Single tree biomass estimation method based on forest quantitative structure model

Abstract

The invention discloses an individual tree biomass estimation method based on a forest quantitative structure model, and belongs to the field of forest resource investigation, forest improved variety breeding and carbon sink metering, and the method comprises the steps: S1, collecting foundation LiDAR data through a foundation laser radar device; s2, carrying out preprocessing on the foundation LiDAR data; s3, single tree point cloud segmentation: clustering the point cloud, and extracting a three-dimensional structure point cloud of a single tree; s4, constructing a forest quantitative structure model by extracting an initial skeleton of a forest and approximately fitting geometric shapes of a trunk and branches by using a cylinder; s5, forest biomass calculation; s6, performing precision verification on the result; according to the method, detailed tree trunk, branch and leaf information can be obtained through foundation LiDAR data, a single-tree three-dimensional space structure model is constructed by comprehensively and deeply analyzing the three-dimensional space distribution rule of single-tree point cloud, and the single-tree biomass estimation capability and precision are enhanced.

Description

Technical Field

[0001] The present invention belongs to the fields of forest resource investigation, tree breeding and carbon sink measurement, and particularly relates to a single tree biomass estimation method based on a forest quantitative structure model. Background Art

[0002] Individual trees are the fundamental units of forests, and accurate estimation of their biomass is a crucial component of forest resource and carbon sink surveys. It is crucial for forest resource surveys, tree breeding, and carbon sink measurement. Furthermore, accurately understanding the biomass distribution patterns of individual trees and further understanding competition within forests and their interactions with the environment is crucial for sustainable forest management and ecosystem solid sequestration. Conventional forest resource surveys fail to consider detailed individual tree biomass distribution. Furthermore, the extraction of survey parameters relies primarily on field surveys and empirical models, which are inefficient and often lack high accuracy, making them difficult to implement across large areas. Ground-based LiDAR data can capture detailed information about tree trunks, branches, and foliage, including detailed information on biomass within the canopy. Furthermore, its extremely high point density facilitates accurate extraction of the three-dimensional spatial structure of tree trunks, branches, and foliage. By accurately fitting the three-dimensional spatial structure of tree trunks and branches, accurate estimates of individual tree biomass can be achieved.

[0003] In the prior art, “Terrestrial laser scanning for non-destructive estimates of liana stem biomass” was published in Volume 456 of Forest Ecology and Management. This is a study on tree biomass estimation based on ground-based LiDAR. The study manually extracted the stem and branch information from the ground-based LiDAR point cloud and combined it with the cylinder fitting method to estimate the tree biomass. The results showed that the accuracy was higher than that of the traditional allometric equation method. In addition, in the prior art, “Development of estimation models for individual tree aboveground biomass based on TLS-derived parameters” was published in Volume 14 of Forests. This study extracted information related to individual tree height and crown width through ground-based LiDAR point clouds, and combined it with a multivariate stepwise regression algorithm to estimate the aboveground biomass of trees.

[0004] However, existing methods generally suffer from indirect extraction of trunk morphological parameters, diverse estimation models, and high uncertainty, making it difficult to achieve universal and accurate forest biomass estimation. Furthermore, there are no biomass estimation methods that comprehensively analyze the three-dimensional spatial distribution of tree branches and comprehensively consider the spatial connectivity between the trunk and branches of individual trees.

[0005] Moreover, the traditional method is based on the statistical characteristics of LiDAR and combines simple regression relationships to build models. It does not use ground-based LiDAR technology that can record the three-dimensional structural information of tree branches in detail, and is unable to accurately depict the three-dimensional structure of tree branches. Summary of the Invention

[0006] In response to the problems mentioned in the background technology, the present invention proposes a method for estimating the biomass of individual trees based on a quantitative structural model of forest trees. By comprehensively and in-depth analyzing the three-dimensional spatial distribution patterns of individual tree point clouds, a three-dimensional spatial structural model of individual trees is constructed, thereby enhancing the ability and accuracy of individual tree biomass estimation.

[0007] Technical solution: In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] A method for estimating individual tree biomass based on a quantitative forest structure model comprises the following steps:

[0009] S1: Collect ground-based LiDAR data through ground-based LiDAR equipment;

[0010] S2: Preprocessing of ground-based LiDAR data;

[0011] S3: Single tree point cloud segmentation, clustering the point cloud to extract the 3D structure point cloud of a single tree;

[0012] S4: A quantitative tree structure model is constructed by extracting the initial skeleton of the tree and using cylinders to approximate the geometric shapes of the trunks and branches;

[0013] S5: Calculation of forest biomass;

[0014] S6: Verify the accuracy of the results.

[0015] Preferably, in S2, the specific process of preprocessing the ground-based LiDAR data is as follows:

[0016] S21: Merge and crop ground-based LiDAR data;

[0017] S22: Denoise the ground-based LiDAR point cloud;

[0018] S23: Perform ground extraction and height normalization on the ground-based LiDAR point cloud.

[0019] Preferably, in S22, a denoising algorithm based on neighborhood distance is used to identify noise in the point cloud. For each point to be measured, a specified number of neighboring points are searched for. The average distance D from the point to the neighboring points is calculated. The median Mean D and standard deviation S of the distance average are calculated. The maximum distance Max D is calculated according to the set standard deviation multiple K. If the distance average D of a point is greater than the median Max D of the distance average, the point is considered to be a noise point. The specific calculation formula is:

[0020] Max D = Mean D + K × S.

[0021] Preferably, in S23, the specific process of performing ground extraction and height normalization processing on the ground-based LiDAR point cloud is as follows:

[0022] For the denoised sample point cloud data, the progressive encryption triangulation filtering algorithm is used to classify the ground points, and the nearest neighbor method is used to construct the DEM based on the classified ground points, and then the point cloud height is normalized. The specific calculation formula is:

[0023]

[0024] Among them, Z i represents the elevation value at point i, represents the DEM pixel value corresponding to point i, and Z represents the normalized height.

[0025] As a preferred method, in S3, the specific process of segmenting the single tree point cloud, clustering the point cloud, and extracting the three-dimensional structure point cloud of a single tree is as follows:

[0026] S31: A density-based noisy spatial clustering algorithm is used to identify the trunk point cloud of each tree. A cylinder is fitted to the trunk point cloud slices of each tree to obtain the DBH of each tree.

[0027] S32: Extract the centroid of each point cloud cluster as the seed point of the tree, and segment the tree crown from bottom to top using the comparative shortest path algorithm based on the seed point of each tree.

[0028] As a preference, the specific process of S4 is:

[0029] S41: Use the minimum spanning tree algorithm to extract the initial skeleton of each tree from the point cloud and perform pruning;

[0030] S42: A series of cylinders are then fitted to approximate the geometry of the trunk and branches.

[0031] Preferably, the specific content of S42 is:

[0032] After selecting the tree trunk points to fit the cylinder, extract the non-overlapping parts from the top, middle, and bottom of the selected point cloud; calculate the cylinder radius of the intersection of the three point clouds and the tree bottom, and finally take the average of the three radii as the initial cylinder radius;

[0033] Determine the intersection of the three cylinder segments extending downward along the central axis with the ground, and take the average of the three intersection points as the position of the tree base. Take the point with the largest normalized height value in each part of the point cloud as the top center, and the point with the smallest normalized height value as the bottom center. Use half the length or width of the minimum bounding box of the part of the point cloud as the radius, and use it as the initial rough cylinder in the tree cylinder fitting process.

[0034] After determining the top center, bottom center, and radius of the initial cylinder, the cylinder is optimized using the nonlinear least squares method. The optimized cylinder becomes the fitting cylinder of the final tree reconstruction model.

[0035] Preferably, in S5, the specific content of forest biomass calculation is:

[0036] According to the quantitative structural model, the volume of the cylinders contained in the trunks and branches of the trees was counted respectively to calculate the volume of the trunks and side branches of the trees. The biomass of individual trees was calculated with the help of the basic wood density of the trees. Specifically,

[0037] Q=V QSM ×D

[0038] Among them, Q represents the biomass of a single tree, V QSM It represents the volume of the trunk or side branches, and D represents the wood density of the corresponding tree species.

[0039] Preferably, in S6, the accuracy of single tree segmentation is evaluated by comparing the single tree positioning record information with the single tree segmentation results; the correctly segmented trees TP, the incorrectly detected trees FP, and the undetected trees FN obtained after segmentation are counted; the tree detection rate r, the correct rate of tree segmentation p, and the overall accuracy F taking into account the misclassification and omission are calculated, and the calculation formula is as follows:

[0040] r=TP / (TP+FN)

[0041] p=TP / (TP+FP)

[0042] F = (r × p) / (r + p)

[0043] The coefficient of determination R 2 , root mean square error RMSE and relative root mean square error rRMSE are used to evaluate the effect and accuracy of single tree biomass estimation, specifically:

[0044]

[0045] Among them, xi represents the measured value of aboveground biomass of a single tree; represents the measured average aboveground biomass of individual trees; represents the estimated value of the aboveground biomass of a single tree; n represents the number of samples; i represents a single tree sample.

[0046] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0047] (1) The ground-based LiDAR data of the present invention can obtain detailed information about tree trunks, branches, and leaves, including detailed information related to biomass within the canopy. Furthermore, its ultra-high point density facilitates accurate extraction of the three-dimensional spatial structure of tree trunks, branches, and leaves. By meticulously fitting the three-dimensional spatial structure of tree trunks, it facilitates accurate estimation of the biomass of individual trees.

[0048] (2) The present invention constructs a three-dimensional spatial structure model of a single tree by comprehensively and deeply analyzing the three-dimensional spatial distribution pattern of the single tree point cloud. Since this method adopts a spatial quantitative structure model, the method enhances the ability and accuracy of estimating the biomass of a single tree. At the same time, the method can fundamentally enhance the ability to estimate the three-dimensional spatial structure of trees, thereby improving the accuracy of estimating the biomass of a single tree. The method takes into account the influence of mutual occlusion of tree canopies on the segmentation of single trees, and adopts a bottom-up single tree canopy point cloud segmentation method, clustering the point cloud according to the arrangement of the main trunk points of the trees, thereby improving the accuracy of single tree segmentation. The verification results show that the overall accuracy of estimating the biomass of a single tree at a scale by the present invention is improved by more than 10% compared with the use of other similar estimation methods to obtain the biomass of a single tree.

[0049] (3) The present invention performs multi-station matching and splicing, denoising, ground point identification and extraction, and DEM creation on the ground-based LiDAR point cloud, identifies single tree segmentation seed points based on the trunk point cloud, and adopts a bottom-up spatial clustering strategy and a comparative shortest path algorithm to segment and extract the single tree point cloud.

[0050] (4) This paper constructs a single tree point cloud supervoxel and uses adaptively sized cylinders to fit the trunk and branches, constructing the trunk and branch topology to reconstruct a three-dimensional quantitative structural model of the trunk and branches of the single tree. The spatial volume of the trunk and branches of the single tree is statistically calculated and combined with the basic density of the forest to estimate the biomass of the single tree. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a single tree segmentation result diagram of the present invention, wherein a is a single tree segmentation result diagram of a high stand density sample plot, b is a single tree segmentation diagram of a medium stand density sample plot, c is a single tree segmentation diagram of a low stand density sample plot, and d is a single tree segmentation diagram of an ultra-low stand density sample plot;

[0052] Figure 2The three-dimensional point cloud visualization of the sample trees of the present invention is shown in FIG1 , wherein (a) is a three-dimensional point cloud diagram of a single tree branch in a high stand density sample plot, (b) is a three-dimensional point cloud diagram of a single tree branch in a medium stand density sample plot, (c) is a three-dimensional point cloud diagram of a single tree branch in a low stand density sample plot, and (d) is a three-dimensional point cloud diagram of a single tree branch in an ultra-low stand density sample plot;

[0053] Figure 3 It is the result diagram of the single tree quantitative structure model of the present invention, wherein a1 is the quantitative structure model diagram of the trunk and the first, second and third order branches of the sample tree in the high stand density sample plot, a2 is the quantitative structure model diagram of the trunk and the first, second and third order branches of the sample tree in the medium stand density sample plot, a3 is the quantitative structure model diagram of the trunk and the first, second and third order branches of the sample tree in the low stand density sample plot, b1 is the quantitative structure model diagram of the trunk and the first and second order branches of the sample tree in the high stand density sample plot; b2 is the quantitative structure model diagram of the trunk and the first and second order branches of the sample tree in the medium stand density sample plot, b3 is the quantitative structure model diagram of the trunk and the first and second order branches of the sample tree in the low stand density sample plot The quantitative structural model diagram of the trunk and primary and secondary branches of the sample trees in the density sample plots, C1 is the quantitative structural model diagram of the trunk and primary branches of the sample trees in the high stand density sample plots, C2 is the quantitative structural model diagram of the trunk and primary branches of the sample trees in the medium stand density sample plots; C3 is the quantitative structural model diagram of the trunk and primary branches of the sample trees in the low stand density sample plots, D1 is the quantitative structural model diagram of the trunk of the sample trees in the high stand density sample plots, D2 is the quantitative structural model diagram of the trunk of the sample trees in the medium stand density sample plots, D3 is the quantitative structural model diagram of the trunk of the sample trees in the low stand density sample plots. DETAILED DESCRIPTION

[0054] The present invention will be further illustrated below with reference to specific examples. The examples are implemented based on the technical solutions of the present invention. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0055] The present embodiment provides a method for estimating the biomass of individual trees based on a quantitative structural model of forest trees. The plantation of Caragana chinensis is selected as the research object. A total station is used for ground survey to determine the coordinates of the four corner points of the sample plot, and the diameter at breast height, tree height and height under branches of each individual tree in the sample plot are measured. The coordinates of the corner points, center coordinates and position information of each individual tree in the sample plot are obtained by a real-time differential device FJD Trion RTK GNSS. The device guarantees a positioning accuracy of 3 cm within a 60° inclination angle. The diameter at breast height is measured in two opposite directions using a steel tape measure and the average is taken. The tree height and height under branches are measured using a Vertex V ultrasonic altimeter and rangefinder ( Swiss) used ultrasonic and laser modes to measure tree height. The two modes were measured twice and the average values ​​were taken. The DBH, tree height and height below the branches obtained from the sample survey and their standard deviations are shown in Table 1.

[0056] Table 1 Summary of sample information

[0057]

[0058] The specific steps include:

[0059] S1: Collect high-density ground-based LiDAR data through ground-based LiDAR equipment;

[0060] Data were acquired using a ground-based LiDAR system, a RIEGL-VZ 400i 3D laser scanner (RIEGL Laser Measurement Systems, Horn, Austria). The RIEGL-VZ 400i 3D laser scanner has a laser transmission frequency of 1200 kHz and achieves an accuracy of 5 mm. TLS point cloud data were acquired on clear, windless days with approximately 1 minute of scanning time per station. Scanning stations were arranged in a "#" pattern, adjusted based on the tree density, canopy density, and tree height of the sample plot.

[0061] S2: Preprocessing of ground-based LiDAR data;

[0062] S21: Merge and crop ground-based LiDAR data;

[0063] After data collection, the Riscan Pro software package was used to match and merge the point clouds from all scanned locations into a single point cloud. The point cloud was then cropped based on the coordinates of the four corner points of the plot to obtain the point cloud data within the target plot.

[0064] S22: Denoise the ground-based LiDAR point cloud;

[0065] A denoising algorithm based on neighborhood distance is used to identify noise in point clouds. The input of the algorithm is the number of neighborhood points (N) and the standard deviation multiple (K). For each point to be measured, it searches for the specified number of neighboring points. The average distance (D) from the point to the neighboring points is calculated. The median (Mean D) and standard deviation (S) of these average distances are calculated. The maximum distance (Max D) is calculated based on the set K. The calculation formula is shown in Equation (1). If the D of a point is greater than Max D, it is considered a noise point and is removed.

[0066] Max D = Mean D + K × S (1)

[0067] S23: perform ground extraction and height normalization on the ground-based LiDAR point cloud;

[0068] For the denoised sample point cloud data, the progressive encryption triangulation filtering algorithm is used to classify the ground points, and the digital elevation model (DEM) is constructed based on the classified ground points using the nearest neighbor method, and then the point cloud height is normalized, that is:

[0069]

[0070] Among them, Z i represents the elevation value at point i, It represents the DEM pixel value corresponding to point i. The normalized point cloud height value represents the height of the point from the ground. Z represents the normalized height.

[0071] S3: Single tree point cloud segmentation, clustering the point cloud to extract the 3D structure point cloud of a single tree;

[0072] S31: First, extract the tree trunk point cloud within the height range of 1.2 to 1.4 meters, and use the density-based noisy spatial clustering algorithm to identify the trunk point cloud of each tree. Fit the 0.2-meter-thick trunk point cloud slice of each tree through a cylinder to obtain the diameter at breast height of each tree.

[0073] S32: The centroid of each point cloud cluster is extracted as the seed point of the tree. Then, based on the seed point of each tree, the shortest path comparison algorithm is used to segment the tree crown from bottom to top. Before segmentation, the distance between tree branches is normalized:

[0074]

[0075] Among them, D v→trunk is the connection distance from point v to the trunk, is the length-normalized connection distance, and DBH is the tree diameter at breast height. Figure 1 shown.

[0076] The shortest path comparison in the present invention is based on the Dijkstra algorithm, and its mathematical model process is as follows:

[0077] d(n)=min(d(u)+w(u,n))

[0078] Where d(n) represents the shortest distance from node n to the starting point, d(u) represents the shortest distance from node u to the starting point, and w(u,n) represents the actual distance from node u to node n.

[0079] S4: A quantitative tree structure model is constructed by extracting the initial skeleton of the tree and using cylinders to approximate the geometric shapes of the trunks and branches;

[0080] S41: The minimum spanning tree algorithm (MST) is used to extract the point cloud of each tree (such as Figure 2) to extract the initial tree skeleton and perform pruning. The MST algorithm uses a greedy algorithm. The basic idea is to start from the lowest point in each tree point cloud and gradually grow a minimum spanning tree. At each step, the algorithm selects a minimum edge connecting the selected point and the unselected points and adds this edge and its points to the minimum spanning tree. This process continues until all points have been added to the minimum spanning tree.

[0081] S42: A series of cylinders are then fitted to approximate the geometry of the trunk and branches.

[0082] At this point, the tree is a set of generalized cylindrical surfaces. After closing the main branch tips, the tree transforms from a generalized cylindrical surface into a closed convex hull polyhedron. Due to the abrupt and nonlinear nature of the tree root point cloud, the trunk base point cloud may also contain a large number of noise points, which may ultimately lead to inaccurate radius of the entire tree cylinder model. To make the quantitative structural model of the tree more accurate, a semi-automated modeling method was used to optimize the modeling results.

[0083] After selecting relatively stable tree trunk points to fit the cylinder, non-overlapping parts are taken from the top, middle, and bottom of the selected point cloud. The cylinder radius of the intersection of the three point clouds and the tree bottom is calculated, and the average of the three radii is finally taken as the initial cylinder radius. Determine the intersection of the three cylinders with the ground after extending downward along the central axis, and take the average of the three intersections as the position of the tree bottom. Take the point with the largest Z value in each part of the point cloud as the top center, and the point with the smallest Z value as the bottom center. Use half of the length or width of the minimum bounding box of the part of the point cloud (take the maximum value) as the radius as the initial rough cylinder in the tree cylinder fitting process. After determining the initial cylinder (top center, bottom center, radius), the nonlinear least squares method is used to optimize the cylinder. The optimized cylinder is the fitting cylinder of the final tree reconstruction model.

[0084] S5: Calculation of forest biomass;

[0085] According to the quantitative structural model, the volume of the cylinders contained in the trunks and branches of the trees is counted to calculate the volume of the trunks and side branches. At the same time, the biomass of individual trees is calculated with the help of the basic wood density of the trees:

[0086] Q = V QSM × D (4)

[0087] Among them, Q is the biomass of a single tree, V QSM is the volume of the trunk or side branches, and D is the wood density of the corresponding tree species.

[0088] S6: Verify the accuracy of the results.

[0089] Finally, the accuracy of the results was verified. On the one hand, the accuracy of individual tree segmentation was evaluated by comparing the individual tree location record information with the individual tree segmentation results. The number of correctly segmented trees (TP), incorrectly detected trees (FP), and undetected trees (FN) obtained after segmentation was statistically analyzed. Based on this, the tree detection rate (r), the correct tree segmentation rate (p), and the overall accuracy (F) that comprehensively considers misclassification and omission were calculated using the following formula:

[0090] r = TP / (TP + FN) (5)

[0091] p = TP / (TP + FP) (6)

[0092] F = (r × p) / (r + p) (7)

[0093] At the same time, this method uses the coefficient of determination (R 2 ), root mean square error (RMSE) and relative root mean square error (rRMSE) were used to evaluate the effect and accuracy of single tree biomass estimation:

[0094]

[0095] Among them, x i is the measured value of aboveground biomass of a single tree; is the measured average aboveground biomass of individual trees; is the estimated value of the aboveground biomass of a single tree; n is the number of samples; i is a single tree sample.

[0096] The results of extracting single tree parameters by the method proposed in this invention are shown in Figure 3 The summary results are shown in Table 2. Figure 3 The trunks of the middle trees and the first, second and third level branches are clearly visible, intuitively displaying the three-dimensional spatial structure information of the segmented and extracted trunks and branches.

[0097] Table 2 Summary of individual tree parameter extraction based on the forest quantitative structure model

[0098]

[0099]

[0100] This method comprehensively considers the three-dimensional distribution of tree trunk point clouds and constructs a refined quantitative structural model of tree trunks, achieving high-precision estimation of trunk biomass at the individual tree scale. Case studies demonstrate that this method performs well in forest stands of varying densities, achieving approximately 10% higher accuracy than traditional allometric equation methods.

[0101] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for estimating individual tree biomass based on a quantitative forest structure model, characterized by: The following steps are involved: S1: Collect ground-based LiDAR data through ground-based LiDAR equipment; S2: Preprocessing of ground-based LiDAR data; S3: Single tree point cloud segmentation, clustering the point cloud to extract the 3D structure point cloud of a single tree; S4: A quantitative tree structure model is constructed by extracting the initial skeleton of the tree and using cylinders to approximate the geometric shapes of the trunks and branches; S5: Calculation of forest biomass; S6: Verify the accuracy of the results.

2. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 1, characterized in that: In S2, the specific process of preprocessing ground-based LiDAR data is as follows: S21: Merge and crop ground-based LiDAR data; S22: Denoise the ground-based LiDAR point cloud; S23: Perform ground extraction and height normalization on the ground-based LiDAR point cloud.

3. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 2, wherein: In S22, a denoising algorithm based on neighborhood distance is used to identify noise in the point cloud. For each point to be measured, a specified number of neighboring points are searched for. The average distance D from the point to the neighboring points is calculated. The median value Mean D and the standard deviation S of the distance average are calculated. The maximum distance Max D is calculated based on the set standard deviation multiple K. If the distance average D of a point is greater than the median value Max D of the distance average, the point is considered to be a noise point. The specific calculation formula is: Max D = Mean D + K × S.

4. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 2, wherein: In S23, the specific process of ground extraction and height normalization of the ground-based LiDAR point cloud is as follows: For the denoised sample point cloud data, the progressive encryption triangulation filtering algorithm is used to classify the ground points, and the nearest neighbor method is used to construct the DEM based on the classified ground points, and then the point cloud height is normalized. The specific calculation formula is: Among them, Z i represents the elevation value at point i, represents the DEM pixel value corresponding to point i, and Z represents the normalized height.

5. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 1, wherein: In S3, the specific process of segmenting the single tree point cloud, clustering the point cloud, and extracting the 3D structure point cloud of a single tree is as follows: S31: A density-based noisy spatial clustering algorithm is used to identify the trunk point cloud of each tree. A cylinder is fitted to the trunk point cloud slices of each tree to obtain the DBH of each tree. S32: Extract the centroid of each point cloud cluster as the seed point of the tree, and segment the tree crown from bottom to top using the comparative shortest path algorithm based on the seed point of each tree.

6. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 1, wherein: The specific process of S4 is: S41: Use the minimum spanning tree algorithm to extract the initial skeleton of each tree from the point cloud and perform pruning; S42: A series of cylinders are then fitted to approximate the geometry of the trunk and branches.

7. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 6, characterized in that: The specific contents of S42 are: After selecting the tree trunk points to fit the cylinder, extract the non-overlapping parts from the top, middle, and bottom of the selected point cloud; calculate the cylinder radius of the intersection of the three point clouds and the tree bottom, and finally take the average of the three radii as the initial cylinder radius; Determine the intersection of the three cylinder segments extending downward along the central axis with the ground, and take the average of the three intersection points as the position of the tree base. Take the point with the largest normalized height value in each part of the point cloud as the top center, and the point with the smallest normalized height value as the bottom center. Use half the length or width of the minimum bounding box of the part of the point cloud as the radius, and use it as the initial rough cylinder in the tree cylinder fitting process. After determining the top center, bottom center, and radius of the initial cylinder, the cylinder is optimized using the nonlinear least squares method. The optimized cylinder becomes the fitting cylinder of the final tree reconstruction model.

8. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 1, wherein: In S5, the specific contents of forest biomass calculation are as follows: According to the quantitative structural model, the volume of the cylinders contained in the trunks and branches of the trees was counted respectively to calculate the volume of the trunks and side branches of the trees. The biomass of individual trees was calculated with the help of the basic wood density of the trees. Specifically, Q=V QSM ×D Among them, Q represents the biomass of a single tree, V QSM It represents the volume of the trunk or side branches, and D represents the wood density of the corresponding tree species.

9. The method for estimating individual tree biomass based on a forest quantitative structure model according to claim 1, wherein: In S6, the accuracy of tree segmentation is evaluated by comparing the tree location record information with the results of tree segmentation. The number of correctly segmented trees (TP), incorrectly detected trees (FP), and undetected trees (FN) obtained after segmentation is statistically calculated. The tree detection rate r, the correct tree segmentation rate (p), and the overall accuracy F, which takes into account both misclassification and omission, are calculated using the following formula: r=TP / (TP+FN) p=TP / (TP+FP) F = (r × p) / (r + p) The coefficient of determination R 2 , root mean square error RMSE and relative root mean square error rRMSE are used to evaluate the effect and accuracy of single tree biomass estimation, specifically: Among them, x i represents the measured value of aboveground biomass of a single tree; represents the measured average aboveground biomass of individual trees; represents the estimated value of the aboveground biomass of a single tree; n represents the number of samples; i represents a single tree sample.

Images