A LiDAR-based non-destructive method for estimating the volume of standing timber

Abstract

This invention belongs to the field of geographic information technology, specifically relating to a non-destructive estimation method for standing timber volume based on LiDAR. The method includes: acquiring three-dimensional information of a portion of sample trees to obtain point cloud data; manually measuring a portion of the sample trees after pruning to obtain manual measurement parameters; preprocessing the point cloud data; calculating the volume of the base segment of the sample trees based on the preprocessed point cloud data; calculating shape parameters based on the manual measurement parameters; calculating the volume of the upper segment of the sample trees using the BCTEM model based on the shape parameters and point cloud data; calculating the sum of the base segment volume and the upper segment volume as the standing timber volume of the sample trees; calculating the tree height and diameter at breast height (DBH) of the sample trees; constructing a standing timber volume equation based on the standing timber volume, DBH, and tree height of the sample trees; and using the standing timber volume equation to achieve non-destructive estimation of the standing timber volume of trees similar to the sample trees. This invention solves the problem of inaccurate or impossible trunk measurement due to canopy shading and has good application prospects.

Description

Technical Field

[0001] This invention belongs to the field of geographic information technology, specifically relating to a non-destructive estimation method for standing timber volume based on LiDAR. Background Technology

[0002] Obtaining standing timber volume with high accuracy and low cost is a crucial issue in forestry research and application; calculation using standing timber volume models is an important method for obtaining this data. A standing timber volume model, also known as a standing timber volume formula, is a regression model established using statistical analysis principles and methods, with standing timber volume as the dependent variable and independent variables such as diameter at breast height (DBH), tree height, and trunk shape. Standing timber volume models are fundamental to the measurement of forest tree biomass and carbon storage, and are an important component of forest carbon sink monitoring. Currently, the construction of standing timber volume equations generally involves obtaining DBH, tree height, and shape number data through logging and analysis, which is highly destructive. However, for some rare tree species, such as the Chinese cypress (Thuja orientalis), which is a very small and endemic population in Chongqing, its importance dictates that large-scale logging and analysis of volume data cannot be used to obtain a large amount of volume data, thus making traditional methods unsuitable for constructing volume equations. In recent years, LiDAR technology, by emitting laser pulses and analyzing the echo signals, can efficiently acquire three-dimensional point cloud data of forest trees, achieving high-precision extraction of individual tree segmentation, DBH, and tree height parameters. The "segmented calculation method" based on lidar models tree trunks as approximate cylinders by segmenting individual tree point clouds and accumulating the volume segment by segment. This method can obtain volume data without felling, overcoming the limitations of tree species and region, and providing a new approach for calculating the volume and carbon storage of mixed forests and non-standard tree species. However, lidar is often unable to obtain trunk information in the obscured areas when acquiring tree parameters, especially for some pyramidal trees where the exposed trunk accounts for less than 30% of the total tree height. LiDAR cannot penetrate the pyramidal canopy, resulting in limited trunk information and often failing to obtain diameter at breast height (DBH) data. Therefore, it is not possible to obtain sufficient individual tree information to construct volume equations.

[0003] In summary, there is an urgent need for a new method for measuring standing timber volume, so as to achieve the non-destructive construction of volume equations without felling analytical trees, while solving the problem of missing information on individual trees caused by canopy shading. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a non-destructive estimation method for standing timber volume based on LiDAR, which includes:

[0005] S1: Use a LiDAR scanner to acquire the 3D information of the sample tree to obtain point cloud data; divide the sample tree into a base segment and an upper segment based on the point cloud data; extract a portion of the sample tree for manual measurement to obtain manual measurement parameters;

[0006] S2: Preprocess the point cloud data to obtain preprocessed point cloud data;

[0007] S3: Calculate the volume of the base segment of the sample tree based on the preprocessed point cloud data;

[0008] S4: Calculate the shape parameters based on manually measured parameters; calculate the upper section volume of the sample tree using the BCTEM model based on the shape parameters and point cloud data; calculate the sum of the base section volume and the upper section volume as the standing timber volume of the sample tree.

[0009] S5: Calculate the tree height and diameter at breast height of the sample tree based on the preprocessed point cloud data;

[0010] S6: Construct a standing timber volume equation based on the standing timber volume, diameter at breast height (DBH), and tree height of the sample trees; use the standing timber volume equation to achieve non-destructive estimation of the standing timber volume of trees of the same type as the sample trees.

[0011] Preferably, the preprocessing of point cloud data includes:

[0012] Point cloud data is subjected to point cloud resampling, point cloud denoising, and ground point filtering to obtain primary point cloud data;

[0013] After performing point cloud normalization on the primary point cloud data, a deep learning algorithm is used to segment individual trees to obtain preprocessed point cloud data.

[0014] Furthermore, the point cloud normalization process includes subtracting the sample tree point cloud from the primary point cloud data from the elevation values ​​generated by the ground points or digital elevation model to obtain the normalized primary point cloud data.

[0015] Preferably, the formula for calculating the volume of the base segment of the sample tree is:

[0016]

[0017] in, This represents the volume of the base segment of the sample tree. Indicates the number of slices at the measurement points. Indicates the first The diameter of the slice at each measurement point. Indicates the first The thickness of the slice at each measurement point.

[0018] Preferably, the formula for calculating the shape parameters is:

[0019]

[0020] in, and These are the first and second shape parameters, respectively. Indicates the weighting coefficient. Indicates the number of measurement points in the upper section. Indicates the number of measurement points in the base section. Indicates the diameter at the base of the tree. Indicates tree height, This represents the measured diameter of the slice at the j-th measurement point of the base segment. This represents the measured diameter of the slice at the k-th measurement point in the upper segment. express The corresponding measured height, express The corresponding measured height.

[0021] Preferably, the volume of the upper section of the sample tree calculated using the BCTEM model is expressed as follows:

[0022]

[0023] in, This represents the volume of the upper section of the sample tree. Indicates tree height, Indicates the diameter at the top of the base segment. Indicates the height of the top of the base segment; and These are the first and second shape parameters, respectively. It is relative height.

[0024] Preferably, the process of calculating the tree height and diameter at breast height of the sample tree includes:

[0025] Obtain the tree height from the preprocessed point cloud data;

[0026] If the height of the base section is not less than 1.3 meters, the diameter at 1.3 meters is directly obtained from the preprocessed point cloud data as the diameter at breast height.

[0027] If the height of the base section is less than 1.3 meters, the diameter at 1.3 meters is calculated based on the BCTEM model as the diameter at breast height.

[0028] Furthermore, the formula for calculating the diameter at 1.3 meters as the breast diameter based on the BCTEM model is as follows:

[0029]

[0030] in, This indicates the diameter at 1.3 meters. Indicates tree height, Indicates the diameter at the top of the base segment. Indicates the height of the top of the base segment; and These are the first and second shape parameters, respectively.

[0031] The beneficial effects of this invention are as follows:

[0032] This invention, based on lidar technology, uses high-precision point cloud data and single-tree segmentation technology to automatically extract parameters such as diameter at breast height (DBH) and tree height. Compared with traditional manual measurement of each tree, it can quickly obtain a large amount of information about each tree, significantly improving work efficiency and reducing labor costs.

[0033] This invention, based on lidar, can quickly calculate the volume of a large number of individual tree trunks through segmented volume calculation, achieving non-destructive measurement, reducing the cost of constructing volume equations, and solving the problem of not being able to cut down rare tree species;

[0034] This invention effectively combines the advantages of precise lidar measurement and model extrapolation. By using the Base Constrained Taper Extrapolation Model (BCTEM), it can calculate the trunk volume in the tree shading area, solving the problem of inaccurate or impossible trunk measurement caused by canopy shading, and providing an innovative technical solution for forestry resource surveys. Attached Figure Description

[0035] Figure 1 This is a flowchart of the non-destructive estimation method for standing timber volume based on LiDAR in this invention;

[0036] Figure 2 This is a schematic diagram of the preprocessed point cloud data in this invention;

[0037] Figure 3 This is a schematic diagram of tree height measurement in this invention;

[0038] Figure 4 This is a schematic diagram illustrating the principle of the lidar point cloud diameter parameter extraction method in this invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] This invention proposes a non-destructive method for estimating the volume of standing timber based on LiDAR, such as... Figure 1 As shown, the method includes the following:

[0041] S1: Use a LiDAR scanner to acquire the three-dimensional information of the sample tree to obtain point cloud data; divide the sample tree into a base segment and an upper segment based on the point cloud data; extract a portion of the sample tree for manual measurement to obtain manual measurement parameters.

[0042] First, data collection is conducted. Specifically, after surveying the overall condition of the sample plot, the LiDAR scanning route is determined. The operator holds the handheld LiDAR scanner at a height of approximately 1.5 meters above chest level and moves at a constant speed along an "S"-shaped path to collect data. This route ensures that the scanner can effectively cover and record the three-dimensional information of every tree in the sample plot, obtaining point cloud data. It should be noted that holding the handheld LiDAR device at chest level, with the device approximately 1.5 meters above the ground, allows the arms to bend naturally and lie close to the body, forming a stable triangular support. This minimizes body sway and arm fatigue, ensuring relative stability of the device during prolonged scanning. Holding it too high or too low will quickly lead to muscle fatigue, introduce unnecessary shaking, and reduce point cloud quality.

[0043] Based on radar scanning data, specifically the point cloud data obtained from the scan, the sample tree is divided into a base segment and an upper segment. The base segment is the trunk without any branches obstructing the view, and the remaining portion is the upper segment. A portion of the sample trees is then manually measured to obtain the manually measured parameters; specifically:

[0044] For tree diameter at breast height (DBH), when measuring DBH manually (using a DBH ruler), for trees on flat ground, the trunk diameter is measured directly at a point 1.3 meters above the ground; for trees on slopes, the measurement is taken 1.3 meters down from the top of the slope (keeping the trunk perpendicular) to avoid measurement errors caused by the slope. Three readings are taken at each location, and the average value is used as the verification data.

[0045] Tree height was measured manually using a tree height gauge. The gauge was held perpendicular to the ground, and the height from the base of the tree to the top was measured. The average of three readings was used as the verification value. The accuracy of the tree height extraction was then verified in conjunction with field measurement data.

[0046] To obtain the shape characteristics of trees, relevant parameters need to be constructed using measured data. Representative trees with similar shapes are selected, branches are removed, and only the trunk is retained. The trunk diameter is then measured at fixed intervals using a diameter at breast height (DBH) measuring tape. For rare tree species that cannot be felled, the tree diameter at different heights can be obtained by removing or bending some branches.

[0047] S2: Preprocess the point cloud data to obtain preprocessed point cloud data.

[0048] The point cloud data is resampled, denoised, and ground point filtered to obtain primary point cloud data; specifically:

[0049] Point cloud resampling:

[0050] The main purpose of point cloud resampling is to thin out the point cloud of a handheld LiDAR, thereby improving the efficiency of algorithms such as single-point separation and parameter extraction, and reducing the impact of data acquisition on point cloud density. This study uses a minimum point spacing method for point cloud resampling. The principle is to set a minimum spacing D between points, in meters, such that the distance between points in the point cloud in two-dimensional space cannot be less than this spacing. The algorithm is as follows: Assume there are N points in the point cloud. Select the i-th point as the target point (initially 0). Calculate whether the distance d between the target point i and its surrounding test points satisfies d>D. Keep the test points that satisfy the condition and delete those that do not. Calculate whether the distance d between the (i+1)-th target point and its surrounding test points satisfies d≥D. Delete the test points that do not satisfy the condition, and so on, until i=N.

[0051] Point cloud noise reduction:

[0052] When scanning forest stands using a handheld LiDAR scanner, non-tree objects are inevitably scanned as well; this portion of the point cloud data is referred to as noise points. The sources of noise points mainly include: ① ground features that extend beyond the target area during data acquisition; ② points generated when people, birds, etc., cross the scanning area during the scanning process; ③ non-target points generated by the reflection or refraction of the sensor's laser pulse signal, such as sunlight beams. These noise points are initially manually marked and removed. However, some discrete and sparse noise points are not easily removed manually. Therefore, denoising algorithms are needed to denoise the point cloud data to better recreate the true scene of the forest stand.

[0053] In some preferred embodiments of the present invention, a spatial distribution-based algorithm is used for point cloud denoising. This algorithm first performs a neighborhood search for each point in the data, finds N points in a specified neighborhood and calculates the average distance d from each point to that point, and then calculates the standard deviation SD corresponding to the median distance Me of each average distance. If the average distance exceeds k times the standard deviation threshold D, then the point is considered a noise point.

[0054] Ground point filtering:

[0055] Handheld lidar point cloud data only contains point cloud location information, lacking attribute information such as vegetation points or ground points. Since handheld lidar data contains rich surface point cloud information, the main purpose of ground point filtering is to separate the terrain and low shrubs and grasses located in the understory of the forest stand, thereby obtaining detailed ground point and terrain information, laying the foundation for extracting forest stand height information.

[0056] In some preferred embodiments of the present invention, an improved progressively denser triangulation filtering algorithm is used for ground point filtering. This algorithm uses morphological windowing and a root mean square error (RMSE) determination method based on plateau fitting to obtain ground seed points. Furthermore, it densifies the triangulation downwards after construction, better preserving the steep terrain features of southern forest areas. The calculation process is as follows: ① Initial seed point selection: Ground points are selected as potential seed points using morphological operations. Points with large residual values ​​are removed using plane fitting, and the remaining ground points are used as seed points for progressive triangulation densification. ② Triangulation construction: A triangulation is constructed using the seed points to generate an initial triangulation model. ③ Densification process: All points to be classified are traversed. The initial triangulation is densified downwards, allowing some unclassified points below the terrain surface to be classified. The initial surface more closely approximates the real terrain before upward triangulation densification is performed to reflect the details of terrain undulation. This step is repeated until all points are classified.

[0057] The primary point cloud data is normalized before being segmented into individual trees using a deep learning algorithm, resulting in preprocessed point cloud data. Specifically:

[0058] Point cloud normalization:

[0059] Normalized point cloud (NPC) is point cloud data that excludes terrain. It is a crucial prerequisite for most forest parameter extraction algorithms and the foundation for individual tree segmentation. The basic idea is to subtract the elevation value of each point from the corresponding DEM elevation value based on the already classified ground points in the point cloud data to obtain a normalized point cloud dataset. Therefore, by subtracting the elevation values ​​of sample trees in the primary point cloud data from the ground points or those generated by the digital elevation model, the influence of terrain undulations on vegetation parameter extraction is reduced, ensuring that the tree height is relative to the ground. This yields the normalized primary point cloud data.

[0060] Single-log division:

[0061] Point cloud data is segmented into individual trees based on their trunks. Deep learning algorithms are used to extract and segment the tree trunks. The preprocessed point cloud data is shown below. Figure 2 As shown.

[0062] S3: Calculate the volume of the base segment of the sample tree based on the preprocessed point cloud data.

[0063] Because a tree trunk is not a single, uniformly thick stick, but a complex three-dimensional structure that tapers gradually from base to tip, a single measurement point (such as measuring only at a diameter at breast height of 1.3 meters) is completely insufficient to describe this three-dimensional shape variation. To accurately grasp its "tapering" pattern (i.e., the tapering pattern), multiple measurement points must be set at different heights on the trunk to capture its complete morphological information. The more and denser the measurement points, the more realistic and accurate the fitted trunk contour curve will be. Let the base height that the lidar can scan be... Divide the height into Each measurement point slice is a point cloud set with a certain thickness. Specifically, the setting of measurement points follows the principle of combining standard forestry measurement methods with practical operability: Base densification: At the base of the trunk (e.g., 0-2 meters), the trunk shape changes most rapidly and is a key area determining the model's accuracy. Therefore, this invention sets up denser measurement points (e.g., 0m, 0.25m, 0.5m, 1.0m, 1.3m, 1.5m, 2.0m). Upper section sparser: As height increases, the trunk's thinning trend slows down, so the number of measurement points can be appropriately reduced (e.g., 2.5m, 3.0m) until the top of the tree. Simultaneously, the measurement must include the standard diameter at breast height (DBH) measurement point (at 1.3m) in forestry to connect with traditional dendrometry. This setup ensures that data best reflecting the true morphology of the trunk is obtained with the highest efficiency and lowest cost; Base section volume... Calculations can be performed directly using point cloud data:

[0064]

[0065] in, This represents the volume of the base segment of the sample tree. Indicates the number of slices at the measurement points. Indicates the first The diameter of the slice at each measurement point. Indicates the first The thickness of the slice at each measurement point.

[0066] S4: Calculate the shape parameters based on manually measured parameters; calculate the upper segment volume of the sample tree using the BCTEM model based on the shape parameters and point cloud data; calculate the sum of the base segment volume and the upper segment volume as the standing timber volume of the sample tree.

[0067] The construction of the BCTEM model is divided into two core stages: (1) Model training stage (establishing general laws): The purpose is to find general mathematical laws that can describe the change of tree trunk shape with height. A batch of fully measured "sample tree" data is used (such as collecting trunk diameter data of a certain number of sample trees at different heights). Each "sample tree" provides accurate diameters from the base to the top of the tree at multiple positions. Then, through mathematical optimization algorithms, the measurement data (tree height, diameter at each position) of all these "sample trees" are substituted into the preset model formula and the values ​​of shape parameters α and β are repeatedly adjusted. When the diameter value calculated by the model best matches the actual measurement values ​​of all "sample trees", we obtain a set of optimal shape parameters (α, β) that represent the average taper law of trees. This set of parameters is the trained model. (2) Model application stage (applying laws for prediction): The purpose is to predict the volume of a new tree that has only been partially scanned. For a new tree, we only need two basic pieces of information that the lidar can scan: base diameter (D0) and tree height (H). Then, the new tree's D0 and H, along with the fixed shape parameters (α, β) trained in the first stage, are substituted into the same model formula. The model generates a complete trunk outline curve from the base to the top of the tree. Based on this virtual outline, which conforms to the general laws of trees, the volume of the entire tree can be calculated with high precision through integration.

[0068] Shape parameters in this invention and The improved nonlinear least squares method is used to determine the accuracy of the basal segment, which prioritizes the fitting accuracy of the basal segment. This part is calculated based on measured data of different tree heights and diameters.

[0069]

[0070] in, and These are the first and second shape parameters, respectively. Indicates the weighting coefficient ( (emphasizing the fitting accuracy of the base segment). Indicates the number of measurement points in the upper section. Indicates the number of measurement points in the base section. Diameter at the base of the tree (usually measured at 0 meters above the tree height). Indicates tree height, This represents the measured diameter of the slice at the j-th measurement point of the base segment. This represents the measured diameter of the slice at the k-th measurement point in the upper segment. express The corresponding measured height, express The corresponding measured height.

[0071] Weighting coefficient The determination of the base segment is an automated, data-driven optimization process, rather than being arbitrarily specified by humans. Its core logic and steps are as follows: (1) Core idea: Since the scanning of the base segment by the lidar is the most accurate and reliable, we must ensure that the model's prediction of the base segment is extremely accurate. It is used to tell the computer "to pay special attention to the fitting accuracy of the base data points" during model training. (2) Determination method (taking k-fold cross-validation as an example):

[0072] Step 1: Randomly divide all the "sample tree" data into k groups (e.g., 5 groups).

[0073] Step 2: Preset a candidate range Values ​​(e.g., 0.6, 0.65, 0.7, ..., 0.95).

[0074] Step 3: For each candidate For the value, perform the following operations:

[0075] a. Take turns using one set of data as a "temporary exam room" and the other four sets as "workbooks".

[0076] b. Using the "workbook" data and the current... The values ​​are used to train the model, resulting in temporary α and β.

[0077] c. Use this temporary model to predict the volume of the base section of trees in the "temporary examination room" and calculate the prediction error.

[0078] d. Repeat 5 times to get this result. The average prediction error under the given value.

[0079] Step 4: Compare all candidates The average prediction error corresponding to the value is used to select the λ value that minimizes the prediction error of the base segment volume, which is then taken as the final and optimal value. .

[0080] therefore, The value of is objectively determined by the data itself through rigorous statistical methods, with the aim of maximizing the reliability and prediction accuracy of the model.

[0081] Traditional nonlinear least squares methods aim to minimize Σ(measured value - model predicted value)², but they treat all measurement points (both base and top) equally, applying equal force. This can lead to the model sacrificing accuracy at the base to better fit the top. The improved nonlinear least squares method of this invention aims to minimize […]. ·Σ(Base measured value - Model predicted value)²+(1- )·Σ(Measured value at the top - Predicted value from the model)².

[0082] The improved nonlinear least squares method of this invention introduces weights. This invention assigns higher weights to the error terms at the base measurement points. This means that during optimization, the computer prioritizes reducing the error in the base data, ensuring the model's accuracy in critical areas. Furthermore, the improved nonlinear least squares method of this invention starts with a set of random initial values ​​for α and β, continuously attempting to fine-tune them and observing whether the weighted total error decreases. This process iterates until the optimal combination of α and β that minimizes the weighted total error is found. Essentially, the improved nonlinear least squares method of this invention incorporates a priority mechanism in the search for the optimal shape parameters (α, β), forcing the model to first satisfy the base segment that the lidar can accurately scan, thereby ensuring the reliability of the model's extrapolation.

[0083] Based on the BCTEM model, the upper segment ( To the top of the tree The diameter variation pattern of ) is as follows:

[0084]

[0085] in, Indicates height The diameter at that point The diameter of the top of the base segment ( (Location).

[0086] The volume of the upper section is calculated by integration:

[0087]

[0088] Substitute into the model expression:

[0089]

[0090] make ,but :

[0091]

[0092] in, This represents the volume of the upper section of the sample tree. It is relative height.

[0093] Solve The process is as follows:

[0094] The volume of the upper section was obtained during the derivation of this model. The exact integral expression:

[0095]

[0096] The core issue is this integral. It is a non-elementary integral. That is, it cannot be solved analytically (i.e., by a simple formula) using combinations of basic elementary functions (such as power functions, exponential functions, logarithmic functions, trigonometric functions, etc.). This type of integral is mathematically considered to have "no closed-form solutions".

[0097] Therefore, an accurate calculation cannot be directly obtained using a single formula. Therefore, we must turn to numerical methods to seek high-precision approximate solutions. Among the many numerical integration methods (such as the trapezoidal rule and Simpson's method), this invention prefers the Gauss-Legend integration method because of its extremely high algebraic accuracy and computational efficiency, and it is particularly suitable for calculating the integral of smooth functions (such as the exponential function in this model) over a fixed interval (such as [0,1]).

[0098] Its principle is briefly described as follows:

[0099] The core idea of ​​the Gauss-Legend integral is not to find the sum of the areas of rectangles by equally dividing the intervals, but to intelligently select some integration points. And assign optimal weights to these points. This allows us to approximate the integral value using a weighted sum of the function values ​​at these points, achieving the highest algebraic accuracy.

[0100]

[0101] Our integration interval is [0, 1], which requires a simple variable transformation. Map it to the standard interval [-1, 1]. After this transformation, we obtain the form used in the model:

[0102]

[0103] here: It is the number of integration points selected in advance (for example, n=5 or n=6 can achieve extremely high precision). It corresponds to the th interval [0, 1]. One Gauss-Legendal integration point, It is the first The weights corresponding to each integration point. and Is it related to a specific function? These are irrelevant constants that depend only on the number of integration points chosen. Its value can be found in mathematical tables or pre-calculated and stored by a standard numerical calculation library.

[0104] To meet the need for rapid, real-time volume estimation in lidar forestry surveys, this invention further optimizes the above numerical process.

[0105] Pre-computation: Before deploying the model, a sufficiently accurate [model / component] is pre-selected. Values ​​(e.g., n=5), and obtain them from authoritative mathematical tables or calculate them yourself. indivual and The value is stored as a constant in the program.

[0106] Establish a parameter lookup table: For the target tree species, its shape parameters have already been fitted during the training phase. and In the application phase, the user inputs the measurements taken by the lidar. and The calculation process was then simplified to:

[0107] Read the pre-stored data from memory. , and the tree species , .

[0108] Calculate a simple weighted sum: .

[0109] Substitute this sum into the formula for weighted sum. You can get .

[0110] The total standing timber volume of a tree is the sum of the volumes of the basal and upper sections:

[0111]

[0112] S5: Calculate the tree height and diameter at breast height of the sample tree based on the preprocessed point cloud data.

[0113] The preprocessed point cloud data is the point cloud data after the segmentation of a single standing tree. Based on the segmentation results of a single tree, the height and diameter at breast height of a single tree can be extracted.

[0114] Obtain the tree height from the preprocessed point cloud data; specifically:

[0115] Tree height H refers to the vertical distance from the ground to the top of the tree, that is, the relative height of the tree top. Figure 3 As shown. Based on the definition of tree height, a Cartesian coordinate system based on point clouds is established. In the point cloud data of a complete tree, the maximum Z-value in the Z-axis direction is defined as Zmax, and the minimum Z-value in the Z-axis direction is defined as Zmin. Then, the tree height H is defined by the following formula:

[0116]

[0117] The trunk diameter is the diameter of a section perpendicular to the trunk's main axis, while the diameter at breast height (DBH) is the trunk diameter at a vertical distance of 1.3 meters from the ground. For example... Figure 4 As shown, the diameter at breast height (DBH) parameter extraction based on LiDAR technology is carried out using a measurement point slice at 1.3m ± 0.1m combined with parameter extraction methods. In cases where the trunk is obscured by branches and leaves, preventing the LiDAR from measuring the diameter at 1.3m, extrapolation is performed based on the base trunk diameter data scanned by the LiDAR and an extrapolation model of base constraint taper. This invention uses a circle fitting method to extract the diameter of scanned trunks.

[0118] If the height of the tree trunk (base section) that the lidar can scan is not less than 1.3m, then the diameter at 1.3m is directly obtained from the preprocessed point cloud data as the diameter at breast height (DBH).

[0119]

[0120] When the height of the tree trunk that the lidar can scan, i.e. the height of the base section, is less than 1.3m, then based on the BCTEM model, the upper section ( To the top of the tree The diameter variation pattern of ) is as follows:

[0121]

[0122] Therefore, the diameter at 1.3m, which is the diameter at breast height, is:

[0123]

[0124] Parameter extraction using radar lasers has good accuracy, as demonstrated by the coefficient of determination R. 2 The root mean square error (RMSE) is used to evaluate the extraction accuracy of tree height and tree diameter. 2 The formula for calculating RMSE is as follows:

[0125]

[0126]

[0127] In the formula: Representative data on the cross-sectional diameter of the trunk or the tree height at different locations in the sample plot, obtained through actual measurements. This represents the radar laser extraction value. This is the average value of the measured values.

[0128] If the accuracy does not meet the preset requirements, the number of samples can be increased or optimized to meet the requirements.

[0129] S6: Construct a standing timber volume equation based on the standing timber volume, diameter at breast height (DBH), and tree height of the sample trees; use the standing timber volume equation to achieve non-destructive estimation of the standing timber volume of trees of the same type as the sample trees.

[0130] Based on the standing timber volume, diameter at breast height (DBH), and height of the sample trees, a bivariate equation in the form of a conventional nonlinear power function can be fitted using the nonlinear least squares method to obtain the standing timber volume equation.

[0131] Once the standing timber volume equation is obtained, it can be used to achieve non-destructive estimation of the standing timber volume of trees of the same type as the sample.

[0132] In summary, to address the technical challenges of existing lidar technology's inability to acquire complete trunk point clouds due to canopy occlusion, and the high costs and inability to harvest rare tree species using traditional methods, coupled with the lack of volume equations, this invention establishes the Base Constrained Taper Extrapolation Model (BCTEM). Based on lidar, it can quickly achieve high-precision whole-tree volume estimation under conditions of incomplete trunk information. This invention offers several advantages: improved accuracy: directly using precisely measured base volume from lidar as the basis reduces error accumulation; clear physical meaning: segmented modeling better reflects the biological characteristics of tree growth; greater adaptability: applicable to different scanning heights and occlusion conditions; computational robustness: direct measurement of the base segment provides strong constraints for the model; deep integration with remote sensing technology: this invention essentially achieves seamless integration of precise ground point cloud measurement and mathematical model extrapolation, representing an upgrade of lidar technology in forestry applications from "measurement" to "perception and prediction." This invention effectively combines the advantages of precise lidar measurement and model extrapolation, solving the problem of volume estimation under conditions of incomplete trunk information in complex forest stands, and providing an innovative technical solution for forestry resource surveys.

[0133] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A LiDAR-based non-destructive estimation method of standing timber volume, characterized by, The method comprises the following steps: S1: collecting three-dimensional information of a sample tree by using a LiDAR scanner to obtain point cloud data; and dividing the sample tree into a base section and an upper section according to the point cloud data; extracting part of the sample tree to obtain artificial measurement parameters through artificial measurement; S2: pre-processing the point cloud data to obtain pre-processed point cloud data; S3: calculating the volume of the base section of the sample tree according to the pre-processed point cloud data; S4: calculating a shape parameter according to the artificial measurement parameters; calculating the volume of the upper section of the sample tree by using a BCTEM model according to the shape parameter and the point cloud data; and calculating the volume of the sample tree by summing the volume of the base section and the volume of the upper section; the formula for calculating the shape parameter is: ； wherein, with are a first and second shape parameter, respectively, denotes a weight coefficient, denotes the number of upper section measurement points, denotes the number of base section measurement points, denotes the tree base diameter, denotes the tree height, denotes the measured diameter of the jth measurement point slice of the base section, denotes the measured diameter of the kth measurement point slice of the upper section, denotes the corresponding measurement height, denotes the corresponding measurement height; the formula for calculating the volume of the upper section of the sample tree by using the BCTEM model is: ； wherein, represents the upper section volume of the sample tree, represents the base section top diameter, represents the base section top height; is the relative height; S5: calculating the height and diameter of the sample tree according to the pre-processed point cloud data; S6: constructing a volume equation according to the volume of the sample tree, the diameter and the height of the sample tree; and realizing nondestructive estimation of the volume of the same kind of trees as the sample tree by using the volume equation.

2. The LiDAR-based non-destructive estimation method of standing tree volume according to claim 1, characterized in that, The pre-processing process of the point cloud data comprises: performing point cloud resampling, point cloud denoising and ground point filtering on the point cloud data to obtain primary point cloud data; performing point cloud normalization on the primary point cloud data and then using a deep learning algorithm to perform single tree segmentation to obtain pre-processed point cloud data.

3. The LiDAR-based non-destructive estimation method of standing tree volume according to claim 2, characterized in that, The point cloud normalization process comprises: subtracting the elevation value of the sample tree point cloud in the primary point cloud data from the elevation value generated by the ground point or the digital elevation model to obtain normalized primary point cloud data.

4. The non-destructive estimation method of standing timber volume based on LiDAR according to claim 1, characterized in that, The formula for calculating the volume of the base section of the sample tree is: ； wherein, represents a base segment volume of the sample tree, represents a number of measurement point slices, represents a diameter of the th measurement point slice, represents a thickness of the th measurement point slice.

5. The LiDAR-based non-destructive estimation method of standing tree volume according to claim 1, characterized in that, The process for calculating the height and diameter of the sample tree comprises: obtaining the height of the sample tree from the pre-processed point cloud data; if the height of the base section is not less than 1.3 meters, directly obtaining the diameter at 1.3 meters from the pre-processed point cloud data as the diameter at breast height; if the height of the base section is less than 1.3 meters, calculating the diameter at 1.3 meters as the diameter at breast height based on the BCTEM model.

6. The LiDAR-based non-destructive estimation method of standing timber volume according to claim 5, characterized in that, The formula for calculating the diameter at 1.3 meters as the diameter at breast height based on the BCTEM model is: ； wherein denotes the diameter at 1.3 meters, denotes the tree height, denotes the top diameter of the base section, denotes the top height of the base section; and are first and second shape parameters, respectively.

Images