Forest plot non-contact precision measurement method based on air-ground laser radar fusion

Abstract

The present application relates to the technical field of forest plot measurement, and particularly relates to a forest plot non-contact precision measurement method based on ground-air laser radar fusion, comprising the following steps: according to scanning data and preset forest aspect map data of a forest plot, boundary positioning and scanning of the forest plot are performed to obtain a three-dimensional terrain model; based on point cloud data in the three-dimensional terrain model, single tree segmentation and parameter extraction are performed to obtain single tree data in the forest plot, and then abnormal single trees are positioned according to the single tree data; after the measured data of the abnormal single trees and the abnormal single tree data corresponding to the abnormal single trees are fused and calibrated, the three-dimensional terrain model is optimized according to the fusion and calibration results to obtain an optimized three-dimensional terrain model; after each single tree data is checked through the optimized three-dimensional terrain model, a checking report is generated according to the checking data to realize high-precision identification and parameter extraction of abnormal single trees in the forest plot without relying on large-scale manual measurement.

Description

Technical Field

[0001] This invention relates to the field of forest plot measurement technology, specifically to a non-contact, precise measurement method for forest plots using ground-to-air lidar fusion. Background Technology

[0002] The measurement of forest plots is an important foundation for obtaining forest stand structure parameters, assessing forest resource status, and making scientific management decisions. The accuracy and efficiency of the measurement results directly affect the reliability of forest resource inventory and dynamic monitoring.

[0003] Currently, the main measurement methods are divided into two categories: traditional manual field measurement and airborne lidar measurement. Traditional methods rely on surveyors venturing deep into forest areas to conduct contact measurements of individual trees using tools such as altimeters and diameter gauges. While this method can obtain relatively reliable individual tree data, it suffers from high workload, low efficiency, poor adaptability to complex terrain and high-risk forest areas, and susceptibility to subjective factors. Airborne lidar technology, using UAVs or manned aircraft as platforms, can achieve large-scale, non-contact, rapid scanning and automatically extract stand parameters from point cloud data, significantly improving fieldwork efficiency. However, this method generally lacks accuracy in identifying and extracting parameters from structurally abnormal trees such as leaning trees, dead standing trees, and crushed trees during data processing. The generated 3D models exhibit significant deviations in describing the true morphology and spatial location of these trees, leading to systematic errors in the estimation of key survey data such as stand volume and biomass. Summary of the Invention

[0004] To address the technical challenge of achieving high-precision identification and parameter extraction of abnormal individual trees in forest plots without relying on large-scale manual measurements, this application provides a non-contact, precise measurement method for forest plots using ground-air lidar fusion.

[0005] The non-contact, precise measurement method for forest plots provided in this application, which integrates ground-air lidar, adopts the following technical solution: A non-contact, precise measurement method for forest plots using ground-to-air lidar fusion includes: Based on the scanning data of the forest plots and the pre-set forest stand map data, the boundaries of the forest plots are located and scanned to obtain a three-dimensional terrain model; Based on point cloud data in a 3D terrain model, individual tree segmentation and parameter extraction are performed to obtain individual tree data within the forest plot. Then, abnormal individual trees are located based on the individual tree data. After fusing and calibrating the measured data of abnormal trees and the corresponding abnormal tree data, the three-dimensional terrain model is optimized based on the fusion and calibration results to obtain the optimized three-dimensional terrain model. After verifying the data of each individual tree using the optimized 3D terrain model, a verification report is generated based on the verification data.

[0006] Furthermore, based on the scanning data of the forest plots and the pre-set forest stand map data, the steps for locating and scanning the boundaries of the forest plots to obtain a three-dimensional terrain model include: The point cloud of the UAV lidar was obtained based on the scanning data, and the boundary polygon of the forest plot was extracted based on the pre-set forest stand map data. Based on the point cloud of the UAV's lidar, the vertex positions of the boundary polygon are corrected to obtain the corrected boundary polygon; Point cloud is extracted within the corrected boundary polygon to obtain point cloud data, and a three-dimensional terrain model is constructed based on the point cloud data.

[0007] Furthermore, based on the UAV lidar point cloud, the vertex positions of the boundary polygon are corrected to obtain the corrected boundary polygon. The steps include: Based on the UAV lidar point cloud, the point cloud located within a preset buffer distance of the boundary polygon is extracted as the boundary point cloud; Calculate the spatial density of the boundary point cloud, and after identifying the density abrupt change lines in the boundary point cloud based on the spatial density, translate the vertices of the boundary polygon adjacent to the density abrupt change lines along the direction of the density abrupt change lines to obtain the initial vertex positions. Based on the initial vertex positions, we perform orientation consistency analysis on the boundary point cloud to obtain the principal direction of the boundary. Then, we align the initial vertex positions according to the principal direction of the boundary to obtain the corrected vertex positions. Establish the connection relationships between the corrected vertex positions to obtain the corrected boundary polygon.

[0008] Furthermore, based on the point cloud data in the 3D terrain model, individual tree segmentation and parameter extraction are performed to obtain individual tree data within the forest plot. The steps for locating abnormal individual trees based on the individual tree data include: Normalize the point cloud data in the 3D terrain model to obtain normalized point cloud data; Clustering and segmenting the normalized point cloud data yields multiple individual point cloud clusters; For each individual tree cluster, parameter calculations are performed to obtain individual tree data including tree height, diameter at breast height, crown width, growth taper, and vertical structure. After integrating each individual tree data into a parameter vector, the similarity between the parameter vectors is calculated, and the individual trees corresponding to the individual tree data with a similarity lower than the preset similarity are identified as abnormal individual trees.

[0009] Furthermore, the steps for fusing and calibrating based on the measured data of abnormal individual trees and the corresponding abnormal individual tree data include: Spatial matching and temporal synchronization alignment of measured data and abnormal single tree data with the same parameters are performed to generate a comparison dataset; The differences of the same parameters in the comparison dataset are calculated item by item to obtain a set of parameter deviations including differences in tree height, diameter at breast height, and crown width. Based on the statistical results obtained from the statistical analysis of the parameter deviation set, the deviation patterns are identified, a correction function is established for each deviation pattern, and the correction function is associated and bound with the corresponding abnormal individual to generate the fusion calibration result.

[0010] Furthermore, the steps for optimizing the 3D terrain model based on the fusion calibration results to obtain the optimized 3D terrain model include: After extracting the anomalous tree point cloud clusters associated with the anomalous trees from the 3D terrain model, the correction function bound to the anomalous trees in the fusion calibration results is applied to the corresponding anomalous tree point cloud clusters to perform functional correction on the 3D coordinates and geometric features of each point in the corresponding anomalous tree point cloud clusters, generating a corrected point cloud cluster. The corrected point cloud clusters are spatially fused with other single-tree point cloud clusters in the 3D terrain model, excluding anomalous single-tree point cloud clusters, to obtain the fused point cloud. Based on the fused point cloud, the 3D terrain model is optimized to obtain an optimized 3D terrain model.

[0011] Furthermore, based on the fused point cloud, the steps to optimize the 3D terrain model to obtain the optimized 3D terrain model include: After separating the ground point cloud from the fused point cloud, a digital ground model is generated based on the ground point cloud, and the corrected terrain surface is obtained based on the digital ground model. Based on the corrected terrain surface, the height values ​​reflected by each point in the fused point cloud are normalized to obtain the terrain-normalized point cloud. Based on the terrain-normalized point cloud, the inflection point coordinates included in the 3D terrain model are subjected to elevation correction and planar position verification to obtain optimized inflection point coordinates. By combining the normalized point cloud of the terrain with the optimized inflection point coordinates, an optimized 3D terrain model is obtained.

[0012] Beneficial effects achieved: This application provides a non-contact, precise measurement method for forest plots using ground-to-air lidar fusion, comprising: locating and scanning the boundaries of the forest plots based on scanning data and pre-set forest stand map data to obtain a three-dimensional terrain model; segmenting individual trees and extracting parameters based on point cloud data in the three-dimensional terrain model to obtain individual tree data within the forest plots, and then locating abnormal individual trees based on the individual tree data; performing fusion calibration based on the measured data of abnormal individual trees and the corresponding abnormal individual tree data, and then optimizing the three-dimensional terrain model based on the fusion calibration results to obtain an optimized three-dimensional terrain model; and finally, using the optimized three-dimensional terrain model, verifying the data of each individual tree and generating a verification report based on the verification data.

[0013] In this application, the initial step of "locating and scanning the boundaries of forest plots based on scanning data and pre-set forest stand map data to obtain a three-dimensional terrain model" achieves contactless and precise digitization of the spatial range and overall three-dimensional structure of forest plots, providing a reliable data foundation for subsequent analysis. Then, "based on point cloud data in the three-dimensional terrain model, individual tree segmentation and parameter extraction are performed to obtain individual tree data within the forest plot, and then anomalous individual trees are located based on this data." This step identifies anomalous individual trees whose parameter performance differs from that of conventional individual trees from the point cloud data, thereby filtering out anomalous individual trees requiring high-precision processing from a massive dataset, significantly narrowing the target area that must rely on ground verification. Subsequently, "after fusing and calibrating the measured data of anomalous individual trees and their corresponding anomalous individual tree data, the three-dimensional terrain model is optimized based on the fusion calibration results to obtain an optimized model." The core of the "optimized 3D terrain model" step lies in conducting targeted ground-based lidar measurements only on the identified anomalous trees, fusing and calibrating the measured data with anomalous tree data extracted by UAVs, and using the calibration results to optimize the 3D terrain model. This fundamentally corrects the morphological and positional deviations of the anomalous trees in the 3D terrain model, enabling the optimized 3D terrain model to accurately represent such trees. Finally, "after checking the data of each tree through the optimized 3D terrain model, a check report is generated based on the check data." In other words, the reliability of parameters for all tree data extracted based on the optimized 3D terrain model, including the original anomalous trees, is guaranteed. This achieves high-precision identification and parameter extraction of forest plots, especially anomalous trees, without requiring large-scale manual measurements. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the steps of a non-contact, precise measurement method for forest plots using ground-air lidar fusion, as described in this application. Figure 2 A flowchart illustrating the steps to obtain a 3D terrain model; Figure 3 A flowchart illustrating the steps for obtaining individual tree data and locating abnormal individual trees within a forest plot; Figure 4 A flowchart illustrating the steps to obtain the optimized 3D terrain model. Detailed Implementation

[0015] The following combination Figures 1 to 4 This application will be described in further detail.

[0016] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0018] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0019] This application discloses a non-contact, precise measurement method for forest plots using ground-to-air lidar fusion.

[0020] Please refer to Figure 1 The non-contact, precise measurement method for forest plots proposed in this embodiment using ground-to-air lidar fusion includes steps S10-S40: Step S10: Based on the scanning data of the forest plots and the pre-set forest stand map data, the boundaries of the forest plots are located and scanned to obtain a three-dimensional terrain model.

[0021] By utilizing the scanning data with geographic coordinates obtained by UAV scanning, spatial registration and coordinate correction are performed on the sample plot boundaries that may have positional deviations in the pre-set forest stand map. This clearly defines the accurate spatial range for subsequent processing and constructs a three-dimensional terrain model within this range. This generates a digital base that simultaneously integrates high-precision geographic boundaries and real three-dimensional terrain undulations, providing a reliable and consistent spatial foundation for all subsequent tree identification, parameter extraction, and analysis calculations. This ensures that the entire measurement process is based on accurate geospatial information from the very beginning.

[0022] Among them, the scanning data of the forest plots are point cloud data that includes three-dimensional spatial information of terrain and vegetation, obtained by ground-to-air lidar; the pre-set forest stand map data is a digital forest distribution map that includes prior information such as the planning boundary of the forest plots and tree species composition.

[0023] Step S20: Based on the point cloud data in the 3D terrain model, perform individual tree segmentation and parameter extraction to obtain individual tree data within the forest plot, and then locate abnormal individual trees based on the individual tree data.

[0024] The function of identifying and quantifying the individual characteristics of each tree from the 3D terrain model is to decompose dense point cloud data into independent tree units (i.e., individual trees) and extract parameters to generate individual tree data including individual tree morphology and spatial information. Based on the point cloud data, abnormal individual trees that deviate significantly from the normal range in parameters such as tree height, crown width, or location can be automatically screened out. This enables the rapid location of abnormal individual trees in forest plots that need to be checked and treated without large-scale manual intervention, providing a clear data foundation for subsequent calibration and model optimization for these abnormal individual trees.

[0025] Step S30: After fusing and calibrating the measured data of the abnormal individual trees and the corresponding abnormal individual tree data, the three-dimensional terrain model is optimized based on the fusion calibration results to obtain the optimized three-dimensional terrain model.

[0026] This method solves the problem of accurately obtaining the true parameters of abnormal trees using only aerial scanning. By fusing and calibrating the measured data of abnormal trees obtained from ground surveys with the corresponding data extracted from the 3D terrain model, a parameter correction relationship is established between the two. Based on this parameter correction relationship, the point cloud and attributes of the abnormal trees and their surrounding areas in the 3D terrain model are optimized in a targeted manner. This allows the use of ground surveys of a small number of abnormal trees to drive and complete the iterative update and accuracy improvement of the entire 3D terrain model. The optimized model can more realistically reflect the actual state of all trees, including the abnormal trees, laying the foundation for the subsequent generation of highly reliable survey data.

[0027] Step S40: After verifying the data of each individual tree using the optimized 3D terrain model, a verification report is generated based on the verification data.

[0028] Based on a more accurate and reliable 3D terrain model after calibration and optimization, the final quantitative statistics of forest resources are completed. Using the high-precision point cloud data and individual tree parameters included in the optimized 3D terrain model that have corrected abnormal individual tree deviations, the spatial and attribute information of each individual tree in the forest plot is systematically extracted, verified, and summarized, thereby generating structured plot inventory data. The effect achieved is that a standardized inventory report that accurately reflects the true stand condition of the plot and can directly support forestry management decisions is produced. This ensures that while the entire measurement process significantly reduces the amount of ground work, the overall accuracy and reliability of the output results meet the requirements of practical applications.

[0029] Specifically, by calling and parsing the point cloud data and individual tree parameters of the trees contained in the optimized 3D terrain model, which have been corrected through the fusion calibration process in step S30, and based on the spatial location and morphological characteristics of each individual tree point cloud cluster in the optimized 3D terrain model, the corrected individual tree parameters, including the tree height, diameter at breast height, and crown width of each tree, are automatically re-extracted or directly called, thus completing the one-by-one investigation and data recording of all individual trees in the forest plot and obtaining the individual tree dataset. Next, based on the horizontal coordinates and crown width information of each tree in the individual tree dataset, the projection range of its crown on the horizontal plane is calculated. By superimposing the crown projections of all individual trees, the ratio of its total coverage area to the total area of ​​the forest plot is calculated, thereby automatically calculating the stand canopy closure. The diameter at breast height (DBH) and tree height of all trees in the individual tree dataset are summed and the arithmetic mean is calculated to directly generate the average DBH and average tree height at the stand level. At the same time, based on the tree species, DBH, and tree height parameters of each tree in the individual tree dataset, a preset binary volume formula matching the tree species is called. This binary volume formula expresses the mathematical relationship between volume and two independent variables: DBH and tree height. The measured DBH and tree height values ​​of the current individual tree are used as input variables and automatically substituted into this binary volume formula. The calculation is performed in the middle, where, For single timber volume, Diameter at breast height, For the height of the tree, , , To determine specific parameters for this tree species through fitting a large amount of sample data, this mathematical operation is performed to obtain the volume of a single tree. This process is then repeated for all trees in the single-tree dataset to calculate the total volume of each tree. This total volume is then divided by the forest plot area to calculate the volume per unit area. The single-tree dataset is categorized and counted according to tree species, automatically calculating the number of trees of each species and their percentage of the total. Based on the horizontal coordinates of each tree in the dataset, the Euclidean distance between it and the horizontal coordinates of all other trees in the forest plot is calculated. The minimum distance is identified and recorded as the nearest neighbor distance for that tree. The average nearest neighbor distance at the stand level is obtained by averaging the average nearest neighbor distances of all trees. This average nearest neighbor distance is then compared with a preset distance value. If the average nearest neighbor distance is less than a preset value, the average nearest neighbor distance is calculated. The preset distance value determines whether individual trees tend to cluster together; if it is greater than the preset distance value, it determines whether individual trees tend to disperse. Then, the calculated stand canopy closure, average diameter at breast height (DBH), average tree height, volume per unit area, tree species composition, and spatial distribution pattern are integrated into stand-level inventory data. Finally, the inventory data is filled into and generated into an inventory report that can be directly reported or used for further analysis, according to the preset forestry resource survey and planning system data interface specifications. The data source of the final inventory report is based on a three-dimensional terrain model that has been calibrated and optimized through ground and air data fusion, thereby ensuring the accuracy and reliability of each stand and individual tree parameter in the inventory report. In particular, the statistical results on abnormal individual trees and their contribution to the overall stand indicators have a credibility comparable to traditional detailed manual surveys, meeting the needs of high-precision forest resource surveys and monitoring.

[0030] The preset distance value is a threshold distance used to determine whether the spatial distribution pattern of trees tends to be clustered or dispersed.

[0031] In one feasible implementation, refer to Figure 2 As shown, step S10 may specifically include steps S11 to S13: Step S11: Obtain the UAV lidar point cloud based on the scanning data, and extract the boundary polygon of the forest plot based on the preset forest stand map data.

[0032] A drone equipped with a lidar, GPS, and inertial measurement unit (IMU) conducted a full-area aerial scan of the forest plot. The lidar recorded the laser ranging value corresponding to the time difference between the emission and reception of each laser pulse; the GPS receiver recorded the antenna phase center position at the corresponding moment; and the IMU recorded the carrier attitude angle at the corresponding moment. All data were timestamped. Subsequently, all data were synchronized based on the timestamps, and the local three-dimensional coordinates of each laser endpoint in the lidar scanner coordinate system were calculated based on the laser ranging value, scanning mirror angle, and installation calibration parameters. Next, based on the timestamp of the laser footpoint, a set of carrier attitude angles matching the timestamp is found from multiple sets of synchronously recorded carrier attitude angles. This set of measurements includes the yaw angle, pitch angle, and roll angle. Based on the geometric principle of coordinate transformation, the yaw angle, pitch angle, and roll angle are used as input parameters. The rotation matrices around the Z-axis, Y-axis, and X-axis of the carrier coordinate system are calculated sequentially and matrix multiplication is performed in turn. Alternatively, the elements of the rotation matrix determined by these three Euler angle parameters can be directly substituted into the calculation formula of the rotation matrix elements to generate a 3x3 rotation matrix. Each element of this rotation matrix is ​​calculated by combining the sine and cosine function values ​​of the carrier attitude angle.

[0033] The coordinates of the corresponding GPS antenna phase center in the geodetic coordinate system are obtained based on the timestamp of the laser footpoint. The offset vector from the antenna phase center to the center of the LiDAR scanner is calculated using the known mounting arm offset parameters. After rotating and correcting this offset vector in the carrier coordinate system according to the attitude angle at the corresponding moment, it is vector synthesized with the GPS position to obtain the position of the LiDAR scanner center in the geodetic coordinate system. The coordinates of this position are used as the components of the translation vector. Then, the coordinates of the laser footpoint in its local scanner coordinate system are rotated to the instantaneous carrier coordinate system using the rotation matrix derived from the corresponding carrier attitude angle. Finally, by adding the translation vector with the scanner center position as the endpoint, the transformation from the local scanner coordinate system to the instantaneous carrier coordinate system with the GPS position as the origin is completed. Since the origin and direction of the instantaneous carrier coordinate system are fixed to the geodetic coordinate system through the GPS position and the carrier attitude angle, the coordinates of the laser footpoint after the above rotation and translation are directly expressed in the geodetic coordinate system, thus completing the transformation from local coordinates to geodetic coordinates. Finally, the above transformation calculations are performed on all laser footpoints, and the geodetic coordinates of each point with latitude, longitude and elevation are output, thus obtaining the UAV lidar point cloud with three-dimensional geographic coordinates.

[0034] Meanwhile, based on the pre-stored forest stand map data containing forest plot planning information stored in digital vector form, the closed polygon vector elements of the corresponding plots are queried and extracted using geographic information system software according to the plot number or specific identifier in the attribute table. The node coordinate sequence of the closed polygon vector elements constitutes the boundary polygon of the forest plot. By simultaneously obtaining the UAV lidar point cloud reflecting the actual three-dimensional condition of the forest plot and the boundary polygon reflecting its theoretical planning range, a data foundation is provided for the subsequent accurate spatial matching and automated correction of the theoretical boundary with the actual scanning data.

[0035] Step S12: Based on the UAV lidar point cloud, correct the vertex positions of the boundary polygon to obtain the corrected boundary polygon.

[0036] The vertex positions of the boundary polygons are corrected based on UAV LiDAR point clouds. By utilizing the 3D spatial information provided by the UAV LiDAR point clouds, potential positioning errors or geometric distortions in the boundary polygons extracted from pre-set forest stand data are corrected. This ensures that the boundary polygons accurately correspond to the actual spatial range of the forest plots. The boundary positions are identified and verified by using the detailed terrain and land feature characteristics contained in the UAV LiDAR point clouds, thereby adjusting and optimizing the spatial positions of the vertices of the boundary polygons. This improves the geometric accuracy and spatial consistency of the boundary polygon definition, resulting in a corrected boundary polygon that better reflects the actual situation. This provides a reliable spatial framework for subsequent point cloud data extraction and the construction of high-precision 3D terrain models within the boundary polygons, thus ensuring the quality of the entire data processing workflow.

[0037] Furthermore, step S12 may include steps S121 to S124: Step S121: Based on the UAV lidar point cloud, extract the point cloud located within a preset buffer distance of the boundary polygon as the boundary point cloud.

[0038] It should be noted that the preset buffer distance is a fixed value predefined based on the possible positioning error range of the initial boundary polygon, used to create a strip-shaped buffer area around the initial boundary polygon.

[0039] For each point in the UAV LiDAR point cloud, spatial position determination is performed, and the shortest perpendicular distance from it to each edge of the boundary polygon is calculated. If the shortest perpendicular distance is less than or equal to the preset buffer distance, the point is determined to be within the buffer and extracted. The set of all extracted points constitutes the boundary point cloud, which provides a data basis for subsequent steps to identify the boundary polygon and correct the vertex position based on the spatial density of the point cloud.

[0040] Step S122: Calculate the spatial density of the boundary point cloud, and after identifying the density abrupt change line in the boundary point cloud based on the spatial density, translate the vertices of the boundary polygon adjacent to the density abrupt change line along the direction of the density abrupt change line to obtain the initial vertex position.

[0041] Extract the strip buffer region where the boundary point cloud is located, establish a series of equally spaced sampling profile lines along the direction parallel to the boundary polygon, and set a fixed statistical width on both sides of each sampling profile line. Calculate the number of boundary point clouds within the statistical width of each sampling profile line in turn, thereby obtaining the spatial density distribution sequence of the point cloud along the direction of the strip buffer region.

[0042] Next, the spatial density distribution sequence of the point cloud was analyzed to identify points where density values ​​changed drastically. Because the point cloud is denser inside the forest plot and sparser in the open areas outside, the density values ​​at these points exhibit a sharp decrease from high to low. Points with abrupt density changes identified on adjacent sampling profiles were connected to form a density abrupt change line. Subsequently, for each vertex on the boundary polygon, its nearest point on the density abrupt change line was found, and the direction vector from that vertex to its nearest point was calculated. This direction vector indicates the direction in which the vertex should be translated. The vertex was moved along this direction to the position of its corresponding nearest point, thus obtaining the corrected vertex position, i.e., the initial vertex position. By utilizing the natural variation characteristics of the boundary point cloud density, the position of the actual boundary polygon of the forest plot was identified, and the vertices of the initially planned boundary polygon were aligned with the vertices of the actual boundary polygon, improving the geometric accuracy of the boundary polygon in spatial location.

[0043] Step S123: Based on the initial vertex positions, perform orientation consistency analysis on the boundary point cloud to obtain the main boundary direction. Then, align the initial vertex positions according to the main boundary direction to obtain the corrected vertex positions.

[0044] After extracting the two-dimensional planar coordinates (usually east and north coordinates) of all points in the boundary point cloud, an n-row, 2-column coordinate matrix is ​​constructed from these coordinates, where n is the number of points, and the two columns represent the east and north coordinates of the points, respectively. The mean east and north coordinates of the center point are calculated, and the mean coordinates of each column are subtracted from the mean coordinates of each point to obtain an n-row, 2-column deviation matrix composed of the deviations of the coordinates of each point. The deviation matrix is ​​then transposed to obtain its transpose matrix, and matrix multiplication is performed between the transpose matrix and the deviation matrix. Finally, the result of the matrix multiplication is divided by (n-1) for unbiased estimation, resulting in a 2-row, 2-column covariance matrix. Subsequently, the characteristic equation of this covariance matrix is ​​constructed, which is to calculate the determinant of the covariance matrix minus the product of a specific variable and the identity matrix and set it equal to zero, resulting in a quadratic polynomial equation in terms of the specific variable. , among which, among which Represents eigenvalues, and These represent the eastward and northward coordinate variances on the diagonal of the covariance matrix, respectively. Let represent the eastward and northward coordinate covariances on the off-diagonal lines. Solving this quadratic equation yields the eigenvalues ​​and eigenvectors of the covariance matrix. For each eigenvalue, substitute it back into the homogeneous linear equation system corresponding to the characteristic equation. Solving this system yields a non-zero solution vector, which is the eigenvector corresponding to that eigenvalue. Finally, this eigenvector is usually normalized to a magnitude of 1, thus obtaining a unit eigenvector. The unit eigenvector corresponding to the largest eigenvalue defines the first principal component direction, which represents the axis with the largest variance in the spatial distribution of the boundary point cloud.

[0045] Then, the initial vertex positions are aligned according to the direction of the first principal component. Specifically, a reference line parallel to the principal direction of the boundary is constructed, and each initial vertex is vertically projected onto the reference line in sequence. The coordinates of the projection points are used as the new positions of the corresponding initial vertices. This results in the corrected vertex positions that are consistent with the principal direction of the boundary in direction and are still close to the density change line in space. This ensures that the sides of the boundary polygon are consistent with the overall outline of the forest plot, eliminating the irregularity and jagged undulations of the boundary line segments caused by local point cloud disturbances or preliminary translations.

[0046] Step S124: Establish the connection relationship between the corrected vertex positions to obtain the corrected boundary polygon.

[0047] According to the order of the corrected vertex positions in the original boundary polygon, adjacent corrected vertices are connected by straight line segments in turn, and the last corrected vertex is connected to the first corrected vertex, thus forming a closed planar geometric figure composed of a series of line segments connected end to end. This planar geometric figure is the corrected boundary polygon. In this way, the discrete vertex sequence after position correction is reconstructed into a complete, continuous and closed geometric boundary expression.

[0048] Step S13: Extract point cloud data within the corrected boundary polygon to obtain point cloud data, and construct a three-dimensional terrain model based on the point cloud data.

[0049] Based on the closed spatial range defined by the corrected boundary polygon, a spatial query is performed on the UAV lidar point cloud. By determining whether the planar coordinates of each laser foot point are located within the corrected boundary polygon, all points falling within the range of the corrected boundary polygon are selected, and the set of these points is used as the point cloud data within the forest plot area.

[0050] Subsequently, the point cloud data is divided into spatial grids, and the lowest point in each grid is found as the initial ground point. Then, a virtual cloth is placed over the highest point of the point cloud and discretized into regular grid nodes. Next, the physical process of the cloth settling downward under gravity and colliding with the inverted point cloud surface is simulated through iterative calculation. In each iteration, the position of the nodes is adjusted according to the elevation relationship between the cloth node and the corresponding point cloud point below, as well as the internal force constraints between adjacent nodes. When the position of the cloth node stabilizes and no longer changes significantly, the settling process stops. At this time, the elevation position corresponding to each cloth node is identified as the simulated ground elevation. Finally, by judging whether the elevation difference between each point in the point cloud data and the cloth node directly above it is less than a preset threshold, points with an elevation difference less than the preset threshold are classified as ground points, and the rest are classified as non-ground points. The preset threshold is a critical parameter for elevation difference used to distinguish between ground points and non-ground points.

[0051] The ground point cloud is treated as a set of data points on a two-dimensional plane. Starting with a triangle containing all data points, each data point is sequentially inserted into this triangle. During insertion, a triangle whose circumcircle contains the insertion point is first found, and the common edges of these triangles are deleted to form an insertion polygon. Then, the insertion point is connected to all vertices of the insertion polygon to form a new triangle. Subsequently, local optimization is performed on the newly generated triangle and its affected neighboring triangles. That is, these triangles are checked to see if they satisfy the Delaunay empty circumcircle criterion. If the quadrilateral formed by two adjacent triangles does not satisfy the criterion, their common diagonals are swapped. Through this iterative local optimization operation, the circumcircle of each triangle in the triangulation is ultimately made so that no other data points are contained inside, thus generating a triangulation network directly composed of ground points that satisfies the Delaunay criterion. This triangulation network is the three-dimensional terrain model, ensuring that the spatial extent of the point cloud data used for subsequent analysis such as single tree segmentation and parameter extraction is completely consistent with the actual boundary of the corrected forest plot. This eliminates the range error introduced by inaccurate boundaries and provides an absolutely reliable data foundation for building a high-precision three-dimensional terrain model and carrying out subsequent accurate measurements.

[0052] In one feasible implementation, refer to Figure 3 As shown, step S20 may specifically include steps S21 to S24: Step S21: Normalize the point cloud data in the 3D terrain model to obtain normalized point cloud data.

[0053] For each point in the point cloud data, interpolation is first performed on the 3D terrain model based on its planar coordinates to obtain the ground elevation value directly below that point. Then, the original elevation value of the point is subtracted from the obtained ground elevation value, and the difference is the normalized height of the point relative to the ground. This calculation is performed on all points in the point cloud data, thereby unifying the elevation reference of all points from absolute elevation to relative height relative to the ground directly below, generating normalized point cloud data. By eliminating the influence of terrain undulations on the vertical distribution of the point cloud, subsequent tree segmentation and parameter extraction algorithms can be performed on a unified horizontal reference plane, thus significantly improving the accuracy and reliability of the algorithm in calculating parameters such as tree height and canopy vertical structure.

[0054] The original elevation value refers to the vertical coordinate value of each point in the point cloud data relative to a certain absolute or original reference surface obtained during measurement.

[0055] Step S22: Cluster the normalized point cloud data to obtain multiple single point cloud clusters.

[0056] Normalized point cloud data is clustered and segmented using density-based spatial clustering algorithms, such as the DBSCAN algorithm. This algorithm defines the cluster formation conditions by setting two key parameters: neighborhood radius and minimum number of points.

[0057] First, an unvisited point is randomly selected from the normalized point cloud data as the core point. Using its location as the center and its neighborhood radius as the search range, the number of other points within this range is calculated. If the number reaches or exceeds the minimum number of points, these points are marked as belonging to the same cluster. This neighborhood search and expansion process is recursively repeated for each newly added point until no new points can be added to the cluster, thus forming a complete single-tree point cloud cluster. Points that do not meet the core point condition are temporarily marked as noise. After traversing all points, multiple independent and spatially separated point cloud clusters are output, each corresponding to a single-tree point cloud cluster. Based on the distribution density and proximity relationships of the point cloud in three-dimensional space, the entire normalized point cloud data is automatically divided into multiple individual units representing independent trees, laying a structured data foundation for subsequent accurate parameter calculations and feature analysis for each tree.

[0058] The minimum number of points is a parameter used in the DBSCAN clustering algorithm to define the core points. It refers to the minimum number of points that must be included in the neighborhood of a point. Its setting is based on the density distribution of normalized point cloud data and experience or prior knowledge of the size of a single point cloud cluster in practical applications.

[0059] Step S23: Calculate parameters for each individual tree cluster to obtain individual tree data including tree height, diameter at breast height, crown width, growth taper, and vertical structure.

[0060] For each individual tree point cloud cluster obtained from the segmentation, its tree height is first calculated by finding the maximum normalized height value of all points in the cluster. Next, the diameter at breast height (DBH) is calculated. Typically, in the trunk point cloud portion, a horizontal slice of point cloud at a certain standard height above the ground is extracted. The coordinates of all points in this horizontal slice are projected onto a horizontal plane to obtain a two-dimensional coordinate point set. Then, an initial estimate of the center coordinates and radius is initialized, and the iterative optimization phase begins. In each iteration, the Euclidean distance from all two-dimensional points to the currently estimated center is calculated, and these Euclidean distances are compared with the currently initialized... The radius values ​​are compared and the difference is calculated. The sum of the squares of all differences is defined as the error function of the current fit. The center coordinates and radius values ​​are adjusted to reduce this error function, that is, the horizontal and vertical components of the center coordinates and the radius value are updated along the direction of error reduction. This iterative process is repeated until the change in the error function is less than a preset threshold or the maximum number of iterations is reached. At this time, the iteration stops. The obtained center coordinates and radius values ​​are the fitted circle. Multiplying the radius of the fitted circle by two gives the diameter of the fitted circle, which is the diameter at breast height (DBH). Next, the crown width is calculated by projecting the point cloud representing the crown onto the horizontal plane to construct... The convex hull polygon is then used to determine the maximum span of the convex hull polygon in the east-west and north-south directions, which are then used as the major and minor axes of the crown width. Next, the growth taper is calculated by analyzing the diameter variation trend of the trunk point cloud at different heights, i.e., quantifying it by calculating the difference or ratio of the diameter at breast height (DBH) to the tree height at a certain ratio. Finally, based on the determined tree height range of each individual tree point cloud cluster, i.e., the height value from ground level zero to the highest point of the point cloud, this tree height range is divided into several equally spaced height layers, for example, each layer is divided at intervals of 0.5 meters or 1 meter. For each of the divided height layers, the falling weight is statistically analyzed. The number of all point clouds within the three-dimensional spatial boundary of this layer reflects the point cloud density of branches and leaves within this layer. By sequentially calculating the point cloud density and distribution characteristics of all height layers, a quantitative profile describing the continuous change of the branch and leaf space occupancy of the single tree point cloud cluster from the ground to the treetop on the vertical gradient can be generated. This accurately describes the configuration of branches and leaves in the vertical space, thereby transforming each single tree point cloud cluster from a set of points in three-dimensional space into a set of tree morphological and structural parameters, thus providing data input for subsequent similarity comparison and abnormal single tree identification based on parameter vectors.

[0061] Step S24: After integrating each individual tree data into a parameter vector, calculate the similarity between the vectors and locate the individual trees corresponding to the individual tree data with a similarity lower than the preset similarity as abnormal individual trees.

[0062] Multiple parameters calculated from the point cloud clusters of each individual tree, including tree height, diameter at breast height (DBH), crown width, growth taper, and parameters describing vertical structure, are combined into a multidimensional numerical sequence according to a predefined uniform order. This multidimensional numerical sequence constitutes a parameter vector representing the comprehensive morphological characteristics of the corresponding individual tree. Next, the similarity between the parameter vectors of all individual trees is calculated. For any two individual tree parameter vectors, when calculating the Euclidean distance, the square of the difference between the feature values ​​at each corresponding position in the two vectors is first calculated. Then, the squares of all differences are summed, and finally, the square root of the sum is taken. The resulting value is the Euclidean distance. The smaller the Euclidean distance, the smaller the absolute numerical difference in morphological characteristics between the two individual trees, i.e., the higher the overall similarity. This is used to assess the overall similarity between any two individual trees in terms of morphological characteristics. Then, by analyzing the pairwise similarity results of all individual trees, a global similarity threshold is set for the entire forest plot as a preset similarity. The specific threshold can be determined based on the statistical distribution characteristics of the similarity values ​​(such as the mean minus a certain number of standard deviations) or empirical values. Finally, during the traversal comparison, if the average similarity between the parameter vector of a certain tree and the parameter vectors of most other trees in the forest plot is lower than the preset similarity, then the tree is determined to have a significant difference in morphology from other trees, and is thus positioned as an abnormal tree. By objectively comparing and identifying abnormal trees in terms of growth status, health status, or point cloud data quality based on quantified multidimensional morphological feature parameters, this provides guidance for subsequent ground verification and data calibration of abnormal trees, realizing the transition from full-area measurement to precise verification.

[0063] In one feasible implementation, refer to Figure 4 As shown, step S30 may specifically include steps S31 to S36: Step S31: Perform spatial matching and time synchronization alignment of the measured data and abnormal single tree data with the same parameters to generate a comparison dataset.

[0064] When acquiring measured data obtained by scanning and parameter calculation of abnormal trees using a ground-based handheld lidar, and extracting abnormal tree data from the tree point cloud clusters corresponding to the 3D terrain model, spatial matching with the same parameters is performed based on identifiable feature points of the same name in the two sets of data (such as significant branching points, treetop apex, and trunk base center point). The local measurement coordinate system where the measured data is located is registered with the geodetic coordinate system where the abnormal tree data is located to ensure that the two describe the same tree's spatial posture and position in a consistent manner.

[0065] When performing time synchronization alignment, since the two sets of data may be collected at different times, it is necessary to compensate for the systematic offset of parameters that may be caused by tree growth or phenological changes based on the timestamps in the data collection log. This is usually done by assuming that changes within a short time interval are negligible, thereby achieving consistency of parameter comparison on the time base.

[0066] After completing the spatial matching and temporal synchronization alignment, the two sets of parameter data that have been registered for each abnormal tree are paired and combined to generate a comparison dataset that can be directly compared item by item. This lays the data foundation for subsequent calculation of parameter deviations and analysis of their patterns. It ensures that each set of data being compared strictly corresponds to the same tree, the same set of features, and the same comparable time state in the real world. This allows the calculated deviation to truly reflect the difference between the automatic extraction by the UAV and the ground-based true value measurement, rather than the error introduced by coordinate misalignment or time asynchrony.

[0067] Step S32: Calculate the differences of each parameter in the comparison dataset to obtain a set of parameter deviations, including differences in tree height, diameter at breast height, and crown width.

[0068] Based on the generated comparison dataset, for each abnormal tree, arithmetic subtraction operations are performed on the corresponding parameters of its paired measured data and abnormal tree data. Specifically, the tree height difference is obtained by subtracting the tree height extracted from the abnormal tree data from the tree height value recorded in the measured data; the diameter at breast height (DBH) difference is obtained by subtracting the DBH value extracted from the abnormal tree data from the measured DBH value; and the crown width difference is obtained by subtracting the crown width value extracted from the abnormal tree data from the measured crown width value. This process is repeated for all parameters of the same type. The series of difference results obtained for each abnormal tree are then summarized to form a parameter deviation set containing the differences of all abnormal trees and all parameter categories. This allows the drone to automatically extract the magnitude and direction of errors in various key parameters for each abnormal tree, providing data input for the next step of statistically identifying and modeling the patterns of these deviations.

[0069] Step S33: Based on the statistical results obtained from the statistical analysis of the parameter deviation set, after identifying the deviation patterns, establish a correction function for each deviation pattern, and associate and bind the correction function with the corresponding abnormal unit to generate the fusion calibration result.

[0070] The core of statistical analysis of parameter deviation sets is to calculate the central tendency (such as mean and median) and dispersion (such as standard deviation) of various parameter differences. By plotting the differences and individual tree characteristic parameters, we can observe whether the deviations change with individual tree characteristics (such as the initially extracted tree height, crown size, or point cloud density) to identify the deviation patterns. For example, we may find that the diameter at breast height deviation is generally negative (systematic underestimation) and its absolute value is positively correlated with the tilt angle of the individual tree, or that the tree height deviation shows a specific proportional relationship in a certain tree species.

[0071] After identifying the patterns of deviation, a correction function is established for each type of deviation. For example, for the pattern that "diameter deviation is related to tilt angle", a linear function is fitted with the tilt angle of a single tree as the independent variable and the diameter correction (i.e., the negative of the deviation) as the dependent variable. Expressed as the corrected diameter at breast height. equal to the tilt angle of a single log Multiply by the regression coefficient Add a constant term For example, tree height deviation is related to point cloud shading rate. That is, when a single tree is in the lower canopy or its point cloud is severely shaded by neighboring trees, a linear function is fitted with point cloud shading rate as the independent variable and tree height deviation as the dependent variable. , expressed as tree height correction amount Equal to point cloud occlusion rate Multiply by the regression coefficient Add a constant term For example, crown width deviation is related to the scanning angle or the density of branches and leaves. For instance, the lack of point cloud due to lateral scanning may cause the extracted crown width value to shrink regularly in a specific direction. A linear function is fitted with the scanned angle as the independent variable and the crown width deviation as the dependent variable. , indicating the crown radius correction amount equal to scanning angle (For example, the angle between the incident direction of the laser beam and the normal to the center of the tree canopy) multiplied by the regression coefficient Add a constant term This function is the correction function for this type of deviation.

[0072] The established correction function is associated with abnormal trees that conform to the deviation pattern. That is, the applicable correction function is recorded for each abnormal tree, forming a mapping relationship set of correction rules for each tree. This set is the fusion calibration result. In this way, the measured deviation value for individual trees is summarized into a correction function. This allows subsequent steps to not only accurately correct the measured abnormal trees, but also logically provide a basis for correction of trees with similar characteristics but not measured. This significantly enhances the batch optimization capability of the entire method for abnormal tree parameters and the robustness of the algorithm.

[0073] Step S34: After extracting the anomalous tree point cloud clusters associated with the anomalous trees from the 3D terrain model, the correction function bound to the anomalous trees in the fusion calibration results is applied to the corresponding anomalous tree point cloud clusters to perform functional correction on the 3D coordinates and geometric features of each point in the corresponding anomalous tree point cloud clusters, generating a corrected point cloud cluster.

[0074] Based on the spatial location of each anomalous tree, a spatial query is performed in the point cloud data corresponding to the 3D terrain model. By utilizing the proximity relationship between the 3D coordinates of the points and their spatial locations, all point clouds belonging to the anomalous tree are extracted. The collection of these point clouds constitutes the anomalous tree point cloud cluster associated with the anomalous tree.

[0075] Next, the correction functions already bound to the anomalous individual trees are obtained, and these correction functions are applied to the corresponding anomalous individual tree point cloud clusters. The core operation is to numerically adjust the three-dimensional coordinates and geometric features of each point in the anomalous individual tree point cloud cluster according to the rules described by the correction functions: for example, for diameter at breast height correction, the adjustment is based on the tilt angle of the anomalous individual tree. A global radial scaling factor is calculated, and the point cloud coordinates of the tree trunk are scaled accordingly. For tree height correction, it may be based on the point cloud occlusion rate. The calculation of a vertical stretch allows for overall or layered adjustments to the point cloud elevation of the tree canopy. By applying a geometric transformation defined by a correction function to each point in the anomalous tree point cloud cluster, a corrected point cloud cluster that more closely resembles the measured ground conditions in shape and size is generated. This transforms the correction function model, established based on deviation patterns, into a geometric correction of the point cloud data, providing a data foundation for subsequently constructing a geometrically more realistic, consistent, and accurate optimized 3D terrain model.

[0076] Step S35: Spatially fuse the corrected point cloud cluster with other single-tree point cloud clusters in the 3D terrain model, excluding the anomalous single-tree point cloud clusters, to obtain the fused point cloud.

[0077] After correcting each anomalous tree point cloud cluster, based on spatial positional relationships, each corrected point cloud cluster is treated as a whole unit and its corresponding spatial position in the point cloud data of the 3D terrain model is replaced. At the same time, all other tree point cloud clusters and ground point cloud data in the 3D terrain model, except for these anomalous tree point cloud clusters, remain completely unchanged. Through this direct data replacement operation of replacing the old with the new, all corrected point cloud clusters and all other unchanged tree point cloud clusters are recombined in 3D space into a point cloud set, which is the fused point cloud.

[0078] Step S36: Based on the fused point cloud, optimize the 3D terrain model to obtain the optimized 3D terrain model.

[0079] By using the fused point cloud to perform holistic iterative reconstruction and optimization of the 3D terrain model, the overall errors accumulated by the 3D terrain model due to local anomalies or systematic biases in the point cloud data are corrected. This results in an optimized 3D terrain model with superior geometric accuracy and spatial consistency. The local accuracy improvement achieved in the previous steps by implementing "measurement-calibration-replacement" on abnormal trees is effectively integrated and transferred to the 3D model describing the entire forest plot topography and structure through a systematic model reconstruction process. The resulting optimized 3D terrain model not only has higher accuracy in identifying abnormal trees, but also systematically improves the accuracy of the digital representation of the overall terrain surface and the spatial location and morphological parameters of all trees due to reconstruction based on a higher quality and more consistent data source. This lays a data foundation for the subsequent generation of highly reliable plot survey reports.

[0080] Furthermore, step S36 also includes steps S361 to S364: Step S361: After separating the ground point cloud from the fused point cloud, a digital ground model is generated based on the ground point cloud, and the corrected terrain surface is obtained based on the digital ground model.

[0081] By simulating the physical process of a virtual cloth settling under gravity and colliding with the inverted, merged point cloud surface, the positions of the cloth nodes are adjusted through iterative calculations. Finally, the merged point cloud is classified into ground points and non-ground points based on whether the elevation difference between the original cloth nodes and the adjusted cloth nodes is less than a preset threshold. Next, based on the ground point cloud, a digital terrain model is generated using an irregular triangular mesh construction algorithm. The core of this algorithm is the use of Delaunay triangulation, ensuring that the circumcircle of each triangle formed by ground points does not contain any other ground points, thus connecting these ground points into a continuous, non-overlapping triangular mesh. This triangular mesh is the corrected terrain surface. The specific generation process is shown in step S13.

[0082] Step S362: Based on the corrected terrain surface, normalize the height values ​​reflected by each point in the fused point cloud to obtain a terrain-normalized point cloud.

[0083] Based on the corrected terrain surface, each point in the fused point cloud is normalized. Specifically, based on the planar coordinates of the points in the fused point cloud, the ground elevation value of the corresponding position directly below the point is calculated by linear interpolation in the triangular mesh of the digital ground model of the corrected terrain surface. Then, the ground elevation value before interpolation is subtracted from the ground elevation value obtained by interpolation, and the difference is the normalized height of the point relative to the corrected terrain surface. This calculation is performed sequentially on all points in the fused point cloud, thereby generating a terrain-normalized point cloud in which the height values ​​of all points have been converted to a unified zero reference based on the corrected terrain surface. This eliminates the influence of terrain undulations on the vertical distribution of the point cloud, so that the height value of each point in the terrain-normalized point cloud can directly represent its true height above the ground. This provides an absolutely consistent height reference for all subsequent operations based on height information, such as single tree height measurement, vertical structure analysis, and 3D model reconstruction.

[0084] Step S363: Based on the terrain-normalized point cloud, perform elevation correction and planar position verification on the inflection point coordinates included in the 3D terrain model to obtain the optimized inflection point coordinates.

[0085] Regarding elevation correction, based on the planar coordinates of each inflection point, all ground points within a preset radius around it (i.e., points whose normalized height values ​​are within a threshold range near zero) are searched and selected in the terrain normalized point cloud. Then, an inverse distance weighted interpolation method is used to calculate the weight of each ground point based on its horizontal distance to the inflection point (the closer the distance, the greater the weight). Next, the actual ground elevation value of each ground point (i.e., its original absolute elevation value, which can be obtained by adding its normalized height value to the ground elevation of its location, or by directly using the elevation attribute of the corresponding point) is multiplied by its corresponding weight. All the product results are summed and then divided by the sum of the weights to calculate the interpolated ground elevation at the inflection point. Finally, the ground elevation value obtained by this interpolated ground elevation is used to replace the original elevation value in the coordinates of the inflection point to complete the elevation correction of the inflection point coordinates. The preset radius is a distance parameter used to delineate a circular search area on the horizontal plane centered on the planar coordinates of the inflection point during the elevation correction.

[0086] Next, regarding planar position verification, a circular neighborhood with a preset radius is defined centered on the inflection point. The terrain-normalized point cloud within this neighborhood is extracted, and its two-dimensional spatial density distribution on the horizontal plane is calculated. This is typically achieved by dividing the neighborhood into a regular grid and counting the number of point clouds within each grid cell. The density distribution is analyzed to identify locations where density values ​​abruptly change. These locations usually correspond to the boundaries of the actual terrain or forest edges. A boundary line reflecting the actual orientation can be formed by connecting adjacent density abrupt change points. Then, based on the identified boundary line, the shortest vertical distance and direction from the current inflection point's planar coordinates to this boundary line are calculated. The planar coordinates of the inflection points are moved along this direction to the boundary line, thereby completing the fine-tuning of the planar coordinates of the inflection points to better fit the actual boundary features revealed by the terrain normalized point cloud. The optimized inflection point coordinates are obtained, ensuring that the coordinates of each inflection point of the forest plot boundary polygon are strictly aligned with the real geographical conditions reflected by the terrain normalized point cloud after fusion calibration and terrain optimization, not only in horizontal position but also in elevation value. This ensures that the spatial range of the forest plot defined by these inflection points has high geometric accuracy, providing a spatial framework for the final construction of an optimized three-dimensional terrain model that is completely consistent in terms of boundary, terrain and single tree structure.

[0087] Step S364: Combine the normalized point cloud of the terrain with the optimized inflection point coordinates to obtain the optimized three-dimensional terrain model.

[0088] The normalized point cloud of terrain and the optimized inflection point coordinates are integrated in terms of data structure and spatial representation. The optimized inflection point coordinates define the spatial boundary framework of the optimized 3D terrain model, while the normalized point cloud of terrain provides the 3D geometric information of all features and terrain details within the spatial boundary framework. The combination of the two is achieved by using the optimized inflection point coordinates as the vertex sequence of the boundary polygon and registering them together with the normalized point cloud of terrain in the same geodetic coordinate system. This constructs a 3D digital model that integrates the boundary framework and the internal dense point cloud. This model is the optimized 3D terrain model. The generated optimized 3D terrain model has both a geometric boundary defined by the inflection points that matches the actual geographical range and a 3D scene described by the point cloud that has been fully calibrated and normalized. This ensures that the optimized 3D terrain model achieves high accuracy at multiple levels, including the overall spatial range, local terrain undulations, and individual tree geometry. It provides a unique, authoritative, and reliable 3D data base for generating the final high-reliability sample plot inventory report.

[0089] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A non-contact, precise measurement method for forest plots using ground-air lidar fusion, characterized in that, include: Based on the scanning data of the forest plots and the pre-set forest stand map data, the boundaries of the forest plots are located and scanned to obtain a three-dimensional terrain model. The scanning data is obtained by scanning the forest plots using a lidar mounted on a UAV. Based on the point cloud data in the three-dimensional terrain model, individual tree segmentation and parameter extraction are performed to obtain individual tree data in the forest plot. Then, abnormal individual trees are located based on the individual tree data. After fusing and calibrating the measured data of the abnormal tree and the corresponding abnormal tree data, the three-dimensional terrain model is optimized based on the fusion calibration results to obtain an optimized three-dimensional terrain model. The measured data is obtained by scanning the abnormal tree with a ground-based handheld lidar and calculating the parameters. The fusion calibration is to fuse and calibrate the measured data obtained from the ground with the corresponding abnormal tree data extracted from the three-dimensional terrain model. After verifying the data of each individual tree using the optimized 3D terrain model, a verification report is generated based on the verification data.

2. The method for non-contact, precise measurement of forest plots using ground-to-air lidar fusion as described in claim 1, characterized in that, The steps of locating and scanning the boundaries of the forest plots based on the scanning data and pre-set forest stand map data to obtain a three-dimensional terrain model include: The point cloud of the UAV lidar is obtained based on the scanning data, and the boundary polygon of the forest plot is extracted based on the preset forest stand map data. Based on the UAV lidar point cloud, the vertex positions of the boundary polygon are corrected to obtain the corrected boundary polygon; Point cloud is extracted within the corrected boundary polygon to obtain point cloud data, and the three-dimensional terrain model is constructed based on the point cloud data.

3. The method for non-contact, precise measurement of forest plots using ground-to-air lidar fusion as described in claim 2, is characterized in that, The step of correcting the vertex positions of the boundary polygon based on the UAV lidar point cloud to obtain the corrected boundary polygon includes: Based on the UAV lidar point cloud, the point cloud located within a preset buffer distance of the boundary polygon is extracted as the boundary point cloud; Calculate the spatial density of the boundary point cloud, and after identifying the density abrupt change line in the boundary point cloud based on the spatial density, translate the vertices of the boundary polygon adjacent to the density abrupt change line along the direction of the density abrupt change line to obtain the initial vertex position; Based on the initial vertex positions, the boundary point cloud is subjected to orientation consistency analysis to obtain the main boundary direction. Then, the initial vertex positions are aligned according to the main boundary direction to obtain the corrected vertex positions. Establish the connection relationship between the corrected vertex positions to obtain the corrected boundary polygon.

4. The method for non-contact, precise measurement of forest plots using ground-to-air lidar fusion as described in claim 1, characterized in that, The steps of segmenting individual trees and extracting parameters based on the point cloud data in the three-dimensional terrain model to obtain individual tree data within the forest plot, and then locating abnormal individual trees based on the individual tree data, include: The point cloud data in the three-dimensional terrain model is normalized to obtain normalized point cloud data; The normalized point cloud data is clustered and segmented to obtain multiple individual point cloud clusters; For each individual tree point cloud cluster, parameter calculations are performed to obtain individual tree data including tree height, diameter at breast height, crown width, growth taper, and vertical structure; After integrating each piece of single-tree data into a parameter vector, the similarity between the parameter vectors is calculated, and the single-tree data corresponding to the single-tree data with a similarity lower than the preset similarity is located as the abnormal single-tree.

5. The method for non-contact, precise measurement of forest plots using ground-to-air lidar fusion as described in claim 1, characterized in that, The step of fusing and calibrating based on the measured data of the abnormal individual tree and the corresponding abnormal individual tree data includes: The measured data and the abnormal single tree data are spatially matched and time-synchronized with the same parameters to generate a comparison dataset; The differences of the same parameters in the comparison dataset are calculated item by item to obtain a set of parameter deviations including differences in tree height, diameter at breast height, and crown width. Based on the statistical results obtained from the statistical analysis of the parameter deviation set, after identifying the deviation patterns, a correction function is established for each deviation pattern, and the correction function is associated and bound with the corresponding abnormal individual to generate a fusion calibration result.

6. The method for non-contact, precise measurement of forest plots using ground-to-air lidar fusion as described in claim 5, is characterized in that, The step of optimizing the three-dimensional terrain model based on the fusion calibration results to obtain the optimized three-dimensional terrain model includes: After extracting the abnormal tree point cloud clusters associated with the abnormal tree from the three-dimensional terrain model, the correction function bound to the abnormal tree in the fusion calibration result is applied to the corresponding abnormal tree point cloud cluster to perform functional correction on the three-dimensional coordinates and geometric features of each point in the corresponding abnormal tree point cloud cluster, thereby generating a corrected point cloud cluster. The corrected point cloud cluster is spatially fused with other single tree point cloud clusters in the three-dimensional terrain model, excluding the abnormal single tree point cloud cluster, to obtain the fused point cloud. Based on the fused point cloud, the 3D terrain model is optimized to obtain the optimized 3D terrain model.

7. The method for non-contact, precise measurement of forest plots using ground-to-air lidar fusion as described in claim 6, characterized in that, The step of optimizing the 3D terrain model based on the fused point cloud to obtain the optimized 3D terrain model includes: After separating the ground point cloud from the fused point cloud, a digital ground model is generated based on the ground point cloud, and a corrected terrain surface is obtained based on the digital ground model. Based on the corrected terrain surface, the height values ​​reflected by each point in the fused point cloud are normalized to obtain a terrain-normalized point cloud. Based on the terrain normalized point cloud, the inflection point coordinates included in the three-dimensional terrain model are subjected to elevation correction and planar position verification to obtain optimized inflection point coordinates. The optimized 3D terrain model is obtained by combining the normalized point cloud of the terrain with the optimized inflection point coordinates.

Images