Lidar-based forest volume precision survey method, device and medium

Abstract

The present application relates to a kind of forest stock volume precision investigation method, equipment and medium based on laser radar, comprising the following steps, by airborne laser radar, target forest area is scanned at multiple angles and noise filtering, obtain forest point cloud data;Based on the forest point cloud data, three-dimensional reconstruction is carried out, obtain tree profile data, and the tree profile data is extracted to geometric parameter, obtain tree characteristic parameter;Based on the tree characteristic parameter, shape factor calculation is carried out to single tree in target forest area, obtain tree shape data, and the tree shape data is estimated to volume, obtain forest stock volume data, solve the technical problem that the existing artificial measurement method is difficult to comprehensively reflect the stock volume distribution of entire forest area.

Description

Technical Field

[0001] This invention relates to the field of lidar technology, and in particular to a method, equipment, and medium for accurate surveying of forest stock volume based on lidar. Background Technology

[0002] Forest stock volume is a core indicator in forest resource surveys. Traditional survey methods primarily rely on a combination of manual field measurements and sample plot investigations. Surveyors need to establish standard sample plots in the forest, using tools such as altimeters and diameter gauges to measure parameters like diameter at breast height (DBH) and tree height for each tree. Then, they calculate the stock volume of individual trees using volume tables or empirical formulas, and finally extrapolate the forest stock volume for the entire area. This method is well-established in plains areas and easily accessible forest regions, providing relatively detailed tree parameter information. With the development of remote sensing technology, some studies have begun to use satellite imagery or aerial photogrammetry for forest resource surveys. By analyzing the spectral characteristics and texture information of the images, they estimate forest stock volume, playing a significant role in large-scale forest resource monitoring.

[0003] However, existing manual measurement methods suffer from low efficiency in complex terrain and high-density forest areas. Furthermore, due to the limited number of measurement points, they fail to comprehensively reflect the distribution of forest stock volume across the entire forest region, resulting in insufficient representativeness of the survey results. While image-based remote sensing estimation methods offer wide coverage, they are limited by the lack of vertical information in two-dimensional imagery, making it impossible to accurately obtain three-dimensional characteristic parameters such as tree height and canopy structure. The calculation of tree shape factors relies on regionally uniform empirical values, ignoring the morphological differences of individual trees, which significantly impacts the accuracy of stock volume estimation. Therefore, a forest stock volume survey method is needed that can rapidly acquire three-dimensional forest structure information and achieve accurate measurement of individual trees. Summary of the Invention

[0004] The main objective of this invention is to provide a precise survey method for forest stock volume based on lidar, which solves the technical problem of how to accurately identify and spatially locate the early stress state of individual trees.

[0005] To achieve the above objectives, this invention provides a method for accurate forest stock volume survey based on lidar, comprising the following steps:

[0006] Forest point cloud data is obtained by scanning the target forest area from multiple angles and filtering noise using an airborne lidar.

[0007] Three-dimensional reconstruction is performed based on the forest point cloud data to obtain tree outline data, and geometric parameters are extracted from the tree outline data to obtain tree feature parameters.

[0008] Based on the tree characteristic parameters, the shape factor of individual trees in the target forest area is calculated to obtain tree shape data, and the volume of the tree shape data is estimated to obtain forest stock volume data.

[0009] Furthermore, the process of using airborne lidar to perform multi-angle scanning and noise filtering of the target forest area to obtain forest point cloud data includes:

[0010] Flight path planning is performed on the boundary range of the target forest area to obtain a multi-angle scanning route. Then, the target forest area is scanned by airborne lidar along the multi-angle scanning route to obtain a laser reflection signal sequence.

[0011] Flight attitude compensation is performed on the laser reflection signal sequence to obtain correction signal data, and three-dimensional coordinate calculation is performed on the correction signal data to obtain a point cloud coordinate set;

[0012] Outliers are removed from the point cloud coordinate set based on the echo intensity threshold to obtain a valid point cloud set. Ground points are then separated from the valid point cloud set to obtain forest point cloud data.

[0013] Furthermore, the step of performing three-dimensional reconstruction based on the forest point cloud data to obtain tree outline data includes:

[0014] Local elevation extremum search is performed on the forest point cloud data to obtain the tree crown vertex position, and regional growth clustering is performed on the forest point cloud data based on the tree crown vertex position to obtain the point cloud of a single tree.

[0015] The point cloud of a single tree is sliced ​​horizontally at equal intervals along the trunk axis to obtain multiple cross-sectional point sets. Edge points are connected and curves are fitted to each cross-sectional point set to obtain tree outline data.

[0016] Furthermore, the step of performing region growing clustering on the forest point cloud data based on the canopy vertex position to obtain a single tree point cloud includes:

[0017] Each of the tree crown vertices is marked as an initial seed point, and a seed point search domain is set for each of the initial seed points;

[0018] Based on the seed point search domain, the distance of the neighboring points in the forest point cloud data is judged and merged to obtain candidate merge points. The candidate merge points are iteratively expanded until no new points are included, and the tree cluster point cloud is obtained.

[0019] Boundary determination is performed on the overlapping areas of adjacent tree cluster point clouds to obtain cluster segmentation boundaries, and independent labels are assigned to each tree cluster point cloud based on the cluster segmentation boundaries to obtain a single tree point cloud.

[0020] Furthermore, the geometric parameter extraction of the tree outline data to obtain tree feature parameters includes:

[0021] The lowest and highest points in the vertical direction of the tree outline data are located to obtain the elevation of the trunk base point and the elevation of the crown apex. The difference between the elevation of the crown apex and the elevation of the trunk base point is calculated to obtain the height of a single tree.

[0022] Based on the tree trunk base point elevation, the cross-sectional profile at a preset distance from the ground in the tree outline data is extracted to obtain the breast height cross-sectional profile. The diameter of the breast height cross-sectional profile and the horizontal projection width of the outer edge of the crown are then calculated to obtain the tree characteristic parameters.

[0023] Furthermore, based on the tree characteristic parameters, shape factors are calculated for individual trees in the target forest area to obtain tree shape data, including:

[0024] The cross-sectional area is calculated based on the diameter at breast height (DBH) value among the tree characteristic parameters to obtain the DBH cross-sectional area. The DBH cross-sectional area is then multiplied by the height of the single tree to obtain the volume of the reference cylinder.

[0025] The area of ​​each layer of the cross-sectional profile in the tree outline data is calculated layer by layer to obtain a layered cross-sectional area sequence. The average cross-sectional area of ​​two adjacent layers in the layered cross-sectional area sequence is calculated and multiplied and summed with the corresponding layer spacing to obtain the measured volume of the tree trunk.

[0026] The ratio of the measured volume of the tree trunk to the volume of the reference cylinder is calculated to obtain the trunk shape coefficient. The trunk shape coefficient is then integrated with the height of the single tree and the diameter at breast height to obtain tree shape data.

[0027] Furthermore, volume estimation is performed on the tree shape data to obtain forest stock volume data, including:

[0028] Based on the diameter at breast height (DBH) value in the tree shape data, the cross-sectional area at DBH is calculated to obtain the cross-sectional area at DBH. The cross-sectional area at DBH is then multiplied by the single tree height value and the trunk shape coefficient to obtain the single tree volume value.

[0029] The volume values ​​of each tree within the target forest area are summed to obtain the total volume of the region. The total volume of the region is then correlated and integrated with the area of ​​the target forest region to obtain forest stock volume data.

[0030] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0032] This invention provides a precise forest stock volume survey method based on lidar, comprising the following steps: multi-angle scanning and noise filtering of a target forest area using an airborne lidar to obtain forest point cloud data; three-dimensional reconstruction based on the forest point cloud data to obtain tree outline data, and extraction of geometric parameters from the tree outline data to obtain tree feature parameters; shape factor calculation of individual trees in the target forest area based on the tree feature parameters to obtain tree shape data, and volume estimation based on the tree shape data to obtain forest stock volume data. This method solves the technical problem that existing manual measurement methods cannot comprehensively reflect the stock volume distribution of the entire forest area, and realizes the extraction of key feature parameters such as diameter at breast height (DBH) and tree height based on high-precision tree outline data, significantly improving the accuracy of parameter measurement, avoiding the systematic bias caused by traditional methods relying on empirical model calculations, and making the information at the individual tree scale more based on actual measurements. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of the steps of a precise forest stock volume survey method based on lidar in one embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of step S1 in one embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of step S2 in one embodiment of the present invention;

[0037] Figure 4 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0038] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0040] like Figure 1 As shown, Figure 1 This invention provides a method for accurate forest stock volume survey based on lidar, comprising the following steps:

[0041] Step S1: The target forest area is scanned from multiple angles and noise is filtered by airborne lidar to obtain forest point cloud data.

[0042] Specifically, above the target forest area, an aircraft equipped with lidar flies along a preset route. The lidar transmitter emits laser pulses to the ground at a frequency of hundreds of thousands of times per second. The laser beams are reflected after encountering targets at different heights, such as tree canopies, trunks, and the ground. The radar receiver records the three-dimensional coordinates and echo intensity information of each reflection point, thus forming a dense spatial point set. Because the aircraft repeatedly scans the same area from different positions and angles, supplementary data can be obtained from the sides or adjacent flight paths for the lower and middle parts of the tree trunks that were originally obscured by the tree canopy, thus acquiring relatively complete tree structure information. The raw point cloud often contains outliers caused by birds, suspended particles, or equipment vibration. These noise points can interfere with subsequent data analysis. Therefore, it is necessary to set distance thresholds and statistical filtering parameters to identify and remove these outliers. For example, if the average distance between a point and its surrounding neighboring points exceeds three times the normal range, it is judged as noise and deleted. The forest point cloud data obtained after this series of processing can be used for tree outline extraction and parameter calculation.

[0043] Step S2: Perform 3D reconstruction based on the forest point cloud data to obtain tree outline data, and extract geometric parameters from the tree outline data to obtain tree feature parameters.

[0044] Specifically, after obtaining the forest point cloud data, the point cloud is first processed by height into layers, separating ground points from vegetation points. Ground points are used to build a digital elevation model as a height benchmark, while vegetation points are used for subsequent tree identification. In the separated vegetation point cloud, individual trees often exhibit a high local point cloud density. By calculating the point cloud density distribution within a certain radius around each point, the location of the crown vertex corresponding to the density peak can be found. The search continues downwards from these vertices, merging adjacent point clouds onto the same tree. This process is similar to watershed division in the watershed algorithm. After completing the individual tree segmentation, the point cloud of each tree is extracted separately to construct tree contour data. During contour reconstruction, the Alpha Shapes algorithm or the convex hull algorithm is used to enclose the discrete point cloud into a continuous three-dimensional surface. With the outline data, we can carry out the calculation of geometric parameters. The tree height is directly taken as the difference between the maximum Z coordinate value of the tree point cloud and the ground elevation. The diameter at breast height (DBH) is measured by making a horizontal slice of the outline at a height of 1.3 meters above the ground. The DBH value is obtained by dividing the perimeter of the fitted cross-section outline by pi. The crown width parameter is obtained by measuring the maximum span after projecting it onto the horizontal plane. These parameters are combined to form the tree characteristic parameters.

[0045] Step S3: Based on the tree characteristic parameters, calculate the shape factor of individual trees in the target forest area to obtain tree shape data, and estimate the volume of the tree shape data to obtain forest stock volume data.

[0046] Specifically, after obtaining the tree characteristic parameters, the next step is to calculate the shape factor for each tree. This coefficient reflects the variation in taper of the trunk from the base to the tip. Traditional forestry surveys often use a fixed empirical value for the shape factor, such as 0.42 for conifers and 0.40 for broadleaf trees. However, in reality, the trunk morphology of different individuals within the same stand varies greatly; some trunks are straight and round, while others are noticeably tapered. This method utilizes the previously extracted tree outline data to create several horizontal sections at different heights on the trunk, measuring the diameter of each section. Assuming a cut is made every 2 meters from the ground upwards, a 20-meter-tall tree will yield 10 diameter data points for each section. Arranging these diameter values ​​in order of height reveals the trend in trunk thickness. A taper curve is then fitted using the least squares method, describing the functional relationship between diameter and height.

[0047] After obtaining the taper curve, integrating it yields the actual volume of the tree trunk. If the trunk is simplified to a cylinder with a base diameter equal to the diameter at breast height (DBH) and a height equal to the tree height, the volume of this cylinder serves as a theoretical reference. The ratio of the actual trunk volume to the cylinder volume is the tree's shape factor, which typically fluctuates between 0.3 and 0.5. For example, a Masson pine tree with a DBH of 26 cm and a height of 18 m has a cylinder volume of π × 0.13² × 18 ≈ 0.955 cubic meters. However, integrating the taper curve yields an actual volume of 0.41 cubic meters. Therefore, the shape factor is 0.41 ÷ 0.955 ≈ 0.43, slightly larger than the empirical value of 0.42, indicating a relatively full trunk. Performing this calculation for every tree within the target forest area creates a tree shape dataset, recording the shape factor of each individual tree.

[0048] With individual shape factors, volume estimation no longer relies on a uniform empirical coefficient. For the i-th tree, its volume Vi is calculated as Vi = fi × π × (Di / 2)² × Hi, where fi is the shape factor calculated earlier, Di is the diameter at breast height (DBH), and Hi is the tree height. This formula looks very similar to the traditional volume formula, but the key difference is that the shape factor is calculated individually for each tree, rather than using the same number for the entire forest. For a practical example, consider a Chinese fir forest with three trees. The first tree has a diameter at breast height (DBH) of 20 cm, a height of 15 meters, and a shape factor of 0.38, so its volume is approximately 0.38 × 3.14 × 0.1² × 15 ≈ 0.179 cubic meters. The second tree has a DBH of 22 cm, a height of 16 meters, and a shape factor of 0.41, so its volume is approximately 0.41 × 3.14 × 0.11² × 16 ≈ 0.249 cubic meters. The third tree has a DBH of 18 cm, a height of 14 meters, and a shape factor of 0.36, so its volume is approximately 0.36 × 3.14 × 0.09² × 14 ≈ 0.129 cubic meters. Adding the volumes of these three trees together gives 0.557 cubic meters, which is the total volume of this small sample plot.

[0049] In practice, a target forest area may contain hundreds or even thousands of trees. The program automatically iterates through each record in the tree shape data, applying the formula above to estimate the volume one by one. Finally, it sums up all the individual tree timber to obtain the forest stock volume data for the entire area. This data not only includes the total stock volume but also outputs detailed indicators such as stock volume per hectare and stock distribution for different diameter classes. For example, statistics show that trees with a diameter at breast height (DBH) of 20-30 cm account for 60% of the total stock volume, and large trees with a DBH of over 30 cm account for 25%. This information is very valuable for forest management decisions. Compared with traditional methods, this process reduces the estimation error by about 15%-20% in complex forest stands because it considers individual tree differences, especially for natural forests with irregular tree shapes, where the accuracy improvement is even more significant. The entire calculation process is based on the previously acquired point cloud data and extracted parameters, without introducing any additional human assumptions, ensuring the objectivity and traceability of the results.

[0050] In a specific embodiment, such as Figure 2 As shown, the process of obtaining forest point cloud data by performing multi-angle scanning and noise filtering of the target forest area using airborne lidar includes:

[0051] S11, Plan a flight path for the boundary range of the target forest area to obtain a multi-angle scanning path, and use an airborne lidar to emit and receive laser pulses along the multi-angle scanning path to obtain a laser reflection signal sequence.

[0052] S12, perform flight attitude compensation on the laser reflection signal sequence to obtain correction signal data, and perform three-dimensional coordinate calculation on the correction signal data to obtain a point cloud coordinate set;

[0053] S13, based on the echo intensity threshold, outliers are removed from the point cloud coordinate set to obtain a valid point cloud set, and ground points are separated from the valid point cloud set to obtain forest point cloud data.

[0054] Specifically, before data collection, flight paths need to be designed based on the boundary of the target forest area. Generally, multiple parallel, round-trip flight paths are deployed, with a 30%-40% overlap between adjacent paths to scan the same area from different angles. This results in a multi-angle scanning route. The airborne lidar on the aircraft continuously emits laser pulses as it flies along these routes. These pulses, upon hitting tree canopies, branches, and the ground, generate echoes. The radar receiver records these echo signals in chronological order, forming a laser reflection signal sequence. Because the aircraft experiences pitch, roll, and yaw changes due to airflow, the raw signal contains these disturbances. Therefore, the attitude angle data measured by the airborne inertial navigation system must be used to compensate for the flight attitude of the laser reflection signal sequence, subtracting the disturbances to obtain corrected signal data. After obtaining the corrected signal data, the distance is calculated by multiplying the laser round-trip time by half the speed of light. Then, combined with the emission angle and the aircraft's position, three-dimensional coordinates are calculated. The XYZ coordinates of each echo point are thus calculated and stored in a point cloud coordinate set. This set also contains some outliers. By setting an echo intensity threshold, points with abnormally low intensity are filtered out, leaving the valid point cloud set. Next, the ground points need to be separated from the vegetation points. A common approach is the progressive triangulation algorithm, which first selects the lowest points as seeds to construct the initial ground, and then progressively searches upwards for points that meet the slope conditions to be added to the ground network. Points that do not meet the conditions are classified as vegetation points. The final separated vegetation portion is the forest point cloud data.

[0055] In a specific embodiment, such as Figure 3 As shown, the process of performing 3D reconstruction based on the forest point cloud data to obtain tree outline data includes:

[0056] S21, perform local elevation extreme value search on the forest point cloud data to obtain the position of the tree crown vertex, and perform regional growth clustering on the forest point cloud data based on the position of the tree crown vertex to obtain the point cloud of a single tree.

[0057] S22, the point cloud of the single tree is sliced ​​horizontally at equal intervals along the trunk axis to obtain multi-layer cross-sectional point sets, and edge point connection and curve fitting are performed on each layer of the cross-sectional point sets to obtain tree outline data.

[0058] Specifically, after obtaining the forest point cloud data, several search grids are first divided horizontally, with each grid typically having a side length of about 2 meters. Then, the point with the largest Z-coordinate within each grid is found. These local highest points roughly correspond to the top of the tree canopy. This step is called local elevation extremum search, and the search result is the location of the tree canopy apex. Starting from each tree canopy apex, the clustering expands outwards, gradually incorporating neighboring points less than a certain threshold and with similar heights into the same cluster. This process is called region growing clustering. The growth stops when it encounters points from other tree canopies or when the height difference exceeds a set value. After this clustering process, the originally mixed forest point cloud data is separated into independent point clouds of individual trees. After obtaining the point cloud of a particular tree, horizontal cuts are made at regular intervals along the vertical direction of the trunk. The cut interval can be set to 0.5 meters or 1 meter. Each cut extracts a ring of points, and the set of these points is the cross-sectional point set. Multiple such cross-sectional point sets are obtained from the tree base to the treetop. For each layer of cross-sectional points, the scattered edge points need to be connected. Common methods include sorting by polar angle and connecting adjacent points sequentially, or fitting these points with a B-spline curve. The resulting smooth, closed curve can more accurately represent the tree trunk outline. By superimposing the outline curves of all layers by height, a complete tree outline data is reconstructed, which describes the morphological characteristics of the tree trunk from bottom to top.

[0059] In a specific embodiment, the step of performing a local elevation extremum search on the forest point cloud data to obtain the canopy vertex position includes:

[0060] The horizontal coverage area of ​​the forest point cloud data is divided into grids of fixed size to obtain point cloud partition grids, and the elevation values ​​of the point clouds in each point cloud partition grid are extracted to obtain a grid elevation set.

[0061] Based on the grid elevation set, the maximum elevation value of each point cloud partition grid is filtered to obtain local elevation extreme points, and the three-dimensional coordinates of the local elevation extreme points are recorded to obtain candidate vertex coordinates, wherein the candidate vertex coordinates include the spatial position of the point with the maximum elevation in each partition.

[0062] The candidate vertex coordinates are compared with the neighborhood elevation difference to obtain the vertex validity label. Based on the vertex validity label, the candidate vertex coordinates are removed from non-extreme points to obtain the tree crown vertex position.

[0063] Specifically, after obtaining the forest point cloud data, the first step is to establish a regular grid on the XY plane to organize these scattered points. The specific method involves first determining the minimum bounding rectangle of the point cloud coverage area, and then dividing this rectangle into several small grids of fixed dimensions. The grid side length is determined based on the forest stand density; for sparse stands, a 3m x 3m grid can be used, while for dense stands, a 1.5m x 1.5m grid is used. These grids form the point cloud partitioning grid. Each grid acts as a container, taking in all the point clouds falling within its area. Next, the Z-coordinate values ​​of the points are extracted from each container. For example, if a grid has 127 points, the elevation values ​​of these 127 points are read one by one and stored in an array. This array is the elevation set of that grid. Repeating this operation for all grids yields the grid elevation set.

[0064] After obtaining the elevation data for each grid, the maximum value needs to be selected from the elevation set of each grid. Assuming the grid with 127 points has an elevation range of 15.2 meters to 23.8 meters, then 23.8 meters is the maximum elevation value for this grid, and the corresponding point is marked as a local elevation extremum. After finding all the extremum points of all grids, their XYZ coordinates are recorded. This coordinate information constitutes the candidate vertex coordinate set. For a practical example, a cedar forest is divided into 200 grids, and after filtering, 178 local elevation extremum points are obtained. The coordinates of these points are (x1, y1, z1), (x2, y2, z2)...(x... 178 y 178 , z 178 Each set of coordinates represents the spatial location of the point with the highest elevation within that partition.

[0065] However, not all of these 178 candidate points are true tree apexes. Some may be protruding branches at the tree canopy edge or spurious peaks caused by data noise, requiring further verification of their validity. The verification method involves checking the elevations of other candidate points within a certain radius around each candidate point. This radius is typically set to about 1.2 times the average canopy width; for example, if the canopy width of this stand is generally around 4 meters, a radius of 5 meters is used. For candidate point i, if there is another candidate point j with an elevation more than 0.5 meters higher within a 5-meter radius, it indicates that point i is actually a lateral branch of the canopy represented by point j, rather than a true apex. In this case, point i is marked as invalid. This process is called neighborhood elevation difference comparison, and the comparison result is recorded in the vertex validity marker, with valid points marked as 1 and invalid points marked as 0.

[0066] Using the example above, after comparing the 178 candidate points with their neighborhoods, 32 points are found to be suppressed by higher-value points in their vicinity; these 32 points are marked as invalid. The program iterates through the vertex validity markers, removing the candidate vertex coordinates marked as 0 from the list; this step is called non-extreme point removal. After removal, 146 coordinate points remain, which are the actual tree crown vertex locations. Each retained vertex location corresponds to an independent tree, and subsequent region growing clustering is performed using these 146 points as seed points.

[0067] The entire search process seems simple, but parameter settings have a significant impact on the results. If the grid size is set too large, a single cell may contain multiple trees, causing some canopies to be missed; if it's set too small, the same canopy may be fragmented, generating redundant candidate points. The radius and elevation difference threshold for neighborhood comparison also need to be adjusted according to the tree species. For coniferous trees with relatively pointed canopies, the elevation difference threshold can be set to 0.3 meters, while for broad-leaved trees with larger canopies and flatter tops, the threshold should be relaxed to 0.8 meters. In practice, some grid cells may not have any point clouds falling into them; these empty grid cells are skipped and do not participate in subsequent calculations to avoid null pointer errors. Another situation is when the understory shrub layer is dense, forming secondary elevation extrema. In this case, an elevation filtering condition can be added, such as requiring the absolute elevation of candidate points to be greater than the ground elevation plus 8 meters, to exclude low shrub vertices in advance.

[0068] This method for searching local elevation extrema is sensitive to terrain undulations. If the ground slope is steep, the ground elevation within the same grid can differ by several meters, interfering with the identification of tree crown apex. To address this, the ground elevation corresponding to that point needs to be subtracted before extracting the elevation value, using relative height rather than absolute height for comparison. This eliminates the influence of terrain. For example, if the Z-coordinate of a point is 45.3 meters, but the ground elevation beneath it is 38.7 meters, then the tree height at that point is 45.3 - 38.7 = 6.6 meters. Using this relative height of 6.6 meters for comparison with the maximum value within the grid is reasonable. After this standardization process, even in mountainous forest areas with a 30-degree slope, the positioning accuracy of the tree crown apex can be maintained within 0.5 meters, meeting the accuracy requirements for subsequent single-tree segmentation.

[0069] In a specific embodiment, the step of performing region growing clustering on the forest point cloud data based on the canopy vertex position to obtain a single tree point cloud includes:

[0070] Each of the tree crown vertices is marked as an initial seed point, and a seed point search domain is set for each of the initial seed points;

[0071] Based on the seed point search domain, the distance of the neighboring points in the forest point cloud data is judged and merged to obtain candidate merge points. The candidate merge points are iteratively expanded until no new points are included, and the tree cluster point cloud is obtained.

[0072] Boundary determination is performed on the overlapping areas of adjacent tree cluster point clouds to obtain cluster segmentation boundaries, and independent labels are assigned to each tree cluster point cloud based on the cluster segmentation boundaries to obtain a single tree point cloud.

[0073] Specifically, the previously located tree canopy apex positions come in handy here. Each vertex is treated as an initial seed point, essentially scattering several seeds in the forest point cloud data. A search area needs to be defined around each seed; this area is called the seed point search domain. It is generally set as a spherical space with the seed point as the center and a radius of 2-3 meters. The radius depends on the average canopy width of the tree species. For example, in a Masson pine forest where the canopy width is mostly around 5 meters, the search domain radius would be set to 2.5 meters, so that it covers the main part of the canopy without encroaching on neighboring trees.

[0074] After the seed point search domain is determined, the program begins searching for all points falling within the sphere in the forest point cloud data. For the i-th seed point, its coordinates are (xi, yi, zi). Iterate through each point pj in the point cloud data, calculating the 3D distance dij between pj and the seed point: dij = √[(xj-xi)² + (yj-yi)² + (zj-zi)²]. If dij is less than the search radius and the point is not yet occupied by other seed points, pj is marked as a candidate merge point. This step is called distance judgment and merging, and its purpose is to include points near the seed point to form a preliminary cluster. Assume that 83 candidate merge points are found around seed point A, and these 83 points are temporarily assigned to the tree represented by A.

[0075] However, tree canopies are not perfect spheres; some branches may extend beyond the search domain. Therefore, the clustering range needs to be expanded outwards. Specifically, the 83 candidate merging points found earlier are treated as new seeds, each with a smaller search domain, typically reduced to a radius of about 1 meter. The search continues to examine the vicinity of these secondary seeds for unclassified points. If new points are found, they are included in the cluster and used as seeds for the next round of expansion. This iterative process is called iterative expansion. Expansion continues until no new points meeting the criteria are found in a given iteration. At this point, the point cloud for the tree is considered complete, and the resulting point cloud set is the tree cluster point cloud.

[0076] Take a certain tree in the Chinese fir forest as an example. The initial seed points collected 83 points in the first round. In the second round of iteration, these 83 points each searched outward and found 46 new points. In the third round, these 46 points continued to expand and found 19 points. In the fourth round, only 3 points were found. In the fifth round, no new points were found, and the iteration terminated. The clustered point cloud of this tree contains a total of 83 + 46 + 19 + 3 = 151 points. There are 146 trees in the whole forest, and each tree grows its own clustered point cloud like this. However, the problem is that the crowns of adjacent trees often touch each other, and their clustered point clouds may overlap at the edges.

[0077] For example, tree A and tree B are 4 meters apart. There is an overlapping band about 0.8 meters wide in the middle area of the clustered point clouds of the two trees. The points in this overlapping area can be attributed to either A or B. The solution is to calculate the distances from each point in the overlapping area to the vertices of the crowns of A and B respectively. Whichever distance is closer, it is attributed to that one. In this way, a dividing line is drawn in the middle of the overlapping band, and this line is called the clustering segmentation boundary. In actual calculation, not only the horizontal distance is considered, but also the influence of the height difference is added, and the three-dimensional Euclidean distance is used to determine the attribution. The coordinates of a controversial point p are (xp, yp, zp), the distance to the vertex of tree A is dA = √[(xp - xA)² + (yp - yA)² + (zp - zA)²], and the distance to the vertex of tree B is dB = √[(xp - xB)² + (yp - yB)² + (zp - zB)²]. If dA < dB, then p is assigned to tree A, otherwise it is assigned to tree B.

[0078] After completing the boundary determination, an independent label number is assigned to the clustered point cloud of each tree. All the points of tree A are marked as ID = 1, the points of tree B are marked as ID = 2, and so on. The point clouds of 146 trees are respectively labeled with numbers from 1 to 146. In this way, the originally mixed forest point cloud data is completely split. The point cloud corresponding to each label forms a single-tree point cloud. These single-tree point clouds are stored in a data structure and can be organized with a list. The i-th element of the list stores all the point coordinates of the tree with ID = i.

[0079] During the process of region-growing clustering, the setting of the search domain radius is crucial. If the radius is too small, it will lead to incomplete clustering and a large number of branches and leaves at the edge of the crown being missed; if the radius is too large, it is easy to mistakenly incorporate the point clouds of adjacent trees. Different parameters need to be set for coniferous trees and broad-leaved trees. Coniferous trees have a small crown width and clear boundaries. It is more appropriate to use an initial search domain of 2 meters and an iterative search domain of 0.8 meters; broad-leaved trees have a large crown width and spreading branches and leaves. The initial search domain needs to be widened to 3 meters and the iterative search domain also needs to be 1.2 meters to include the outer branches. There is also an optimization strategy to add a height constraint during iterative expansion, requiring that the newly incorporated points cannot be more than 2 meters lower than the lowest point in the current cluster, so as to avoid mistakenly calculating ground shrubs or fallen trees into the crown.

[0080] For areas with exceptionally high stand density, such as young forests where the trees are only 1.5 meters apart and adjacent canopies are almost merged, it's difficult to clearly define boundaries based solely on distance. In such cases, additional criteria can be introduced, such as intensity information. The laser echo intensity of the leaves on the canopy surface is often higher than that of the inner branches. If a sudden drop in echo intensity is observed at a clustering boundary, it indicates that the echo may have crossed the canopy surface and entered an adjacent tree. An intensity threshold can be used as an auxiliary condition for cluster segmentation. Another situation involves dead trees or trees with broken branches. The apex elevation of these trees is abnormally low, and previous extreme value searches may have missed them, causing the point cloud of the entire tree to be absorbed by neighboring healthy trees. To address this issue, a residual point check can be performed after region growth. Points that have never been classified can be isolated. If they spatially cluster together and possess tree height characteristics, they can be identified as additional individual trees.

[0081] After processing using this method, a 5-hectare Chinese fir forest, which originally contained 270,000 point clouds, was divided into 146 individual tree point clouds. The largest tree contained 2,834 points, while the smallest contained only 412 points, with the difference in the number of points reflecting the variation in tree size. Each individual tree point cloud serves as the basic material for subsequent 3D reconstruction. Once these are obtained, fine-grained measurement work such as cross-sectional slicing and contour fitting can be carried out.

[0082] In a specific embodiment, the step of performing distance judgment and merging on neighboring points in the forest point cloud data based on the seed point search domain to obtain candidate merge points includes:

[0083] Based on the seed point search domain, the spatial range of the forest point cloud data is retrieved to obtain the point set within the search domain, and the point coordinates of the point set within the search domain are extracted to obtain the coordinates of the neighboring points.

[0084] The Euclidean distance between the coordinates of the neighboring points and the corresponding initial seed points is calculated to obtain the point distance value. The point distance value is then filtered based on a distance threshold to obtain a set of nearest neighbor points.

[0085] Seed point attribution is determined for each point in the nearest neighbor set to obtain a point attribution label. Points with the same point attribution label are merged to obtain candidate merge points, wherein the candidate merge points include the seed point number and point cloud coordinates.

[0086] Specifically, after obtaining the seed point search domain parameters for a given initial seed point, such as the center of the sphere being (125.3, 68.7, 22.4) and a radius of 2.5 meters, the program begins filtering points in the forest point cloud data that meet the spatial conditions. To improve retrieval efficiency, a KD-tree or octree index structure is usually pre-built for the point cloud. During a query, the range search function is directly called, passing in the center coordinates and radius of the search domain. The function can then quickly return all points falling within this sphere's range. Assuming 217 points are retrieved within a certain search domain, the set of these points constitutes the point set within the search domain. Next, the coordinates of these 217 points are read out one by one. The coordinates of the j-th point are denoted as (xj, yj, zj), and all coordinates are organized into an array to obtain the coordinates of the neighboring points.

[0087] With these neighboring point coordinates, we need to calculate the actual distance from each point to the seed point. Let the seed point coordinates be (xs, ys, zs). For the j-th neighboring point, calculate dj = √[(xj-xs)² + (yj-ys)² + (zj-zs)²]. This dj is the point distance value. Although a spherical range search was performed previously, theoretically the distance between these 217 points should be less than 2.5 meters. However, in practice, there might be slight errors due to index precision or boundary point determination. Therefore, a distance threshold needs to be set for further filtering. The distance threshold is usually the search domain radius, which is 2.5 meters. Points with dj > 2.5 are removed, leaving, for example, 214 points that pass the filter. This set of points is called the nearest neighbor set.

[0088] These 214 nearest neighbors cannot be directly assigned to seed points yet because there may be multiple seed points in the forest point cloud data, and their search domains may overlap. Some points may fall within two or more search domains simultaneously. For each point in the nearest neighbor set, it is necessary to check whether it has already been claimed by other seed points. If a point p is still unclaimed, it is marked with the current seed point number. For example, if this is seed point number 15, "Seed point number = 15" is recorded in the attributes of point p. This record is the point's ownership tag. If point p has already been marked by seed point number 8, the distances from point p to these two seed points are compared. The one that is closer is retained, and the other is discarded.

[0089] In a pine forest, after the above determination, seed point 15 has 206 out of 214 neighboring points confirmed to belong to it, while the other 8 points are assigned to other seed points because they are closer to them. The program packages the coordinates and ownership numbers of these 206 points together, with each record in the format of "point number, X coordinate, Y coordinate, Z coordinate, belonging seed point number". For example, "point_3827, 126.1, 69.3, 21.8, 15" means that point 3827 is located at (126.1, 69.3, 21.8) and belongs to seed point 15. These points with ownership tags are merged and organized to form the candidate merge point set for seed point 15. The same operation is repeated on all 146 seed points, and each seed point collects a batch of candidate merge points. These points will then serve as the basis for iterative expansion, eventually gathering the point cloud of the entire tree. The core of the entire distance judgment and merging process lies in Euclidean distance calculation and competitive attribution mechanism. The former ensures the spatial continuity of the point cloud, while the latter solves the problem of point attribution conflict when the search domains overlap.

[0090] In a specific embodiment, the step of extracting geometric parameters from the tree outline data to obtain tree feature parameters includes:

[0091] The lowest and highest points in the vertical direction of the tree outline data are located to obtain the elevation of the trunk base point and the elevation of the crown apex. The difference between the elevation of the crown apex and the elevation of the trunk base point is calculated to obtain the height of a single tree.

[0092] Based on the tree trunk base point elevation, the cross-sectional profile at a preset distance from the ground in the tree outline data is extracted to obtain the breast height cross-sectional profile. The diameter of the breast height cross-sectional profile and the horizontal projection width of the outer edge of the crown are then calculated to obtain the tree characteristic parameters.

[0093] Specifically, after obtaining the tree outline data for a particular tree, first scan the vertical direction to find the extreme values ​​of the Z-coordinate. The smallest value is the trunk base elevation. For example, in the outline data of a certain cedar tree, the lowest point is 38.2 meters and the highest point is 56.7 meters. These two values ​​correspond to the trunk base elevation and the crown apex elevation, respectively. Subtracting these two numbers, 56.7 - 38.2 = 18.5 meters, gives the height of the individual tree. Next, the cross-section at breast height (DBH) needs to be extracted. In forestry tree measurement standards, DBH is defined as the trunk diameter at 1.3 meters above the ground. Therefore, a horizontal cut is made at a position 1.3 meters above the trunk base elevation, which is 38.2 + 1.3 = 39.5 meters. Points near this height are extracted from the outline data, allowing a deviation of 5 centimeters to accommodate data dispersion. The final set of points is the DBH cross-section outline. The cross-sectional profile is typically approximately circular or elliptical. A circle or ellipse is fitted to the points on the profile, and the diameter of the fitted shape is the diameter at breast height (DBH). For example, the diameter of the fitted circle is 24.6 cm. The profile points of the crown are projected onto a horizontal plane, and the maximum span of the projected area in the east-west and north-south directions is measured. The average or maximum of these two spans is taken as the crown width parameter. Assuming an east-west span of 4.8 meters and a north-south span of 5.2 meters, the average crown width is 5.0 meters. Summarizing these data—tree height 18.5 meters, DBH 24.6 cm, and crown width 5.0 meters—constitutes the tree's characteristic parameters.

[0094] In a specific embodiment, shape factor calculation is performed on individual trees in the target forest area based on the tree feature parameters to obtain tree shape data, including:

[0095] The cross-sectional area is calculated based on the diameter at breast height (DBH) value among the tree characteristic parameters to obtain the DBH cross-sectional area. The DBH cross-sectional area is then multiplied by the height of the single tree to obtain the volume of the reference cylinder.

[0096] The area of ​​each layer of the cross-sectional profile in the tree outline data is calculated layer by layer to obtain a layered cross-sectional area sequence. The average cross-sectional area of ​​two adjacent layers in the layered cross-sectional area sequence is calculated and multiplied and summed with the corresponding layer spacing to obtain the measured volume of the tree trunk.

[0097] The ratio of the measured volume of the tree trunk to the volume of the reference cylinder is calculated to obtain the trunk shape coefficient. The trunk shape coefficient is then integrated with the height of the single tree and the diameter at breast height to obtain tree shape data.

[0098] Specifically, after extracting the diameter at breast height (DBH) from the tree's characteristic parameters, for example, if a Masson pine tree has a DBH of 24.6 centimeters (0.246 meters), its area is calculated using a circular cross-section. The formula is π × (D / 2)². Substituting the value, we get π × (0.246 / 2)² = π × 0.123² ≈ 0.0475 square meters. This number is the DBH cross-sectional area. After obtaining the DBH cross-sectional area and the tree height of 18.5 meters, we multiply them: 0.0475 × 18.5 ≈ 0.879 cubic meters. This result represents the ideal volume assuming the trunk is a perfectly uniform cylinder with a constant thickness from bottom to top—the volume of a baseline cylinder. In reality, the trunk is definitely tapered, thicker at the bottom and thinner at the top, and its actual volume will be significantly smaller than this cylindrical shape. Therefore, precise measurements are needed to calculate the actual volume.

[0099] When performing horizontal slicing on the point cloud of a single tree, cuts were made along the trunk from base to top every 0.5 meters or 1 meter, resulting in many layers of cross-sectional contours. Now, we need to calculate the area of ​​each layer. Assuming this 18.5-meter-tall tree was cut into 19 layers at 1-meter intervals, the bottom layer has a fitted diameter of 26.3 centimeters and an area of ​​π × 0.1315² ≈ 0.0544 square meters; the second layer has a diameter of 25.1 centimeters and an area of ​​0.0495 square meters; the third layer has a diameter of 23.8 centimeters and an area of ​​0.0445 square meters, and so on, until the treetop. The 19th layer at the treetop has a diameter of only 8.2 centimeters and an area of ​​0.0053 square meters. Arranging these 19 area values ​​in ascending order gives us the layered cross-sectional area sequence.

[0100] After obtaining this sequence, we need to use it to calculate the actual volume of the tree trunk. The principle is similar to integration; we can consider the trunk as a series of short truncated cones, and the volume of each truncated cone is approximated by multiplying the average of the cross-sectional areas at both ends by the height. For example, the segment between the first and second layers has an upper area of ​​0.0495 square meters and a lower area of ​​0.0544 square meters. The average of these two areas is (0.0495 + 0.0544) / 2 = 0.0520 square meters. The spacing between the layers in this segment is 1 meter, so the volume is 0.0520 × 1 = 0.0520 cubic meters. Between the second and third layers, the average cross-sectional area is (0.0495 + 0.0445) / 2 = 0.0470 square meters, and the volume is 0.0470 × 1 = 0.0470 cubic meters. Calculate the volume of all 18 segments using this method: 0.0520 cubic meters for the first and second layers, 0.0470 cubic meters for the second and third layers, 0.0425 cubic meters for the third and fourth layers, and so on, up to 0.0065 cubic meters for the 18th and 19th layers. Finally, add up the volumes of all 18 segments: 0.0520 + 0.0470 + 0.0425 + ... + 0.0065. Let's assume the total volume is 0.375 cubic meters. This is the actual measured volume of the tree trunk.

[0101] We have two volume data points: a baseline cylinder volume of 0.879 cubic meters and a measured trunk volume of 0.375 cubic meters. Dividing the measured value by the baseline value, 0.375 / 0.879 ≈ 0.427. This ratio is the trunk shape factor, also known as the form factor. The trunk shape factor reflects the degree to which the trunk shape deviates from a cylinder; a smaller value indicates a more pointed trunk, while a larger value indicates a fuller trunk. A value of 0.427 is considered normal for conifers. For some plantation-grown fir trees with exceptionally good straightness, the trunk shape factor might reach 0.45 or even 0.48, while trees in natural secondary forests that grow crookedly might only have a trunk shape factor of 0.35.

[0102] After obtaining the trunk shape coefficient, it needs to be combined with other previously extracted parameters. Specifically, a data record is created with a format similar to "tree number, tree height, diameter at breast height (DBH), crown width, trunk shape coefficient," such as "tree_0042, 18.5m, 24.6cm, 5.0m, 0.427." This record fully describes the morphological characteristics of tree number 42. These records, when compiled, form a tree shape dataset. In a practical application in a Chinese fir forest, the trunk shape coefficient of each of the 146 trees was calculated, with values ​​ranging from 0.38 to 0.46, a mean of 0.42, and a standard deviation of 0.018. It can be seen that even within the same species and age stand, there are significant differences in the trunk shape coefficient between individual trees. This difference mainly stems from the combined influence of site conditions, competition, and genetic characteristics.

[0103] It is worth noting that the spacing of the stratified slices affects the accuracy of volume calculations. A 1-meter spacing may miss details in areas with drastic changes in taper; reducing the spacing to 0.5 meters improves accuracy but doubles the computational load. In practice, the volume difference between 0.5-meter and 1-meter spacings is usually within 3%, and considering computational efficiency, a 1-meter spacing is sufficient in most cases. However, for some special tree species, such as eucalyptus trees with a distinctly tapering shape at the top and bottom, or trees with significant constriction in the middle of the trunk due to snow and ice damage, it is best to reduce the spacing to 0.3-0.5 meters to accurately capture morphological changes. Additionally, the base of the trunk often has a swollen root collar; including the lowest swollen portion in the calculation will overestimate the volume. The common practice is to start slicing at 0.3 meters above the ground to avoid the influence of the root collar. After this standardization process, the calculation error of the shape factor can be controlled within 5%, meeting the accuracy requirements of forest resource surveys.

[0104] In a specific embodiment, the volume of the tree shape data is estimated to obtain forest stock volume data, including:

[0105] Based on the diameter at breast height (DBH) value in the tree shape data, the cross-sectional area at DBH is calculated to obtain the cross-sectional area at DBH. The cross-sectional area at DBH is then multiplied by the single tree height value and the trunk shape coefficient to obtain the single tree volume value.

[0106] The volume values ​​of each tree within the target forest area are summed to obtain the total volume of the region. The total volume of the region is then correlated and integrated with the area of ​​the target forest region to obtain forest stock volume data.

[0107] Specifically, the program reads the diameter at breast height (DBH) of a tree (24.6 cm) from the tree shape data. This is first converted to 0.246 meters, and then the DBH cross-sectional area is calculated using the formula π×(D / 2)², yielding π×0.123²≈0.0475 square meters. The tree's height is 18.5 meters, and its trunk shape coefficient is 0.427. Multiplying these three numbers together, we get 0.0475×18.5×0.427≈0.375 cubic meters. This result is the volume per tree, reflecting the yield of timber from that tree. The program iterates through each record in the tree shape data, applying this formula to each of the 146 trees. The first tree yields 0.375 cubic meters, the second 0.512 cubic meters, the third 0.289 cubic meters, and so on, calculating the volume per tree for all trees. Next, add up all 146 numbers: 0.375 + 0.512 + 0.289 + ... Let's assume the sum is 62.8 cubic meters. This is the total volume of the region. The measured area of ​​the target forest region is 5 hectares, or 50,000 square meters. Connecting the total volume of 62.8 cubic meters with the area of ​​5 hectares, we can calculate the volume per unit area as 62.8 ÷ 5 = 12.56 cubic meters per hectare. The final output forest volume data includes both the total volume of 62.8 cubic meters and the intensity indicator of 12.56 cubic meters per hectare. It can also include detailed information such as diameter at breast height (DBH) distribution and the proportion of dominant tree species, providing a quantitative basis for forest management.

[0108] Reference Figure 4 This invention also provides a computer device whose internal structure can be as follows: Figure 4As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0109] Those skilled in the art will understand that Figure 4 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0110] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0111] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0112] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0113] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for precise surveying of forest stock volume based on lidar, characterized in that, Includes the following steps: Forest point cloud data is obtained by scanning the target forest area from multiple angles and filtering noise using an airborne lidar. Three-dimensional reconstruction is performed based on the forest point cloud data to obtain tree outline data, and geometric parameters are extracted from the tree outline data to obtain tree feature parameters. Based on the tree characteristic parameters, the shape factor of individual trees in the target forest area is calculated to obtain tree shape data, and the volume of the tree shape data is estimated to obtain forest stock volume data.

2. The method for accurate forest stock volume survey based on lidar according to claim 1, characterized in that, The process involves multi-angle scanning and noise filtering of the target forest area using an airborne lidar to obtain forest point cloud data, including: Flight path planning is performed on the boundary range of the target forest area to obtain a multi-angle scanning route. Then, the target forest area is scanned by airborne lidar along the multi-angle scanning route to obtain a laser reflection signal sequence. Flight attitude compensation is performed on the laser reflection signal sequence to obtain correction signal data, and three-dimensional coordinate calculation is performed on the correction signal data to obtain a point cloud coordinate set; Outliers are removed from the point cloud coordinate set based on the echo intensity threshold to obtain a valid point cloud set. Ground points are then separated from the valid point cloud set to obtain forest point cloud data.

3. The method for accurate forest stock volume survey based on lidar according to claim 1, characterized in that, The process of performing 3D reconstruction based on the forest point cloud data to obtain tree outline data includes: Local elevation extremum search is performed on the forest point cloud data to obtain the tree crown vertex position, and regional growth clustering is performed on the forest point cloud data based on the tree crown vertex position to obtain the point cloud of a single tree. The point cloud of a single tree is sliced ​​horizontally at equal intervals along the trunk axis to obtain multiple cross-sectional point sets. Edge points are connected and curves are fitted to each cross-sectional point set to obtain tree outline data.

4. The method for accurate forest stock volume survey based on lidar according to claim 3, characterized in that, The step of performing region growing clustering on the forest point cloud data based on the canopy vertex position to obtain a single tree point cloud includes: Each of the tree crown vertices is marked as an initial seed point, and a seed point search domain is set for each of the initial seed points; Based on the seed point search domain, the distance of the neighboring points in the forest point cloud data is judged and merged to obtain candidate merge points. The candidate merge points are iteratively expanded until no new points are included, and the tree cluster point cloud is obtained. Boundary determination is performed on the overlapping areas of adjacent tree cluster point clouds to obtain cluster segmentation boundaries, and independent labels are assigned to each tree cluster point cloud based on the cluster segmentation boundaries to obtain a single tree point cloud.

5. The method for accurate forest stock volume survey based on lidar according to claim 1, characterized in that, The geometric parameter extraction of the tree outline data to obtain tree feature parameters includes: The lowest and highest points in the vertical direction of the tree outline data are located to obtain the elevation of the trunk base point and the elevation of the crown apex. The difference between the elevation of the crown apex and the elevation of the trunk base point is calculated to obtain the height of a single tree. Based on the tree trunk base point elevation, the cross-sectional profile at a preset distance from the ground in the tree outline data is extracted to obtain the breast height cross-sectional profile. The diameter of the breast height cross-sectional profile and the horizontal projection width of the outer edge of the crown are then calculated to obtain the tree characteristic parameters.

6. The method for accurate forest stock volume survey based on lidar according to claim 5, characterized in that, Based on the tree characteristic parameters, shape factors are calculated for individual trees in the target forest area to obtain tree shape data, including: The cross-sectional area is calculated based on the diameter at breast height (DBH) value among the tree characteristic parameters to obtain the DBH cross-sectional area. The DBH cross-sectional area is then multiplied by the height of the single tree to obtain the volume of the reference cylinder. The area of ​​each layer of the cross-sectional profile in the tree outline data is calculated layer by layer to obtain a layered cross-sectional area sequence. The average cross-sectional area of ​​two adjacent layers in the layered cross-sectional area sequence is calculated and multiplied and summed with the corresponding layer spacing to obtain the measured volume of the tree trunk. The ratio of the measured volume of the tree trunk to the volume of the reference cylinder is calculated to obtain the trunk shape coefficient. The trunk shape coefficient is then integrated with the height of the single tree and the diameter at breast height to obtain tree shape data.

7. The method for accurate forest stock volume survey based on lidar according to claim 6, characterized in that, The volume of the trees is estimated from the tree shape data to obtain forest stock volume data, including: Based on the diameter at breast height (DBH) value in the tree shape data, the cross-sectional area at DBH is calculated to obtain the cross-sectional area at DBH. The cross-sectional area at DBH is then multiplied by the single tree height value and the trunk shape coefficient to obtain the single tree volume value. The volume values ​​of each tree within the target forest area are summed to obtain the total volume of the region. The total volume of the region is then correlated and integrated with the area of ​​the target forest region to obtain forest stock volume data.

8. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images