Forest parameter extraction method and system based on incomplete lidar point cloud data

Abstract

The present invention relates to a forest parameter extraction method and system based on incomplete laser radar point cloud data, the method comprising performing laser radar scanning on a pre-set sample plot, obtaining a laser radar data point cloud and preprocessing it; performing single-tree segmentation on the laser radar data point cloud, obtaining a single-tree point cloud and extracting a single-tree breast diameter; extracting a single-tree theoretical tree height according to the single-tree point cloud, and calculating single-tree estimation information according to the single-tree theoretical tree height and the single-tree breast diameter. The present invention scans the sample plot, preprocesses the acquired laser radar data point cloud, and then performs single-tree segmentation, extracts the single-tree breast diameter and the single-tree theoretical tree height according to the obtained single-tree point cloud, and then obtains the single-tree estimation information. Only using incomplete laser radar data, for sample plots with different complex environments, extracts forest stand parameters such as breast diameter, tree height, single-tree volume and sample plot stocking, and provides data and technical support for sustainable planning and management of forests.

Description

Technical Field

[0001] The present invention relates to the technical field of forest sustainable planning and management research, and in particular to a forest parameter extraction method and system based on incomplete laser radar point cloud data. Background Art

[0002] Sustainable planning and management of forests require accurate forest resource survey information. The traditional method is to conduct a large number of field surveys on standard plots manually, and use parameters such as the breast diameter, canopy diameter and tree height of individual trees in the sample data to estimate the stand information of the entire forest, which consumes a lot of human resources and time costs. In addition, the tree height is generally determined by visual experience during field measurements, and there are certain errors in its value. Therefore, the traditional forest resource survey method can no longer meet the requirements of obtaining forest resource information efficiently and accurately.

[0003] Light detection and ranging (Li DAR) technology has been used in forest resource surveys in the past two decades. At the plot level, ground-based LiDAR has been proven to be a method that can obtain high-quality point cloud data and extract various forest stand parameters from it, but it cannot strike a good balance between cost and efficiency. Obtaining better data means that the instrument needs to be set up multiple times, which increases manpower and time costs. To solve this problem, handheld laser mines provide an innovative and non-destructive method for obtaining forest resource information. It can collect forest 3D point cloud data at the plot scale more quickly and efficiently, and has begun to be applied to survey practices. However, most of my country's forests are large, densely distributed, and have high canopy density, making it difficult to obtain single tree information. Therefore, handheld LiDAR technology has not yet been truly applied to forest surveys. At the same time, when obtaining handheld LiDAR, data collection is difficult due to complex terrain and complex forest structure, resulting in poor data quality. Since the equipment scans from bottom to top and the scanning range is 15m, it is impossible to obtain a complete three-dimensional point cloud for trees with a height greater than 15m. Therefore, the traditional method of using the highest point minus the lowest point as the tree height is not feasible. When obtaining the volume of a single tree or the volume of a sample plot, parameters other than LiDAR are usually used, such as the binary volume equation, wood density, and stand density, which often requires sample plot survey data or empirical data, increasing the workload. Sustainable forest planning and management requires more efficient and accurate forest resource survey methods. Although LiDAR technology has potential in forest resource surveys, there are still some challenges that need to be addressed before it can be widely used.

[0004] Through the above analysis, the problems and defects of the prior art are as follows:

[0005] (1) Handheld LiDAR is not widely used in forest resource surveys, and the technical methods for obtaining forest stand parameter information using it are still not mature enough.

[0006] (2) Due to the scanning method and scanning range of the handheld laser radar, it is impossible to obtain a complete three-dimensional point cloud for trees taller than 15 meters. The traditional method of subtracting the lowest point from the highest point to get the tree height is not feasible.

[0007] (3) When calculating the volume of a single tree and the volume of a stand, plot survey data or empirical formulas other than lidar will be used, which will greatly increase the workload. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a forest parameter extraction method and system based on incomplete lidar point cloud data in view of the deficiencies of the above-mentioned prior art.

[0009] The technical solution of the present invention to solve the above technical problem is as follows: A forest parameter extraction method based on incomplete lidar point cloud data comprises the following steps:

[0010] Performing laser radar scanning on a preset sample plot to obtain a laser radar data point cloud, and preprocessing the laser radar data point cloud;

[0011] Using a density clustering algorithm to perform single-tree segmentation on the laser radar data point cloud to obtain a single-tree point cloud, and extracting a single-tree breast diameter based on the single-tree point cloud;

[0012] The theoretical tree height of a single tree is extracted based on the single tree point cloud, and the estimated information of the single tree is obtained based on the theoretical tree height of the single tree and the breast height diameter of the single tree.

[0013] The beneficial effects of the present invention are as follows: the forest parameter extraction method based on incomplete lidar point cloud data of the present invention scans the sample plot, pre-processes the acquired lidar data point cloud, and then performs single tree segmentation, extracts the single tree breast diameter and the single tree theoretical tree height according to the obtained single tree point cloud, and then obtains the single tree estimation information, without the help of sample plot survey data and empirical data, only using incomplete lidar data, for sample plots in environments of different complexity, extracts the breast diameter, tree height, single tree volume and sample plot stocking and other forest stand parameters, thereby providing data and technical support for sustainable planning and management of forests.

[0014] On the basis of the above technical solution, the present invention can also be improved as follows:

[0015] Further: the preprocessing of the laser radar data point cloud specifically includes the following steps:

[0016] Performing point cloud registration on the laser radar data point cloud, and cropping the laser radar data point cloud according to preset control points;

[0017] Orient the laser radar data point cloud, perform voxel downsampling processing on the laser radar data point cloud at a set spatial resolution, and perform filtering processing on the laser radar data point cloud after the voxel downsampling processing;

[0018] Ground points are separated from the laser radar data point cloud, and elevation normalization is performed on the laser radar data point cloud according to the heights of the separated ground points.

[0019] The beneficial effect of the above further scheme is: by performing point cloud registration on the lidar data point cloud, and performing cropping, voxel downsampling and filtering processing, the amount of data can be greatly reduced, and the ground points can be separated and normalized by the lidar data point cloud, thereby eliminating the influence of terrain on ground vegetation estimation, and facilitating the subsequent calculation of the accurate relative height of the tree point cloud.

[0020] Further: the step of using a density clustering algorithm to segment the laser radar data point cloud into individual trees to obtain individual tree point clouds, and extracting individual tree parameter breast diameter according to the individual tree point clouds specifically includes the following steps:

[0021] Using the density clustering DBSCN algorithm, the LiDAR data point cloud is clustered according to the preset inter-layer thickness to screen out the point cloud of suspected tree trunks in each layer;

[0022] The clustering parameter threshold is optimized according to the screening results and the original point cloud trunk of the sample plot, and the layer with the best clustering result is selected as the trunk layer for preservation, wherein the clustering parameter threshold includes the minimum number of point clouds and the search radius;

[0023] Extracting the center position information of the trunk from the trunk point cloud of the trunk layer, and segmenting and clipping the trunk point cloud with a buffer determined by the center position of the trunk to obtain a candidate single tree point cloud;

[0024] Taking the center position of the trunk as the center, the buffer zone is cylindrically sheared according to a set length as the shearing radius to extract a single tree point cloud;

[0025] The breast diameter layer point cloud of the single tree point cloud is intercepted according to the preset interception height, and the breast diameter of the single tree is calculated using the cylindrical fitting algorithm. The calculation formula is:

[0026] r 2 =(x-x0) 2 +(y-y0) 2 +(z-z0) 2 -[a(x-x0)+b(y-y0)+c(z-z0)]

[0027] Among them, (a, b, c) is the cylinder axis vector, (x0, y0, z0) is a point on the cylinder axis, and r is the radius of the cylinder base.

[0028] The beneficial effect of the above further scheme is: the trunk point cloud clustering processing is performed on the laser radar data point cloud by the density clustering DBSCN algorithm, and the trunk layer point cloud is screened out, and then the candidate single tree point cloud is determined, and the cylinder shearing is performed to obtain the single tree point cloud, and the single tree breast diameter is calculated by combining the breast diameter layer point cloud obtained by the preset interception height.

[0029] Further: the extraction of single tree theoretical height according to the single tree point cloud specifically includes the following steps:

[0030] A region growing algorithm is used to extract the single tree point cloud to obtain a single tree trunk point cloud;

[0031] The Kunz-Gan curve equation is constructed with the center line of the trunk as the X-axis direction, the radius of the trunk cross section as the Y-axis direction, and the treetop as the origin:

[0032]

[0033] Among them, y i represents the radius of the cross section of the i-th layer of the trunk, the theoretical height of a single tree is H, and the height of the i-th cross section from the ground is h i , P is the coefficient to be determined;

[0034] The radius y of each cross section in the single tree trunk point cloud i and height from ground h i Substituting it into the equation of the Kunz trunk curve, the optimal trunk curve is fitted and the theoretical tree height H of a single tree is calculated.

[0035] The beneficial effect of the above further scheme is: the point clouds with similarity are grouped together to form a single tree trunk point cloud through the region growing algorithm, and then the Kunz curve equation is constructed, and then the radius y of each cross section in the single tree trunk point cloud is used to calculate the i and height from ground h i The optimal trunk curve is fitted and then the theoretical tree height H of a single tree is calculated.

[0036] Further: the extraction of theoretical tree height of a single tree according to the single tree point cloud also includes the following steps:

[0037] Conducting actual measurements on a preset sample plot to obtain actual measurement information of individual trees in the sample plot, wherein the actual measurement information of individual trees at least includes the name of the individual tree species, the actual measurement coordinates of the individual tree, the actual measurement breast diameter of the individual tree, and the actual measurement height of the individual tree;

[0038] Compare the theoretical tree height H of the single tree with the actual measured tree height of the single tree, and determine whether the absolute error between the two is less than a preset threshold;

[0039] When the absolute error between the theoretical tree height H of a single tree and the measured tree height of a single tree is less than a preset threshold, the theoretical tree height H of a single tree and the corresponding breast diameter of the single tree are fitted with a tree height curve equation to obtain a tree height curve equation:

[0040] H=a1+b1*log(D)

[0041] Wherein, a1 and b1 are parameters of the tree height curve equation;

[0042] When the absolute error between the theoretical tree height H of a single tree and the actual measured tree height of a single tree is greater than or equal to a preset threshold, the single tree diameter at breast height corresponding to the theoretical tree height H of a single tree is substituted into the tree height curve equation to calculate and update the corresponding theoretical tree height H of a single tree;

[0043] The theoretical tree height H of a single tree whose absolute error with the actual measured tree height is less than a preset threshold and the theoretical tree height H of a single tree whose absolute error with the actual measured tree height is greater than or equal to the preset threshold and updated are merged to obtain the theoretical tree height of all single trees in the sample plot.

[0044] The beneficial effect of the above further scheme is: by comparing the theoretical tree height H of a single tree with the actual measured tree height of a single tree, the theoretical tree height of a single tree with better extraction accuracy can be screened out, and the tree height curve equation is fitted with the corresponding single tree breast diameter to obtain the tree height curve equation, and then the theoretical tree height of a single tree with poor extraction accuracy is updated by the tree height curve equation, so as to obtain a more accurate theoretical tree height of a single tree, thereby facilitating subsequent calculations to obtain more accurate single tree estimation information.

[0045] Further: the step of calculating the single tree estimated information based on the single tree theoretical tree height and the single tree diameter at breast height specifically includes the following steps:

[0046] Based on the binary volume table, the measured volume of a single tree and the measured stock volume of the sample plot are calculated according to the measured information of the single tree;

[0047] Substitute the measured timber volume of the single tree into the timber volume integral calculation formula to calculate the fitting parameter a of the timber volume integral calculation formula, and calculate the theoretical timber volume of all single trees and the theoretical volume of the sample plot according to the timber volume integral calculation formula. The timber volume integral calculation formula is:

[0048]

[0049] V=a2·LI a2

[0050] Among them, V is the theoretical volume of a single tree, LI is the laser radar index, and H isT is the trunk height, a2 and a3 are fitting parameters, and D(H) is the trunk diameter when the height is H.

[0051] The beneficial effect of the above further scheme is: the fitting parameters a2 and a3 in the volume integral calculation formula are solved by the actual measured volume of the single tree, so that the theoretical volume of all single trees can be accurately calculated according to the volume integral calculation formula, and the theoretical volume of the sample plot can be calculated.

[0052] Further: the forest parameter extraction method based on incomplete lidar point cloud data also includes the following steps:

[0053] The single tree measured information is matched with the single tree estimated information, and the accuracy of the single tree segmentation result is evaluated according to the matching result. The specific calculation formula is:

[0054]

[0055]

[0056]

[0057] Among them, P is the precision, R is the recall, F is the overall precision, TP is the number of matched trees corresponding to the estimated information of individual trees that can match the measured information of individual trees, FP is the number of individual trees corresponding to the estimated information of individual trees that were not measured, that is, the number of misdetected individual trees, and FN is the number of individual trees corresponding to the measured but not extracted estimated information of individual trees, that is, the number of missed individual trees;

[0058] Calculate the determination coefficient R according to the single tree measured information and the single tree estimated information 2 And the root mean square error RMSE, the calculation formula is:

[0059]

[0060]

[0061] in, is the estimated parameter value of single tree i, y i is the measured value of the parameter of single tree i, is the arithmetic mean of the measured values.

[0062] The beneficial effects of the above further scheme are: the accuracy, recall rate and overall accuracy are used to respectively characterize the proportion of correct single tree segmentation to the number of extracted single trees, the proportion of the number of correct single tree segmentation to the actual number of single trees, and the overall accuracy that comprehensively considers misclassification and omission, as well as the determination coefficient and root mean square error are calculated based on the actual measured information of single trees and the estimated information of single trees, so that the extraction results of single tree parameters can be accurately evaluated.

[0063] The present invention also provides a forest parameter extraction system based on incomplete laser radar point cloud data, including a scanning processing module, a segmentation extraction module, a tree height extraction module and an estimation calculation module;

[0064] The scanning and processing module is used to perform a laser radar scan on a preset sample plot, obtain a laser radar data point cloud, and pre-process the laser radar data point cloud;

[0065] The segmentation and extraction module is used to perform single-tree segmentation on the laser radar data point cloud using a density clustering algorithm to obtain a single-tree point cloud, and extract the single-tree breast diameter according to the single-tree point cloud;

[0066] The tree height extraction module is used to extract the theoretical tree height of a single tree according to the single tree point cloud;

[0067] The estimation calculation module is used to calculate the single tree estimation information based on the single tree theoretical tree height and the single tree breast diameter.

[0068] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the forest parameter extraction method based on incomplete lidar point cloud data.

[0069] The present invention also provides a forest parameter extraction device based on incomplete laser radar point cloud data, comprising a communication interface, a memory, a communication bus and a processor, wherein the processor, the communication interface and the memory communicate with each other through the communication bus;

[0070] The memory is used to store computer programs;

[0071] The processor is used to implement the steps of the forest parameter extraction method based on incomplete lidar point cloud data when executing the program stored in the memory. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of a process of extracting forest parameters based on incomplete laser radar point cloud data according to an embodiment of the present invention;

[0073] Figure 2 A schematic diagram showing the comparison of laser radar data point cloud before and after preprocessing according to an embodiment of the present invention;

[0074] Figure 3 A schematic diagram of point cloud extraction of a single tree trunk according to an embodiment of the present invention;

[0075] Figure 4This is a schematic diagram of the results of a forest parameter extraction system based on incomplete lidar point cloud data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0077] like Figure 1 As shown, a forest parameter extraction method based on incomplete lidar point cloud data includes the following steps:

[0078] S1: performing a laser radar scan on a preset sample plot, obtaining a laser radar data point cloud, and preprocessing the laser radar data point cloud;

[0079] S2: using a density clustering algorithm to perform single-tree segmentation on the laser radar data point cloud to obtain a single-tree point cloud, and extracting a single-tree breast diameter based on the single-tree point cloud;

[0080] S3: extracting the theoretical height of a single tree according to the single tree point cloud;

[0081] S4: Calculate the estimated information of a single tree based on the theoretical tree height and the breast height diameter of the single tree.

[0082] The forest parameter extraction method based on incomplete lidar point cloud data of the present invention scans the sample plot, pre-processes the acquired lidar data point cloud, and then performs single tree segmentation, extracts the single tree breast diameter and the single tree theoretical tree height according to the obtained single tree point cloud, and then obtains the single tree estimation information. Without relying on sample plot survey data and empirical data, only using incomplete lidar data, for sample plots in environments of different complexity, the forest stand parameters such as breast diameter, tree height, single tree volume and sample plot stocking are extracted, providing data and technical support for sustainable planning and management of forests.

[0083] In one or more embodiments of the present invention, the preprocessing of the laser radar data point cloud specifically includes the following steps:

[0084] S11: performing point cloud registration on the laser radar data point cloud, and cropping the laser radar data point cloud according to preset control points;

[0085] Here, the ZEB-Revo handheld lidar scanner is used to collect lidar data, and the GeoSLAM-Hub is used to solve the point cloud to generate three-dimensional point cloud data in a standard format.

[0086] S12: coordinate-orienting the laser radar data point cloud, performing voxel-downsampling processing on the laser radar data point cloud at a set spatial resolution, and filtering the laser radar data point cloud after the voxel-downsampling processing;

[0087] Point cloud data processing does not require such a large number of points. In order to improve the speed of computer operation, the point cloud data of a single plot needs to be thinned. The point cloud is thinned by voxelizing the point cloud. Under a given spatial resolution, the present invention selects a random echo point in each voxel range of 0.02m×0.02m×0.02m to be retained to reduce the number of points. In order to reduce the amount of calculation, the three-dimensional laser radar point cloud data is downsampled.

[0088] In the measurement process, due to the systematic error of the instrument and the physical characteristics of the measured object, unreasonable noise points are often generated, which are removed by point cloud denoising. After downsampling, the amount of point cloud data is reduced, but some redundant information such as outliers and messy points generated by measurement errors still exist. The present invention uses a statistical filter (Statistical Outlier Removal) to remove this redundant information.

[0089] S13: Separate ground points from the laser radar data point cloud, and perform elevation normalization processing on the laser radar data point cloud according to the heights of the separated ground points.

[0090] Due to the complex terrain of the field sample plots, the trees have different starting heights and growth directions. In order to eliminate the impact of the terrain on the estimation of ground vegetation, the point cloud needs to be normalized according to the height of the ground points. The present invention uses the Cloth Simulation Filter (CSF) algorithm to separate the ground points. CSF simulates a piece of cloth covering the flipped point cloud surface, and then generates an approximate surface shape by analyzing the interaction between the cloth nodes and the corresponding lidar points, thereby achieving point cloud ground classification. The height z value of each point is subtracted from the elevation of the ground point corresponding to the point, so that the bottom of the single tree is at the same height, so as to obtain a normalized point cloud, which is convenient for calculating the accurate relative height of the tree point cloud. Figure 2 As shown, Figure 2 a and Figure 2 b are schematic diagrams of the lidar data before and after preprocessing, respectively.

[0091] By performing point cloud registration, cropping, voxel downsampling and filtering on the LiDAR data point cloud, the amount of data can be greatly reduced. By separating ground points and normalizing the LiDAR data point cloud, the influence of terrain on ground vegetation estimation can be eliminated, facilitating the subsequent calculation of the accurate relative height of the tree point cloud.

[0092] In one or more embodiments of the present invention, the step of using a density clustering algorithm to segment the laser radar data point cloud into individual trees to obtain individual tree point clouds, and extracting individual tree parameter diameter at breast height according to the individual tree point clouds specifically includes the following steps:

[0093] S21: using a density clustering DBSCN algorithm, performing tree trunk point cloud clustering processing on the laser radar data point cloud according to a preset inter-layer thickness, and screening out point clouds suspected to be tree trunks in each layer;

[0094] The density clustering DBSCN algorithm was used to cluster the preprocessed sample site lidar data point cloud. The tree trunk point cloud was recursively clustered with a thickness of 0.3m per layer, and the point cloud that might be the tree trunk in each layer was preliminarily screened out.

[0095] S22: optimizing the clustering parameter threshold according to the screening results and the original point cloud trunk of the sample plot, and selecting a layer with the best clustering result as the trunk layer for saving, wherein the clustering parameter threshold includes the minimum point cloud number and the search radius;

[0096] Here, whether the clustering parameter threshold setting is reasonable is judged according to the matching result between the clustering result and the original point cloud of the sample plot. If it is unreasonable, the clustering parameter threshold is adjusted and the process returns to S21 until the expected clustering effect is achieved.

[0097] S23: extracting the center position information of the trunk from the trunk point cloud of the trunk layer, and segmenting and clipping the trunk point cloud using a buffer determined by the center position of the trunk to obtain a candidate single tree point cloud;

[0098] Here, the center position information of each trunk, i.e., the (x, y) coordinates, is extracted from the trunk layer with the best clustering result, and a 0.5m buffer is set according to the (x, y) coordinates. The point cloud of the entire buffer is cropped out as a candidate single tree point cloud, and the single tree point cloud of each single tree is stored in a separate point cloud file for the next experimental operation.

[0099] S24: taking the center position of the trunk as the center, performing cylindrical shearing on the buffer zone according to a set length as a shearing radius, and extracting a single tree point cloud;

[0100] For the single tree point cloud after single tree segmentation, the point cloud of the tree is roughly extracted by cutting a cylinder centered on the trunk position, with a cutting radius of 1.5m if the DBH is less than 0.4m and a cutting radius of 3m if the DBH is greater than 0.4m. This cylindrical subset of the single tree point cloud contains the entire target tree, but also includes the ground and adjacent vegetation that need to be removed. CloudCompare software is used to delete all points belonging to the ground, adjacent plants, and other points that are considered not to belong to the target single tree.

[0101] S25: intercepting the breast diameter layer point cloud of the single tree point cloud according to a preset interception height, and calculating the breast diameter of the single tree using a cylindrical fitting algorithm, the calculation formula is:

[0102] r 2 =(x-x0) 2 +(y-y0) 2 +(z-z0) 2 -[a(x-x0)+b(y-y0)+c(z-z0)]

[0103] Among them, (a, b, c) is the cylinder axis vector, (x0, y0, z0) is a point on the cylinder axis, and r is the radius of the cylinder base. Input the three-dimensional single tree DBH layer point cloud, and use the cylinder fitting algorithm to solve the above parameters, so that the sum of the squares of the distance from the discrete point of the DBH layer point cloud slice to the cylinder centerline and the estimated radius is minimized, and finally the corresponding single tree DBH value can be obtained according to the fitted cylinder radius value r.

[0104] After obtaining the single tree point cloud by the above method, the DBH layer point cloud is intercepted at 1.2-1.4m from the ground surface. The present invention selects a cylinder fitting algorithm to realize the DBH estimation of a single tree. This method uses the three-dimensional spatial distribution law of the point cloud to perform cylinder detection, which can well overcome the shortcoming of the two-dimensional DBH estimation method that it cannot accurately estimate the DBH when the tree is tilted.

[0105] The density clustering DBSCN algorithm is used to cluster the trunk point cloud of the lidar data point cloud, and the trunk layer point cloud is screened out to determine the candidate single tree point cloud, and the cylinder is sheared to obtain the single tree point cloud. The single tree breast diameter is calculated by combining the breast diameter layer point cloud obtained at the preset interception height.

[0106] Due to the performance limitations of handheld lidar equipment, its maximum scanning radius in the wild is between 10m and 15m. It is basically impossible to measure the height of trees approaching or exceeding 15m in height. In addition, in a forest environment with a high density of canopy, the handheld lidar cannot fully obtain the canopy information. However, in the course of the experiment, it was found that although the canopy information could not be fully obtained, the trunk point cloud information under the canopy was easier to obtain. Based on this situation, the present invention uses the extracted trunk point cloud combined with the Koontz trunk curve formula and the tree height curve equation to fit and extract the height of a single tree.

[0107] In one or more embodiments of the present invention, extracting the theoretical height of a single tree according to the single tree point cloud specifically includes the following steps:

[0108] S31: extracting the single tree point cloud by using a region growing algorithm to obtain a single tree trunk point cloud;

[0109] The region growing algorithm gathers similar point clouds into a single point cloud set. First, a seed point is found for each point cloud to be segmented as the starting point of growth. Then, points with the same or similar properties as the seed in the neighborhood around the seed point are merged into the region where the seed point is located. The new point continues to grow around as a seed until no more points that meet the conditions can be included. Then, a region is grown. The present invention first selects the trunk seed point based on the extracted breast diameter layer point cloud, and then uses the region growing algorithm to grow the entire trunk point cloud in the single tree trunk point cloud based on the trunk seed point. Figure 3 shown.

[0110] S32: With the center line of the trunk as the X-axis direction, the radius of the trunk cross section as the Y-axis direction, and the treetop as the origin, the equation of the Kunz-Gan curve is constructed:

[0111]

[0112] Among them, y i represents the radius of the cross section of the i-th layer of the trunk, the theoretical height of a single tree is H, and the height of the i-th cross section from the ground is h i , P is the coefficient to be determined;

[0113] S33: The radius y of each cross section in the single tree trunk point cloud is i and height from ground h i Substituting it into the equation of the Kunz trunk curve, the optimal trunk curve is fitted and the theoretical tree height H of a single tree is calculated.

[0114] The region growing algorithm is used to group similar point clouds together to form a single tree trunk point cloud, and then the Kunz curve equation is constructed. The radius y of each cross section in the single tree trunk point cloud is used to calculate the i and height from ground h i The optimal trunk curve is fitted and then the theoretical tree height H of a single tree is calculated.

[0115] It should be pointed out that the tree height obtained by using the Koontz curve fitting is not completely accurate, because some trunk point clouds have poor quality, such as the trunk is not straight, the trunk point cloud extraction is incomplete, etc., which will result in poor fitting effect. Therefore, the present invention uses the tree height curve equation to fit the extracted tree height (the absolute error is less than 1.5m compared with the measured tree height) and the trunk diameter at breast height, and uses the fmin_powell function to obtain the optimal solution of the two equation parameters a1 and b1 in the tree height curve equation, and obtains the optimal tree height curve equation for this type of plot. Finally, the diameter at breast height value corresponding to the tree height with poor Koontz curve fitting is substituted into the equation to obtain the tree height.

[0116] Specifically, in one or more embodiments of the present invention, the extracting the theoretical height of a single tree according to the single tree point cloud further includes the following steps:

[0117] S34: Conducting actual measurement on a preset sample plot to obtain actual measurement information of individual trees in the sample plot, wherein the actual measurement information of individual trees at least includes the name of the individual tree species, the actual measurement coordinates of the individual tree, the actual measurement breast diameter of the individual tree, and the actual measurement height of the individual tree;

[0118] Specifically, firstly, the single tree measured parameter information and unit measured position information of single trees in the sample plot are obtained, wherein the single tree measured parameter information at least includes the tree species name, coordinates, breast diameter and tree height, and the sample plot types include simple forest environment, low complexity environment and medium complexity environment; then, the binary volume equation is used to calculate the single tree measured volume and forest stand stock according to the single tree measured parameter information and unit measured position information; at the same time, the sample plot is divided into types according to the complexity according to the preset sample plot type classification standards and the sample plot understory visibility conditions and shrub density.

[0119] S35: comparing the theoretical tree height H of the single tree with the measured tree height of the single tree, and determining whether the absolute error between the two is less than a preset threshold;

[0120] S36: When the absolute error between the theoretical tree height H of a single tree and the measured tree height of a single tree is less than a preset threshold, the theoretical tree height H of a single tree and the corresponding breast diameter of the single tree are fitted with a tree height curve equation to obtain a tree height curve equation:

[0121] H=a1+b1*log(D)

[0122] Wherein, a1 and b1 are parameters of the tree height curve equation;

[0123] S37: When the absolute error between the theoretical tree height H of a single tree and the measured tree height of a single tree is greater than or equal to a preset threshold, the single tree diameter at breast height corresponding to the theoretical tree height H of a single tree is substituted into the tree height curve equation to calculate and update the corresponding theoretical tree height H of a single tree;

[0124] S38: Merge the theoretical tree height H of a single tree whose absolute error with the actual measured tree height is less than a preset threshold and the theoretical tree height H of a single tree whose absolute error with the actual measured tree height is greater than or equal to the preset threshold and updated, to obtain the theoretical tree height of all single trees in the sample plot.

[0125] By comparing the theoretical tree height H of a single tree with the actual measured tree height of a single tree, the theoretical tree height of a single tree with better extraction accuracy can be screened out, and the tree height curve equation is fitted with the corresponding single tree breast diameter to obtain the tree height curve equation. The theoretical tree height of a single tree with poor extraction accuracy is then updated using the tree height curve equation to obtain a more accurate theoretical tree height of a single tree, thereby facilitating subsequent calculations to obtain more accurate single tree estimation information.

[0126] After obtaining the measured information of individual trees, the measured tree diameter at breast height and tree height of individual trees in the sample plot are used to calculate the measured volume of individual trees according to the binary volume table, and finally the measured volume of a single sample plot is obtained. For the individual tree diameter at breast height and tree height extracted from the LiDAR data point cloud, the integral method is used to use the measured volume of individual trees calculated using the measured data as the true value V, which is brought into the volume integral calculation formula, and the parameter a2 is fitted. Then, the volume of all individual trees is calculated using the volume integral calculation formula, and the volume of all individual trees in the sample plot is calculated by adding up the volume of all individual trees in the sample plot to obtain the volume of the sample plot.

[0127] Specifically, in one or more embodiments of the present invention, the step of calculating the single tree estimation information according to the single tree theoretical tree height and the single tree diameter at breast height comprises the following steps:

[0128] S41: Based on the binary volume table, the measured volume of a single tree and the measured stock volume of the sample plot are calculated according to the measured information of the single tree;

[0129] S42: Substituting the measured timber volume of the single tree into the timber volume integral calculation formula to calculate the fitting parameter a of the timber volume integral calculation formula, and calculating the theoretical timber volume of all single trees and the theoretical volume of the sample plot according to the timber volume integral calculation formula, the timber volume integral calculation formula is:

[0130]

[0131] V=a2·LI a2

[0132] Among them, V is the theoretical volume of a single tree, LI is the laser radar index, and H is T is the trunk height, a2 and a3 are fitting parameters, and D(H) is the trunk diameter when the height is H.

[0133] The fitting parameters a2 and a3 in the volume integral calculation formula are solved by the measured volume of the single trees, so that the theoretical volume of all single trees can be accurately calculated according to the volume integral calculation formula, and the theoretical volume of the sample plot can be calculated.

[0134] After the extraction is completed, the accuracy of tree segmentation needs to be evaluated. First, the key information extracted from the LiDAR data source, such as tree ID, tree coordinates (x, y), and tree DBH, is sorted out, and then the matching error threshold is set to match the ground survey data. For each tree, the DBH, coordinates, and the DBH and tree coordinates in the LiDAR data point cloud are combined to find the most similar tree. Then, the corresponding tree ID, tree coordinates (x, y), etc. are merged into one file, and the paired tree information of each plot is output.

[0135] The accuracy of single tree segmentation is tested by matching single trees corresponding to the single tree breast diameter and single tree coordinates extracted from the laser radar data point cloud. The matching trees, missed trees and wrongly classified trees of correct single tree segmentation are calculated. The proportion of the number of correct segmentations to the extracted number is defined as the accuracy P (precision), the proportion of the number of correct segmentations to the actual number is defined as the recall R (recall), and the overall accuracy F (F-Measure) which comprehensively considers the wrong and missed classifications is used.

[0136] Optionally, in one or more embodiments of the present invention, the forest parameter extraction method based on incomplete lidar point cloud data further includes the following steps:

[0137] S5: Matching the single tree measured information with the single tree estimated information, and evaluating the accuracy of the single tree segmentation result according to the matching result. The specific calculation formula is:

[0138]

[0139]

[0140]

[0141] Among them, P is the precision, R is the recall, F is the overall precision, TP is the number of matched trees corresponding to the estimated information of individual trees that can match the measured information of individual trees, FP is the number of individual trees corresponding to the estimated information of individual trees that were not measured, that is, the number of misdetected individual trees, and FN is the number of individual trees corresponding to the measured but not extracted estimated information of individual trees, that is, the number of missed individual trees;

[0142] S6: Calculate the determination coefficient R based on the single tree measured information and the single tree estimated information 2 And the root mean square error RMSE, the calculation formula is:

[0143]

[0144] in, is the estimated parameter value of single tree i, y i is the measured value of the parameter of single tree i, is the arithmetic mean of the measured values.

[0145] In the present invention, the coefficient of determination (R2) and the root mean square error (RMSE) are used as evaluation indicators for the results of forest parameter extraction. The coefficient of determination R2 is a test of the degree of fit of the regression line, reflecting the degree of correlation between the dependent variable and the independent variable. The value range is 0-1. The closer the value is to 1, the higher the correlation is, the stronger the explanatory power of the independent variable on the dependent variable is, and the better the degree of fit of the model is; the root mean square error RMSE represents the square root of the ratio of the sum of squares of the errors between the estimated value and the measured value to the number of samples n. The smaller the value is, the better the model prediction effect is, and the closer the estimated value is to the measured value.

[0146] The accuracy, recall rate and overall precision are used to represent the proportion of correct tree segmentation to the number of extracted trees, the proportion of correct tree segmentation to the actual number of trees, and the overall precision that takes into account misclassification and omission. The determination coefficient and root mean square error are calculated based on the actual tree measurement information and tree estimation information, so that the extraction results of tree parameters can be accurately evaluated.

[0147] In the present invention, based on the results of matching the laser radar data point cloud with the measured information of single trees in the sample plot, the segmented single tree point cloud is superimposed with the original sample point cloud data according to the cloudcompare software to verify the accuracy of visual interpretation. The results are shown in Table 1:

[0148] Table 5 Verification results of single tree segmentation accuracy in different types of plots

[0149]

[0150] According to the experimental data of the present invention (see Table 1 for details), the accuracy of single tree segmentation is the best in simple forest environments and low-complexity forest environments, with the detection accuracy Precision, segmentation accuracy Recall, and comprehensive accuracy F-Measure all reaching more than 91%. In the Platycladus orientalis plantations and Cryptomeria fortunei plantations, the comprehensive accuracies were 96.7% and 93.6%, respectively. In comparison, the comprehensive accuracies of poplar plantations and Phellodendron amurense plantations were slightly lower, at 95.1% and 91.8%, respectively. The accuracy of single tree segmentation in moderately complex environments is relatively low, with the comprehensive accuracies of mixed broad-leaved forests and mixed coniferous and broad-leaved forests being 83% and 83.2%, respectively.

[0151] In general, the present invention investigated sample plots in forest areas of different complexity, and measured a total of 1,280 trees, of which 1,058 matched the individual trees extracted from the LiDAR data point cloud. The results showed that the accuracy of individual tree segmentation using LiDAR data point cloud was high, with a detection accuracy of 90.8%, a segmentation accuracy of 89.9%, and an overall accuracy of 90.4%. In addition, the overall accuracy of individual tree segmentation for coniferous species was better than that for broad-leaved species.

[0152] like Figure 4 As shown, the present invention also provides a forest parameter extraction system based on incomplete laser radar point cloud data, including a scanning processing module, a segmentation extraction module, a tree height extraction module and an estimation calculation module;

[0153] The scanning and processing module is used to perform a laser radar scan on a preset sample plot, obtain a laser radar data point cloud, and pre-process the laser radar data point cloud;

[0154] The segmentation and extraction module is used to perform single-tree segmentation on the laser radar data point cloud using a density clustering algorithm to obtain a single-tree point cloud, and extract the single-tree breast diameter according to the single-tree point cloud;

[0155] The tree height extraction module is used to extract the theoretical tree height of a single tree according to the single tree point cloud;

[0156] The estimation calculation module is used to calculate the single tree estimation information based on the single tree theoretical tree height and the single tree breast diameter.

[0157] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the forest parameter extraction method based on incomplete lidar point cloud data.

[0158] The present invention also provides a forest parameter extraction device based on incomplete laser radar point cloud data, comprising a communication interface, a memory, a communication bus and a processor, wherein the processor, the communication interface and the memory communicate with each other through the communication bus;

[0159] The memory is used to store computer programs;

[0160] The processor is used to implement the steps of the forest parameter extraction method based on incomplete lidar point cloud data when executing the program stored in the memory.

[0161] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A forest parameter extraction method based on incomplete lidar point cloud data, characterized in that: The steps include: Performing laser radar scanning on a preset sample plot to obtain a laser radar data point cloud, and preprocessing the laser radar data point cloud; Using a density clustering algorithm to perform single-tree segmentation on the laser radar data point cloud to obtain a single-tree point cloud, and extracting a single-tree breast diameter based on the single-tree point cloud; Extracting theoretical tree height of a single tree according to the single tree point cloud; Obtaining single tree estimation information based on the single tree theoretical tree height and single tree breast diameter; The extraction of single tree theoretical height according to the single tree point cloud specifically includes the following steps: A region growing algorithm is used to extract the single tree point cloud to obtain a single tree trunk point cloud; The center line of the trunk is taken as the X-axis direction, the radius of the trunk cross section is taken as the Y-axis direction, and the treetop is taken as the origin to construct the equation of the Kunz-Gan curve: Among them, y i represents the radius of the cross section of the i-th layer of the trunk, the theoretical height of a single tree is H, and the height of the i-th cross section from the ground is h i , P is the coefficient to be determined; The radius y of each cross section in the single tree trunk point cloud i and height from ground h i Substitute it into the equation of the Kunz trunk curve, fit the optimal trunk curve, and calculate the theoretical tree height H of a single tree; The extracting of the theoretical height of a single tree according to the single tree point cloud further comprises the following steps: Conducting actual measurements on a preset sample plot to obtain actual measurement information of individual trees in the sample plot, wherein the actual measurement information of individual trees at least includes the name of the individual tree species, the actual measurement coordinates of the individual tree, the actual measurement breast diameter of the individual tree, and the actual measurement height of the individual tree; Compare the theoretical tree height H of the single tree with the actual measured tree height of the single tree, and determine whether the absolute error between the two is less than a preset threshold; When the absolute error between the theoretical tree height H of a single tree and the measured tree height of a single tree is less than a preset threshold, the theoretical tree height H of a single tree and the corresponding breast diameter of the single tree are fitted with a tree height curve equation to obtain a tree height curve equation: H=a1+b1*log(D) Wherein, a1 and b1 are parameters of the tree height curve equation; When the absolute error between the theoretical tree height H of a single tree and the actual measured tree height of a single tree is greater than or equal to a preset threshold, the single tree diameter at breast height corresponding to the theoretical tree height H of a single tree is substituted into the tree height curve equation to calculate and update the corresponding theoretical tree height H of a single tree; The theoretical tree height H of a single tree whose absolute error with the actual tree height is less than a preset threshold and the theoretical tree height H of a single tree whose absolute error with the actual tree height is greater than or equal to the preset threshold and updated are combined to obtain the theoretical tree height of all single trees in the sample plot. The step of calculating the estimated information of a single tree according to the theoretical tree height and the breast diameter of the single tree specifically comprises the following steps: Based on the binary volume table, the measured volume of a single tree and the measured stock volume of the sample plot are calculated according to the measured information of the single tree; Substitute the measured timber volume of the single tree into the timber volume integral calculation formula to calculate the fitting parameter a of the timber volume integral calculation formula, and calculate the theoretical timber volume of all single trees and the theoretical volume of the sample plot according to the timber volume integral calculation formula. The timber volume integral calculation formula is: <h2 style=";text-align:left;direction:ltr">V=a2·LI<h2 style=";text-align:left;direction:ltr"> a3 Among them, V is the theoretical volume of a single tree, LI is the laser radar index, and H is T is the trunk height, a2 and a3 are fitting parameters, and D(H) is the trunk diameter when the height is H.

2. The forest parameter extraction method based on incomplete lidar point cloud data according to claim 1 is characterized in that: The preprocessing of the laser radar data point cloud specifically includes the following steps: Performing point cloud registration on the laser radar data point cloud, and cropping the laser radar data point cloud according to preset control points; Orient the laser radar data point cloud, perform voxel downsampling processing on the laser radar data point cloud at a set spatial resolution, and perform filtering processing on the laser radar data point cloud after the voxel downsampling processing; Ground points are separated from the laser radar data point cloud, and elevation normalization is performed on the laser radar data point cloud according to the heights of the separated ground points.

3. The forest parameter extraction method based on incomplete laser radar point cloud data according to claim 1 is characterized in that: The method of using a density clustering algorithm to segment the laser radar data point cloud into individual trees to obtain individual tree point clouds, and extracting individual tree parameter breast diameter according to the individual tree point clouds specifically includes the following steps: Using the density clustering DBSCN algorithm, the LiDAR data point cloud is clustered according to the preset layer thickness to screen out the point cloud of suspected tree trunks in each layer; The clustering parameter threshold is optimized according to the screening results and the original point cloud trunk of the sample plot, and the layer with the best clustering result is selected as the trunk layer for preservation, wherein the clustering parameter threshold includes the minimum number of point clouds and the search radius; Extracting the center position information of the trunk from the trunk point cloud of the trunk layer, and segmenting and clipping the trunk point cloud with a buffer determined by the center position of the trunk to obtain a candidate single tree point cloud; Taking the center position of the trunk as the center, the buffer zone is cylindrically sheared according to a set length as the shearing radius to extract a single tree point cloud; The DBH layer point cloud of the single tree point cloud is intercepted according to the preset interception height, and the DBH of the single tree is calculated using the cylindrical fitting algorithm. The calculation formula is: r 2 =(x-x0) 2 +(y-y0) 2 +(z-z0) 2 -[a(x-x0)+b(y-y0)+c(z-z0)] Among them, (a, b, c) is the cylinder axis vector, (x0, y0, z0) is a point on the cylinder axis, and r is the radius of the cylinder base.

4. The forest parameter extraction method based on incomplete lidar point cloud data according to any one of claims 1 to 3, characterized in that: The following steps are also included: The single tree measured information is matched with the single tree estimated information, and the accuracy of the single tree segmentation result is evaluated according to the matching result. The specific calculation formula is: Among them, P is the precision, R is the recall rate, F is the overall accuracy of misclassification and omission, TP is the number of matched trees corresponding to the estimated information of individual trees that can match the measured information of individual trees, FP is the number of individual trees corresponding to the estimated information of individual trees that were not measured but extracted, that is, the number of misdetected individual trees, and FN is the number of individual trees corresponding to the measured but not extracted estimated information of individual trees, that is, the number of missed individual trees; Calculate the determination coefficient R according to the single tree measured information and the single tree estimated information 2 And the root mean square error RMSE, calculated as: in, is the estimated parameter value of single tree i, y i is the measured value of the parameter of single tree i, is the arithmetic mean of the measured values.

5. A forest parameter extraction system based on incomplete lidar point cloud data, characterized in that: It includes a scanning processing module, a segmentation and extraction module, a tree height extraction module and an estimation and calculation module; The scanning and processing module is used to perform a laser radar scan on a preset sample plot, obtain a laser radar data point cloud, and pre-process the laser radar data point cloud; The segmentation and extraction module is used to perform single-tree segmentation on the laser radar data point cloud using a density clustering algorithm to obtain a single-tree point cloud, and extract the single-tree breast diameter according to the single-tree point cloud; The tree height extraction module is used to extract the theoretical tree height of a single tree according to the single tree point cloud; The estimation calculation module is used to calculate the single tree estimation information according to the single tree theoretical tree height and the single tree breast diameter; The specific implementation of the tree height extraction module for extracting the theoretical tree height of a single tree according to the single tree point cloud is as follows: A region growing algorithm is used to extract the single tree point cloud to obtain a single tree trunk point cloud; The Kunz-Gan curve equation is constructed with the center line of the trunk as the X-axis direction, the radius of the trunk cross section as the Y-axis direction, and the treetop as the origin: Among them, y i represents the radius of the cross section of the i-th layer of the trunk, the theoretical height of a single tree is H, and the height of the i-th cross section from the ground is h i , P is the coefficient to be determined; The radius y of each cross section in the single tree trunk point cloud i and height from ground h i Substitute it into the equation of the Kunz trunk curve, fit the optimal trunk curve, and calculate the theoretical tree height H of a single tree; The tree height extraction module extracts the theoretical tree height of a single tree according to the single tree point cloud and is also used for: Conducting actual measurements on a preset sample plot to obtain actual measurement information of individual trees in the sample plot, wherein the actual measurement information of individual trees at least includes the name of the individual tree species, the actual measurement coordinates of the individual tree, the actual measurement breast diameter of the individual tree, and the actual measurement height of the individual tree; Compare the theoretical tree height H of the single tree with the actual measured tree height of the single tree, and determine whether the absolute error between the two is less than a preset threshold; When the absolute error between the theoretical tree height H of a single tree and the measured tree height of a single tree is less than a preset threshold, the theoretical tree height H of a single tree and the corresponding breast diameter of the single tree are fitted with a tree height curve equation to obtain a tree height curve equation: H=a1+b1*log(D) Wherein, a1 and b1 are parameters of the tree height curve equation; When the absolute error between the theoretical tree height H of a single tree and the actual measured tree height of a single tree is greater than or equal to a preset threshold, the single tree diameter at breast height corresponding to the theoretical tree height H of a single tree is substituted into the tree height curve equation to calculate and update the corresponding theoretical tree height H of a single tree; Merge the theoretical tree height H of a single tree whose absolute error with the actual measured tree height of a single tree is less than a preset threshold and the theoretical tree height H of a single tree whose absolute error with the actual measured tree height of a single tree is greater than or equal to the preset threshold and updated, to obtain the theoretical tree height of a single tree for all the single trees in the sample plot; The specific implementation of the estimation calculation module to obtain the single tree estimation information based on the single tree theoretical tree height and the single tree breast diameter is as follows: Based on the binary volume table, the measured volume of a single tree and the measured stock volume of the sample plot are calculated according to the measured information of the single tree; Substitute the measured timber volume of the single tree into the timber volume integral calculation formula to calculate the fitting parameter a of the timber volume integral calculation formula, and calculate the theoretical timber volume of all single trees and the theoretical volume of the sample plot according to the timber volume integral calculation formula. The timber volume integral calculation formula is: <h2 style=";text-align:left;direction:ltr">V=a2·LI<h2 style=";text-align:left;direction:ltr"> a3 Among them, V is the theoretical volume of a single tree, LI is the laser radar index, and H is T is the trunk height, a2 and a3 are fitting parameters, and D(H) is the trunk diameter when the height is H.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the forest parameter extraction method based on incomplete lidar point cloud data described in any one of claims 1 to 4 is implemented.

7. A forest parameter extraction device based on incomplete lidar point cloud data, characterized in that: It includes a communication interface, a memory, a communication bus and a processor, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the steps of the forest parameter extraction method based on incomplete lidar point cloud data as described in any one of claims 1 to 4 when executing the program stored in the memory.

Images