point cloud.

[0014] The data is preprocessed, including point cloud denoising, point cloud resampling, and ground point extraction and classification, to obtain a digital elevation model (DEM). The point cloud is normalized using DEM, and then the drone laser point cloud data is aligned with the mobile device point cloud data to obtain complete sample point cloud data.

[0015] Preferably, the mean shift algorithm is used to simplify the sample point cloud data to obtain the structural points of the sample site. First, a KDTree index structure is established for the single tree point cloud data to improve the efficiency of searching for neighboring points; then the point with the minimum Z component (i.e., the point at the bottom of the trunk) is used as the starting center point; other points near this point are searched by radius search, and after searching the points of the specified radius, the mean of the coordinates of all points is calculated. The mean point is the structural point of the cluster; the distance to the previous structural point is determined. If the distance is less than a certain threshold, the iteration is stopped, otherwise the next structural point is calculated. In the actual iteration process, the maximum number of iterations and the minimum number of neighboring points searched by the neighboring points will be set. When the maximum number of iterations is exceeded and the minimum number of neighboring points is less than the minimum number of neighboring points, the iteration stops.

[0016] In the process of three-dimensional modeling of the sample point cloud, an innovative mean shift operation was performed on the pre-processed sample point cloud data to obtain the simplified sample site structural point cloud. This method can greatly simplify the calculation. By calculating the shortest path of the structural point cloud and counting the number of times each structural point passes, the skeleton of single tree branches can be directly extracted from the sample point cloud, and then the sample site branch skeleton can be cylindrically fitted to complete the three-dimensional modeling of single tree branches.

[0017] Preferably, the specific method for obtaining the seed point and breast diameter of a single tree in the sample plot is: first, intercept the trunk point cloud of 1.2-1.4m on the Z coordinate axis of the sampling point cloud, cluster the trunk point cloud of each single tree, calculate its center of mass point, use the center of mass point as the seed point of each single tree, and then vertically project the trunk point cloud of each single tree 1.2-1.4m to the XY plane, and use the least squares circle fitting algorithm to fit the diameter of the circle to be the breast diameter of the single tree.

[0018] Preferably, the specific calculation method of the shortest path from a single tree structure point and a single tree structure point to a corresponding single tree seed point is: calculate the shortest path from each seed point to other structure points, and convert the distance from the structure point to the trunk seed point according to the relationship between tree branching and breast diameter in metabolic ecology theory. The conversion relationship is as follows:

[0019]

[0020] in represents the distance from any point m to the starting seed point i after transformation, D m→i Indicates the distance from the point m before the transformation to the starting seed point i, DBH 2 / 3 It is a transformation scale related to the diameter at breast height. After the conversion is completed, the structural point can be divided into the corresponding single tree according to the minimum value of the shortest path from the point to the seed point, thereby obtaining the single tree structural point.

[0021] Preferably, a three-dimensional model of a single tree in the sample plot is obtained by performing cylindrical fitting on the branch skeleton. During the model fitting process, the volume of the fitted cylinder is statistically fitted to obtain the volume of the branches of the single tree. The volumes of the branches of the single trees in the sample plot are accumulated, and the accumulated volume of the main branches is used as the accumulation of the single tree, and the volume of all the branches of the single tree is used as the biomass, thereby directly obtaining the accumulation and biomass of the sample plot.

[0022] The second aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for extracting parameters of individual trees in a sample plot based on three-dimensional modeling when executing the program.

[0023] A third aspect of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for extracting single tree parameters from a sample plot based on three-dimensional modeling as described in any one of claims 1 to 7.

[0024] A fourth aspect of the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method for extracting parameters of individual trees in a sample plot based on three-dimensional modeling.

[0025] The present invention has the following effects:

[0026] The present invention realizes a new method for obtaining parameters of individual trees in a sample plot based on a three-dimensional model. Different from the traditional manual measurement method, the method of the present invention can replace the work of manually measuring the parameters of individual trees in a sample plot, greatly reducing the workload of field work and saving costs.

[0027] The method of the present invention firstly requires three-dimensional modeling of the sample point cloud data, and extracting relevant parameters through the three-dimensional model of the single tree. Therefore, it is extremely important to perform three-dimensional modeling of the single tree. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Schematic diagram of the process of the present invention;

[0029] Figure 2 This is a sample point cloud data map after preprocessing in Example 2 of the present invention;

[0030] Figure 3 This is a schematic diagram of a single timber structure point in a sample plot in Example 2 of the present invention;

[0031] Figure 4 This is a schematic diagram of a single tree trunk skeleton in Example 2 of the present invention;

[0032] Figure 5 This is a schematic diagram of the single tree modeling results in Example 2 of the present invention. DETAILED DESCRIPTION

[0033] For a better understanding of the present invention, the following examples are provided to further illustrate the present invention, but the present invention is not limited to the following examples.

[0034] like Figure 1 As shown, a method for extracting single tree parameters from a sample plot based on three-dimensional modeling includes the following steps:

[0035] S1. Determine the location of the sample plot and obtain radar point cloud data of the trees in the sample plot;

[0036] First, determine the location of the sample plot, draw a route map including the sample plot on the map, and the drone obtains point cloud data including the sample plot based on the route map; then use a mobile lidar device such as a handheld lidar device to obtain the understory point cloud data of the sample plot, and finally export the two types of data to obtain the drone lidar point cloud data and the mobile device lidar point cloud data.

[0037] Since the scanning range of mobile LiDAR and UAV LiDAR equipment is much larger than the sample plot range when scanning the target, the point cloud data containing the sample plot is first cropped out using point cloud processing software before proceeding to the next step.

[0038] S2, preprocessing the acquired data to obtain complete sample point cloud data;

[0039] S2.1. Point cloud resampling: The original point cloud data volume is large, and there are duplicate point clouds and noisy point clouds. Therefore, these problems in the original point cloud will have an adverse effect on the subsequent processing of the point cloud, and resampling can remove duplicate point clouds and some noise to a certain extent. In addition, point cloud resampling can compress the massive original point cloud data, so that the resampled data can reduce the data volume while maintaining the key features of the original point cloud.

[0040] S2.2, Point cloud denoising: In the process of point cloud data collection, due to internal reasons of the instrument or external unstable factors (such as weather conditions, human operation), some discrete points with no practical significance will be generated, and these points are not helpful for the subsequent point cloud processing, and may even affect the further processing of point cloud data. Therefore, it is necessary to remove these discrete points. There are many algorithms for removing noise, such as statistical outlier removal. This method is based on the principle of statistics. By calculating the statistical characteristics of each point and its neighboring points, such as the mean and standard deviation, points that deviate from the normal range are regarded as noise points. Distance-based filtering, this method is mainly used to remove outliers, that is, by calculating the distance between points, if the points are too far away from other points, then these points are likely to be noise. This paper first uses statistical filtering to filter the sample data, and then uses the distance-based filtering method to remove some outliers in the point cloud. For some noise that is difficult to remove by some algorithms, the relevant noise points are manually removed.

[0041] S2.3, ground point classification: In the processing of forestry point cloud data, the correct identification of ground points is the basis for subsequent point cloud processing. Therefore, algorithms for ground point classification are constantly emerging. Ground point classification is mainly divided into the following categories: slope-based filtering algorithm, block-minimum filtering algorithm, surface-based filtering algorithm, and clustering / segmentation algorithm. Among them, the more famous ones are the filtering algorithm based on the iterative encryption TIN model and the Cloth Simulation Filter (CSF). Experiments have shown that the filtering algorithm based on the iterative encryption TIN model has a better effect on the classification of ground points of UAV point cloud data, but is not ideal for handheld laser device point cloud data. The CSF algorithm has a better effect on mobile laser radar point cloud data with dense ground points. Therefore, this solution uses the filtering algorithm based on the iterative encryption TIN model for the acquisition of ground points of UAV point cloud data, and the CSF algorithm is used for the acquisition of ground points of mobile point cloud data.

[0042] S2.4. Point cloud data normalization: Due to the influence of terrain undulation, the starting elevations of different objects will be inconsistent, which is not conducive to the subsequent single tree segmentation and DBH extraction. Therefore, it is necessary to use ground point data to perform elevation normalization on the entire sample point cloud data, so that the starting elevations of the targets in the sample point cloud are consistent and the influence of terrain undulation is eliminated.

[0043] S2.5, Point cloud registration: Mobile LiDAR point cloud data can clearly reflect the conditions under the forest, but it is difficult to obtain the real data of the forest canopy at a certain height; UAV LiDAR point cloud data can clearly describe the forest canopy, but it is difficult to obtain the real situation under the forest. In summary, a single point cloud data is difficult to fully express the real situation of the sample site, so the registration of mobile point cloud data and UAV point cloud data is expected to solve the above problems.

[0044] Figure 2 Shown is the sample point cloud data after preprocessing.

[0045] S3. Obtain the seed points and DBH of individual trees in the plot, simplify the cloud data of the plot to obtain the structural points of the plot, combine the structural points with the seed points and the DBH to establish the shortest path, obtain the single-tree structural points and the shortest path from the single-tree structural points to the corresponding single-tree seed points based on the shortest path, and count the number of times each structural point is passed. The specific method is as follows:

[0046] First, the trunk point cloud of the sampling point cloud with the Z coordinate axis of 1.2-1.4m was intercepted, and then the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm was used to cluster the trunk point cloud of each single tree, and its centroid point was calculated. The centroid point was used as the seed point of each single tree, and then the trunk point cloud of each single tree with the length of 1.2-1.4m was vertically projected to the XY plane, and the diameter of the circle was fitted as the DBH of the single tree using the least squares circle fitting algorithm. Secondly, the mean shift algorithm was used to obtain the structural points of the entire sampling point cloud, and the actual point cloud corresponding to the structural points was recorded, and the structural points were used to replace the entire sampling point cloud to speed up the calculation. Then the Dijkstra algorithm was used to calculate the shortest path from each seed point to other structural points, and the number of times the shortest path of each structural point passed was counted. Finally, according to the relationship between tree branching and DBH in the metabolic ecology theory, the distance from the structural point to the trunk seed point was converted according to the scale. The conversion relationship is as follows:

[0047]

[0048] in represents the distance from any point m to the starting seed point i after transformation, D m→i Indicates the distance from the point m before the transformation to the starting seed point i, DBH 2 / 3 It is a transformation scale related to the diameter at breast height. After the conversion is completed, the structural point can be divided into the corresponding single tree according to the minimum value of the shortest path from the point to the seed point, thereby obtaining the single tree structural point.

[0049] Figure 3 This is a schematic diagram of a single wooden structure point in the sample plot.

[0050] S4, obtaining the branch skeleton of each tree by path backtracking and frequency statistics according to the number of times the shortest path of each tree's structural point passes through;

[0051] After the single-tree structure points are obtained, the points passed by the shortest path from each single-tree starting seed point to the single-tree structure point and the number of times these points are passed will be obtained, and then the skeleton of the branch will be obtained according to the following path backtracking and frequency statistics steps.

[0052] First, for each shortest path, obtain the other points on the path except the starting point and the end point, which are called seed points. In nature, the nutrients of trees are always transported from the roots along the trunk to the leaves. Therefore, non-seed points include a large number of leaf points. Since leaf points are generally located at the end of branches, the distance from the starting point to the branch point will be smaller than the distance from the starting point to the leaf point. After the seed point is confirmed, the radius neighbor search is used to find the neighbor points near the seed point, and the shortest path from the starting point to the seed point and the shortest path from the starting point to the seed point and the shortest path from the starting point to the neighbor point are determined. If the distance of the former is greater than or equal to the latter, the neighbor point is added to the non-leaf point set, otherwise it is added to the leaf point set. Based on this condition, the leaf points can be roughly extracted, and the conditional expression is as follows:

[0053]

[0054] In the above formula, p i Represents the seed point Neighboring points, dis(graph, begin, p i ) represents the shortest path distance from the starting point to the neighboring point, Represents the shortest path distance from the starting point to the seed point.

[0055] Secondly, in the shortest path from the starting point to the leaf point, the branch points often have a higher access frequency, which is consistent with the phenomenon that tree nutrients are transported to the leaves through the branches in nature. Therefore, in view of the high access rate of branch points, when calculating the shortest path, the number of points passed is recorded, and the branch points are extracted by setting a threshold. In order to facilitate the setting of the threshold, the access frequency is converted into a logarithmic form, and the expression is as follows:

[0056]

[0057] In the above formula, p i represents a point in the point cloud of a single timber structure, freq(p i ) represents point p i The access frequency of point cloud is σ, σ is the set threshold value, the value range is 0-1, log(max(freq(p))) represents the logarithm of the highest access frequency in the point cloud, and n represents the number of point clouds.

[0058] After the above two steps, the branch skeleton structure points of each single tree can be obtained, and then the actual point cloud corresponding to the structure points is restored to obtain a complete branch skeleton.

[0059] like Figure 4 Shown is a schematic diagram of the branch skeleton of a single tree.

[0060] S5. Fit the branch skeleton to obtain a three-dimensional model of a single tree in the sample plot, and then extract relevant parameters from the model.

[0061] The obtained branch skeleton is extracted and the point cloud is fitted with a cylinder segment by segment, and finally a complete single tree model is obtained. First, starting from the bottom point cloud, the Levenberg-MarQuart algorithm is used to perform nonlinear least squares cylinder fitting. In order to improve the quality of the fitting, the nonlinear least squares process is iterated, and weights are introduced for each point in the second iteration. The purpose of introducing weights is to exert a greater influence on points closer to the cylinder. The weight expression is as follows:

[0062]

[0063] In the above formula, dis(p i ) represents the Euclidean distance from the point to the cylinder, dis max Indicates the maximum distance from all points to the cylinder.

[0064] The three-dimensional models of all single trees in the sample plot are obtained by following the steps in sequence. In the process of model fitting, the volume of the fitted cylinder is statistically analyzed to obtain the volume of the single tree trunk, and then the parameters such as the accumulation and biomass of the sample plot can be obtained.

[0065] Figure 5 Shown is a schematic diagram of the single tree modeling results.

[0066] Example 2

[0067] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for extracting single tree parameters from a sample plot based on three-dimensional modeling is implemented.

[0068] Example 3

[0069] A non-transitory computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for extracting single tree parameters from a sample plot based on three-dimensional modeling as described in any one of claims 1 to 7 is implemented.

[0070] Example 4

[0071] A computer program product comprises a computer program, wherein when the computer program is executed by a processor, the method for extracting parameters of individual trees in a sample plot based on three-dimensional modeling is implemented.

[0072] The above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and changes can be made without departing from the principle of the present invention, and these improvements and changes are also regarded as the protection scope of the present invention.

Claims

1. A method for extracting single tree parameters from a sample plot based on three-dimensional modeling, characterized in that: The following steps are involved: Determine the location of the sample plot and obtain the LiDAR point cloud data of the trees in the sample plot; Preprocess the acquired data to obtain complete sample point cloud data; Obtain the seed points and DBH of individual trees in the sample plot, simplify the cloud data of the sample plot to obtain the structural points of the sample plot, combine the structural points with the seed points and the DBH to establish the shortest path, obtain the single-tree structural points and the shortest path from the single-tree structural points to the corresponding single-tree seed points based on the shortest path, and count the number of times each structural point is passed; The branch skeleton of each tree is obtained by path backtracking and frequency statistics according to the number of times the shortest path passes through the structural point of each tree. The three-dimensional model of a single tree in the sample plot is obtained by fitting the branch skeleton, and then the relevant parameters are extracted from the model.

2. The method for extracting single tree parameters from a sample plot based on three-dimensional modeling according to claim 1, characterized in that: The laser radar point cloud data of the trees in the sample plot is obtained by the drone according to the route map, and the point cloud data of the forest under the sample plot is obtained by the mobile laser radar equipment.

3. The method for extracting single tree parameters from a sample plot based on three-dimensional modeling according to claim 1, characterized in that: The obtained data is preprocessed including resampling, denoising, ground point extraction, obtaining a digital elevation model and using the digital elevation model to normalize and align the point cloud.

4. The method for extracting single tree parameters from a sample plot based on three-dimensional modeling according to claim 1, characterized in that: The mean shift algorithm is used to simplify the sample point cloud data to obtain the structural points of the sample site.

5. The method for extracting single tree parameters from a sample plot based on three-dimensional modeling according to claim 1, characterized in that: The specific method for obtaining the seed point and breast diameter of a single tree in the sample plot is as follows: first, intercept the trunk point cloud of 1.2-1.4m on the Z coordinate axis of the sampling point cloud, cluster the trunk point cloud of each single tree, calculate its centroid point, and use the centroid point as the seed point of each single tree, then vertically project the trunk point cloud of each single tree 1.2-1.4m to the XY plane, and use the least squares circle fitting algorithm to fit the diameter of the circle as the breast diameter of the single tree.

6. The method for extracting single tree parameters from a sample plot based on three-dimensional modeling according to claim 5, characterized in that: The specific calculation method of the shortest path from a single tree structure point to the corresponding single tree seed point is as follows: calculate the shortest path from each seed point to other structure points, and convert the distance from the structure point to the trunk seed point according to the relationship between tree branching and breast diameter in metabolic ecology theory. The conversion relationship is as follows: in represents the distance from any point m to the starting seed point i after transformation, D m→i Indicates the distance from the point m before the transformation to the starting seed point i, DBH 2 / 3 It is a transformation scale related to the diameter at breast height. After the conversion is completed, the structural point can be divided into the corresponding single tree according to the minimum value of the shortest path from the point to the seed point, thereby obtaining the single tree structural point.

7. The method for extracting single tree parameters from a sample plot based on three-dimensional modeling according to claim 1, characterized in that: The three-dimensional model of a single tree in the sample plot was obtained by fitting the branch skeleton with a cylinder. During the model fitting process, the volume of the fitted cylinder was statistically calculated to obtain the volume of the single tree branch and the accumulation and biomass parameters of the sample plot.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for extracting single tree parameters from a sample plot based on three-dimensional modeling as described in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for extracting parameters of individual trees in a sample plot based on three-dimensional modeling as described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for extracting parameters of individual trees in a sample plot based on three-dimensional modeling as described in any one of claims 1 to 7 is implemented.

Images