A forest tree number and position automatic detection method based on an HDBSCAN algorithm

Abstract

The application provides a forest tree number and position automatic detection method based on an HDBSCAN algorithm, and the method comprises the following steps: performing a multi-level threshold separation denoising operation on original point cloud data to obtain forest tree point cloud data; performing initial forest tree identification on the forest tree point cloud data through a density clustering algorithm with dynamic parameter optimization to obtain an initial identification result; performing three-dimensional shape correction and adjacent target fusion on the initial identification result to obtain a predicted identification result; and performing multi-dimensional feature verification on the predicted identification result to obtain a final identification result. Through the multi-level threshold separation, heterogeneous filtering technology and multi-dimensional verification, the forest tree detection task is accurately and efficiently completed, and the detection precision and anti-interference capability are improved.

Description

Technical Field

[0001] This invention relates to the field of forest tree detection and identification technology, and in particular to an automatic detection method for the number and location of trees based on the HDBSCAN algorithm. Background Technology

[0002] With the rapid development of remote sensing technology, lidar has become a core tool for forestry resource surveys due to its high-precision three-dimensional information acquisition capabilities. Traditional forest tree detection mainly relies on manual field measurements or visual interpretation based on optical images, which suffers from low efficiency, high cost, and strong subjectivity. In recent years, automated detection methods based on point cloud data have gradually become a research hotspot, especially density clustering algorithms (such as DBSCAN and HDBSCAN), which have shown potential in single-tree segmentation due to their adaptability to irregularly shaped data.

[0003] Traditional DBSCAN algorithms require manual setting of global parameters and cannot adapt to changes in forest stand structure. Although the traditional HDBSCAN algorithm solves some problems through hierarchical clustering, it still has the following shortcomings: 1) It does not consider the vertical distribution characteristics of trees, making it difficult to distinguish different tree species and obtain the true number of trees; 2) It is sensitive to point cloud density, and is prone to missed detections in sparse areas. Therefore, it is essential to design an automatic tree count and location detection method based on the HDBSCAN algorithm. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic tree count and location detection method based on the HDBSCAN algorithm. By employing multi-level threshold separation, heterogeneous filtering technology, and multi-dimensional verification, the method accurately and efficiently completes the tree detection task, thereby improving detection accuracy and anti-interference capability.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] An automatic method for detecting the number and location of trees in forests based on the HDBSCAN algorithm includes the following steps:

[0007] Multi-level threshold separation and noise reduction operations are performed on the raw point cloud data to obtain forest point cloud data;

[0008] Initial tree identification was performed on the forest point cloud data using a density clustering algorithm optimized with dynamic parameters, and the initial identification results were obtained.

[0009] The initial recognition results are subjected to three-dimensional morphological correction and neighboring target fusion to obtain the predicted recognition results;

[0010] The predicted recognition results are validated using multi-dimensional features to obtain the final recognition result.

[0011] Optionally, multi-level threshold separation and denoising operations are performed on the original point cloud data to obtain forest point cloud data, including:

[0012] The original point cloud data is separated by an adaptive elevation threshold to obtain dense and sparse point cloud regions.

[0013] Based on density guidance, virtual points are used to complete sparse areas of the point cloud to obtain the completed area;

[0014] Forest point cloud data is obtained by eliminating outlier noise points in dense areas and complete areas of the point cloud through anisotropic density filtering.

[0015] Optionally, the original point cloud data is separated using an adaptive elevation threshold to obtain dense and sparse point cloud regions, including:

[0016] The original point cloud data was divided into multiple analysis windows with a radius of 3m;

[0017] Use 25% of the point cloud elevation distribution within the analysis window as the window baseline value;

[0018] The window baseline value is dynamically adjusted based on the maximum elevation difference and average slope within the analysis window to obtain the window threshold.

[0019] The separation threshold is obtained by applying a Gaussian weighted average to the window threshold.

[0020] The original point cloud data is divided into dense point cloud regions and sparse point cloud regions based on the separation threshold.

[0021] Optionally, virtual point completion is performed on sparse regions of the point cloud based on density guidance to obtain the completed region, including:

[0022] The virtual direction is obtained by extending along the gradient direction of the sparse region of the point cloud;

[0023] Virtual height is obtained by interpolating the height trend surface of dense point cloud regions adjacent to sparse point cloud regions.

[0024] Virtual points are determined based on virtual direction and virtual height, and then inserted into sparse areas of the point cloud to obtain the completed area.

[0025] Optionally, outlier noise points in dense and completed areas of the point cloud are eliminated by anisotropic density filtering to obtain forest point cloud data, including:

[0026] An effective region is constructed with any point in the dense point cloud region or the completed region as the center point; the horizontal radius of the effective region is 3m and the vertical radius is 0.5m.

[0027] The spatial proximity factor of the effective region is obtained based on the Gaussian weighted distance decay function;

[0028] The difference outliers in the effective region are obtained by using the point cloud height variance, and an exponential penalty term is applied to the difference outliers to obtain the high consistency factor.

[0029] The composite density is obtained by normalizing and multiplying the spatial proximity factor and the high consistency factor;

[0030] Points with a composite density lower than the primary filtering threshold are removed to obtain the primary filtering data; the primary filtering threshold is 40% of the average density of the effective area.

[0031] The primary filtering threshold is adjusted based on the density change rate of the primary filtering data to obtain the secondary filtering threshold. The primary filtering threshold is then replaced with the secondary filtering threshold for data filtering to obtain forest point cloud data.

[0032] Optionally, the steps for dynamically optimizing the density clustering algorithm include:

[0033] Calculate the spatial dispersion and vertical concentration of forest point cloud data;

[0034] The core parameters of the HDBSCAN algorithm are updated based on spatial dispersion and vertical concentration; the core parameters include: minimum cluster size and core distance threshold.

[0035] Optionally, the formulas for calculating the minimum cluster size and the core distance threshold are as follows: M min =[15·(2-e -η )]·ζ 0.5 ; Among them, M min ε is the minimum cluster size. c η is the core distance threshold, η is the spatial dispersion, and ζ is the vertical concentration.

[0036] Optionally, the initial recognition result is subjected to three-dimensional morphological correction and neighbor target fusion to obtain the predicted recognition result, including:

[0037] Construct a triangular spatial topology network of the initial identification results;

[0038] Morphological compatibility is calculated based on the triangular spatial topology network; morphological compatibility includes: deviation of the principal axis direction angle, proportion of projected overlapping area, and similarity of height distribution;

[0039] When the morphological compatibility is higher than the fusion threshold, the adjacent initial recognition results are fused and the morphological compatibility is recalculated until the morphological compatibility is lower than the fusion threshold.

[0040] Optionally, the formula for calculating morphological compatibility is: Among them, C l For morphological compatibility, Sb H represents the ratio of the projected overlapping area. s For height distribution similarity, Δθ is the angle deviation along the main axis direction.

[0041] Optionally, the predicted recognition results are validated using multi-dimensional features to obtain the final recognition result. Specifically, a decision function is constructed based on the eigenvalues ​​of the covariance matrix, the projection and convex hull area, and the height standard deviation. The predicted recognition results are then validated using the decision function to obtain the final recognition result. The expression of the decision function is as follows: Among them, V t To verify the standard, λ1 and λ2 are two eigenvalues ​​of the covariance matrix, A. p Let A be the projected area. c Let σ be the area of ​​the convex hull. z For high standard deviation, The average tree height is represented by α, β, and γ, which are weighting coefficients.

[0042] According to specific embodiments provided by the present invention, the following technical effects are disclosed: The automatic detection method for the number and location of trees based on the HDBSCAN algorithm provided by the present invention includes: performing multi-level threshold separation and denoising operations on the original point cloud data to obtain tree point cloud data; performing initial tree identification on the tree point cloud data using a density clustering algorithm with optimized dynamic parameters to obtain initial identification results; performing three-dimensional morphological correction and neighbor target fusion on the initial identification results to obtain predicted identification results; and performing multi-dimensional feature verification on the predicted identification results to obtain the final identification results. This method accurately and efficiently completes the tree detection task through multi-level threshold separation, heterogeneous filtering technology, and multi-dimensional verification, improving detection accuracy and anti-interference capability. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of the automatic detection process for the number and location of trees according to the present invention;

[0045] Figure 2 This is a flowchart of the multi-level threshold separation and denoising process of the present invention;

[0046] Figure 3 This is a flowchart illustrating the data separation process of the present invention.

[0047] Figure 4 This is a flowchart of the virtual point completion process of the present invention;

[0048] Figure 5 This is a flowchart of the anisotropic density filtering process of the present invention;

[0049] Figure 6 This is a flowchart of the three-dimensional morphology correction process of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] like Figure 1 As shown, this invention provides an automatic detection method for the number and location of trees based on the HDBSCAN algorithm, comprising the following steps:

[0053] Step 100: Perform multi-level threshold separation and denoising operations on the original point cloud data to obtain forest point cloud data; specific steps are as follows. Figure 2 As shown, it includes:

[0054] Step 101: Separate the original point cloud data using an adaptive elevation threshold to obtain dense and sparse point cloud regions; specific steps are as follows. Figure 3 As shown, it includes:

[0055] Step 101.1: Divide the raw point cloud data into multiple analysis windows with a radius of 3m;

[0056] Specifically, the original point cloud is projected onto a horizontal plane, and a square grid with a side length of 3m is created to cover the entire point cloud area. The sliding window step of the square grid is 1.5m.

[0057] Step 101.2: Use 25% of the point cloud elevation distribution within the analysis window as the window baseline value;

[0058] Specifically, an octree data structure is used to perform fast spatial retrieval of the point cloud within the analysis window, obtaining the number of point clouds, elevation distribution, and coordinate range within each window. All point clouds within each window are arranged in ascending order of elevation, and the 25th percentile of the arrangement is used as the window reference value, effectively eliminating interference from low-lying ground points while preserving the point cloud at the base of vegetation.

[0059] Step 101.3: Dynamically adjust the window baseline value based on the maximum elevation difference and average slope within the analysis window to obtain the window threshold;

[0060] Specifically, the maximum elevation difference is obtained by subtracting the maximum and minimum elevations. A least-squares plane is fitted to the point cloud within the analysis window, and the average slope is determined based on the angle between the plane's normal vector and the vertical direction. The window threshold is obtained by adding the window baseline value, the maximum elevation difference, and the average slope using elevation difference compensation weights and slope correction amounts. The formula for calculating the window threshold is:

[0061] z th =z 25 +k1·Δz max +k2·tan(δ);

[0062] Among them, z th z is the window threshold. 25 Δz is the window reference value. max The maximum elevation difference is given by tan(δ), the average slope is given by δ, the angle between the normal vector of the plane and the vertical direction is given by δ, and k1 and k2 are the elevation difference compensation weight and slope correction amount, respectively, which are 0.15 and 0.3 in this embodiment.

[0063] Step 101.4: Perform a Gaussian weighted operation on the window threshold to obtain the separation threshold;

[0064] Specifically, a 3×3 window matrix is ​​established centered on the current window, and each window matrix is ​​assigned a Gaussian weight. The window thresholds are then weighted and fused using Gaussian weights to achieve a smoothing effect that eliminates abrupt changes in single-window thresholds.

[0065] Step 101.5: Divide the original point cloud data into dense point cloud regions and sparse point cloud regions according to the separation threshold.

[0066] Specifically, if the point cloud density is greater than the separation threshold + 0.5m, the point cloud is identified as a dense point cloud region; otherwise, it is identified as a sparse point cloud region.

[0067] Step 102: Perform virtual point completion on sparse areas of the point cloud based on density guidance to obtain the completed area; specific steps are as follows: Figure 4 As shown, it includes:

[0068] Step 102.1: Extend along the gradient direction of the sparse region of the point cloud to obtain the virtual direction;

[0069] Specifically, the sparse region is divided into 0.5m × 0.5m grid cells. After counting the number of point clouds in each cell, an initial density distribution map is generated. Then, Gaussian smoothing is used to eliminate local fluctuation noise, resulting in a continuous density field. The Sobel operator is used to perform gradient detection on the continuous density field, calculating the horizontal gradient vector of each grid cell, and then selecting the direction with the largest gradient magnitude as the extension direction.

[0070] Step 102.2: Obtain the virtual height by interpolating the height trend surface of the dense point cloud region adjacent to the sparse point cloud region;

[0071] Specifically, a 1m wide transition zone of point cloud is extracted as an interpolation sample at the boundary between sparse and dense point cloud regions. A quadratic polynomial surface is used to fit the elevation distribution of the transition zone point cloud to obtain the elevation trend surface equation, which is then solved using the least squares method, treating the solution as a virtual elevation.

[0072] Step 102.3: Determine virtual points based on virtual direction and virtual height, and insert the virtual points into sparse areas of the point cloud to obtain the completed area.

[0073] Step 103: Eliminate outlier noise points in dense and completed areas of the point cloud using anisotropic density filtering to obtain forest point cloud data. Specific steps are as follows: Figure 5 As shown, it includes:

[0074] Step 103.1: Construct an effective region with any point in the dense point cloud region or the completed region as the center point;

[0075] Specifically, the effective area has a horizontal radius of 3m to match the typical canopy projection range and a vertical radius of 0.5m to match the continuity characteristics of the tree trunk. The area size also needs to be dynamically adjusted based on the local point cloud density; if the local point cloud density is lower than 50 points / m², the area will be adjusted accordingly. 3 If the local point cloud density is higher than 200 points / m, then the horizontal radius will be expanded to 4m to capture potential sparse tree trunk points; 3 If the slope is greater than 25°, the horizontal radius will be reduced to 2m to prevent adjacent tree trunks from sticking together; if the slope is greater than 25°, the vertical radius will be expanded to 0.8m to accommodate the tilted tree trunks.

[0076] Step 103.2: Obtain the spatial proximity factor of the effective region based on the Gaussian weighted distance decay function;

[0077] Specifically, the horizontal distance weights between each point and the center point are calculated with a standard deviation of 1.2m, where the weight decreases by 30% for every 1m increase in horizontal distance. The vertical distance weights between each point and the center point are calculated with a standard deviation of 0.3m, where the weight decreases to 10% for every 0.1m increase in vertical distance.

[0078] Step 103.3: Obtain the outliers in the effective region through the point cloud height variance, and apply an exponential penalty term to the outliers to obtain the high consistency factor;

[0079] Specifically, if the variance is > 0.5m 2 If an anomaly is detected, points whose actual height deviates from the regional average height by 1.2 times the variance are marked as outliers. The expression for the exponential penalty term is: Where Δh is the deviation between the actual height and the average height.

[0080] Step 103.4: Normalize and multiply the spatial proximity factor and the high consistency factor to obtain the composite density;

[0081] Step 103.5: Remove points with a composite density lower than the primary filtering threshold to obtain the primary filtering data;

[0082] Specifically, the primary filtering threshold is 40% of the average density of the effective area. When the composite density is within the range of the average density of the effective area and 1.2 times the average density of the effective area, if at least 3 of the 8 points adjacent to that point have a composite density greater than 0.6 times the average density of the effective area, then that point is retained; otherwise, it is also removed.

[0083] Step 103.6: Adjust the primary filtering threshold according to the density change rate of the primary filtering data to obtain the secondary filtering threshold, and replace the primary filtering threshold with the secondary filtering threshold to filter the data and obtain forest point cloud data.

[0084] Specifically, the density change rate within a local window (5m×5m) is calculated with each retention point as the center point. If the density gradient within the window is <-0.2 / m, the primary filtering threshold is increased by 30%; if the density gradient within the window is >0.1 / m, the primary filtering threshold is decreased by 20%.

[0085] Step 200: Initial tree identification is performed on the forest point cloud data using a density clustering algorithm optimized with dynamic parameters to obtain initial identification results;

[0086] Specifically, the steps for dynamic parameter optimization of the density clustering algorithm are as follows: calculate the spatial dispersion and vertical concentration of the forest point cloud data, and update the core parameters of the HDBSCAN algorithm based on the spatial dispersion and vertical concentration; the core parameters include: minimum cluster size and core distance threshold.

[0087] Furthermore, the formulas for calculating the minimum cluster size and the core distance threshold are as follows:

[0088] M min =[15·(2-e - η)]·ζ0.5 ;

[0089]

[0090] Among them, M min ε is the minimum cluster size. c η is the core distance threshold, η is the spatial dispersion, and ζ is the vertical concentration.

[0091] Step 300: Perform 3D morphological correction and neighbor target fusion on the initial recognition results to obtain the predicted recognition results; specific steps are as follows: Figure 6 As shown, it includes:

[0092] Step 301: Construct the triangular spatial topology network of the initial recognition results;

[0093] Specifically, the three-dimensional centroid coordinates of each initial identification result (cluster) are calculated and used as nodes in the topological network. Then, based on the Delaunay triangulation algorithm, adjacent nodes are connected in three-dimensional space to form a triangular spatial topological network.

[0094] Furthermore, the triangular spatial topology retains only triangular sides with a length less than 6m (the upper limit of typical canopy spacing) to eliminate invalid long-distance connections. However, for targets with overlapping vertical projections, topological connections are forcibly established even if their centroid spacing exceeds the threshold. In areas with a terrain slope > 15°, the horizontal projection distance is converted into the actual distance along the slope.

[0095] Step 302: Calculate the morphological compatibility based on the triangular space topology; morphological compatibility includes: deviation of the principal axis direction angle, proportion of projected overlapping area, and similarity of height distribution;

[0096] Specifically, principal component analysis is performed on each target point cloud, the direction of the first principal component is extracted as the principal axis, the angle θ between adjacent target principal axes is obtained, and the deviation of the principal axis direction angle is calculated. The calculation formula is as follows: Project the target point cloud onto a horizontal plane to generate a two-dimensional convex hull polygon. The ratio of the overlapping area of ​​the polygon to the area of ​​its union is used as the projection overlap area ratio. Divide the target point cloud into bins at 1m vertical intervals to generate height distribution histograms. The cosine similarity between two target histograms is considered as the height distribution similarity.

[0097] More specifically, the formula for calculating morphological compatibility is:

[0098]

[0099] Among them, C l For morphological compatibility, S b H represents the ratio of the projected overlapping area. sFor height distribution similarity, Δθ is the angle deviation along the main axis direction.

[0100] Step 303: When the morphological compatibility is higher than the fusion threshold, the adjacent initial recognition results are fused and the morphological compatibility is recalculated until the morphological compatibility is lower than the fusion threshold.

[0101] It should be noted that by fusing the overlapping features of the principal axis direction and projection, over-segmentation caused by interlacing tree branches and leaves is avoided, thus improving the recognition accuracy. The influence of ground slope is considered during the construction of the triangular spatial topology network, reducing the overall recognition error.

[0102] Step 400: Perform multi-dimensional feature verification on the predicted recognition results to obtain the final recognition results.

[0103] Specifically, a decision function is constructed based on the eigenvalues ​​of the covariance matrix, the projection and convex hull area, and the height standard deviation. The validation criterion is obtained through the decision function. When the value of the validation criterion is greater than 0.82, the predicted identification result is determined to be a valid single tree.

[0104] More specifically, the expression for the decision function is:

[0105]

[0106] Among them, V t To verify the standard, λ1 and λ2 are two eigenvalues ​​of the covariance matrix, A. p Let A be the projected area. c Let σ be the area of ​​the convex hull. z For high standard deviation, The average tree height is represented by α, β, and γ, which are weighting coefficients, and in this embodiment, they are 0.6, 0.3, and 0.1, respectively.

[0107] The beneficial effects of this invention are as follows:

[0108] 1) Precise point cloud segmentation was achieved through multi-level threshold separation and dynamic elevation adjustment;

[0109] 2) Virtual point completion and heterogeneous filtering techniques improved the integrity of sparse regions and suppressed noise interference;

[0110] 3) The parameter optimization process driven by spatial dispersion and vertical concentration enables the clustering process to adapt to different tree species, thereby improving the detection accuracy;

[0111] 4) Verifying the detection results based on multi-dimensional features greatly improves the accuracy and reliability of the final results.

[0112] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0113] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An automatic method for detecting the number and location of trees based on the HDBSCAN algorithm, characterized in that, Includes the following steps: Multi-level threshold separation and noise reduction operations are performed on the raw point cloud data to obtain forest point cloud data; The initial tree identification was performed on the forest point cloud data using a density clustering algorithm optimized with dynamic parameters, and the initial identification results were obtained. The initial recognition result is subjected to three-dimensional morphological correction and neighbor target fusion to obtain the predicted recognition result; The predicted recognition results are verified using multi-dimensional features to obtain the final recognition result; Multi-level threshold separation and denoising operations are performed on the raw point cloud data to obtain forest point cloud data, including: The original point cloud data is separated by an adaptive elevation threshold to obtain dense and sparse point cloud regions. Based on density guidance, virtual points are used to complete the sparse regions of the point cloud to obtain the completed regions. The forest point cloud data is obtained by eliminating outlier noise points in the dense areas and the completed areas of the point cloud through anisotropic density filtering. The original point cloud data is separated by an adaptive elevation threshold to obtain dense and sparse point cloud regions, including: The original point cloud data is divided into multiple analysis windows with a radius of 3m; Use 25% of the point cloud elevation distribution within the analysis window as the window baseline value; The window baseline value is dynamically adjusted based on the maximum elevation difference and average slope within the analysis window to obtain the window threshold. The separation threshold is obtained by performing a Gaussian weighted operation on the window threshold. The original point cloud data is divided into dense point cloud regions and sparse point cloud regions based on the separation threshold.

2. The automatic detection method for the number and location of trees based on the HDBSCAN algorithm according to claim 1, characterized in that, Based on density guidance, virtual point completion is performed on the sparse regions of the point cloud to obtain the completed regions, including: The virtual direction is obtained by extending along the gradient direction of the sparse region of the point cloud; The virtual height is obtained by interpolating the height trend surface of the dense region of the point cloud adjacent to the sparse region of the point cloud. Virtual points are determined based on the virtual direction and the virtual height, and the virtual points are inserted into the sparse region of the point cloud to obtain the completed region.

3. The automatic detection method for the number and location of trees based on the HDBSCAN algorithm according to claim 1, characterized in that, The forest point cloud data is obtained by eliminating outlier noise points in the dense regions and the completed regions of the point cloud through anisotropic density filtering, including: An effective region is constructed with any point in the dense point cloud region or the completed region as the center point; the horizontal radius of the effective region is 3m and the vertical radius is 0.5m. The spatial proximity factor of the effective region is obtained based on the Gaussian weighted distance decay function. The difference outliers in the effective region are obtained by using the point cloud height variance, and an exponential penalty term is applied to the difference outliers to obtain the high consistency factor. The composite density is obtained by normalizing and multiplying the spatial proximity factor and the high consistency factor. Points with a composite density lower than the primary filtering threshold are removed to obtain the primary filtering data; the primary filtering threshold is 40% of the average density of the effective area. The primary filtering threshold is adjusted based on the density change rate of the primary filtering data to obtain the secondary filtering threshold. The primary filtering threshold is then replaced with the secondary filtering threshold for data filtering to obtain the forest point cloud data.

4. The automatic detection method for the number and location of trees based on the HDBSCAN algorithm according to claim 1, characterized in that, The steps for dynamically optimizing the density clustering algorithm include: Calculate the spatial dispersion and vertical concentration of the forest point cloud data; The core parameters of the HDBSCAN algorithm are updated based on the spatial dispersion and the vertical concentration; the core parameters include: minimum cluster size and core distance threshold.

5. The automatic detection method for the number and location of trees based on the HDBSCAN algorithm according to claim 4, characterized in that, The formulas for calculating the minimum cluster size and the core distance threshold are as follows: ; ;in, Minimum cluster size, The core distance threshold, For spatial dispersion, This represents the vertical concentration.

6. The automatic detection method for the number and location of trees based on the HDBSCAN algorithm according to claim 1, characterized in that, The initial recognition result is subjected to three-dimensional morphological correction and neighbor target fusion to obtain a predicted recognition result, including: Construct a triangular spatial topology network of the initial identification results; The morphological compatibility is calculated based on the triangular spatial topology network; the morphological compatibility includes: the deviation of the principal axis direction angle, the proportion of the projected overlapping area, and the similarity of the height distribution; When the morphological compatibility is higher than the fusion threshold, adjacent initial recognition results are fused and the morphological compatibility is recalculated until the morphological compatibility is lower than the fusion threshold; the formula for calculating the morphological compatibility is: ;in, For morphological compatibility, The ratio of the projected overlapping area. For high distribution similarity, The deviation of the included angle in the main axis direction.

7. The automatic detection method for the number and location of trees based on the HDBSCAN algorithm according to claim 1, characterized in that, The predicted recognition result is validated using multi-dimensional features to obtain the final recognition result. Specifically, a decision function is constructed based on the eigenvalues ​​of the covariance matrix, the area of ​​the projection and convex hull, and the standard deviation of the height. The predicted recognition result is then validated using the decision function to obtain the final recognition result. The expression of the decision function is as follows: ;in, To verify the standard, and These are the two eigenvalues ​​of the covariance matrix. For the projected area, Let be the area of ​​the convex hull. For high standard deviation, For average tree height, , and All are weighting coefficients.

Images