Single tree segmentation method, device and equipment based on unmanned aerial vehicle laser radar, medium and product

Abstract

The invention discloses a single tree segmentation method, device and equipment based on unmanned aerial vehicle laser radar, a medium and a product, and relates to the field of single tree segmentation, and the method comprises the steps: obtaining unmanned aerial vehicle laser radar point cloud data of a target forest farm; acquiring position coordinates of each single tree to be segmented in the target forest farm by adopting an RTK positioning technology; processing the unmanned aerial vehicle laser radar point cloud data of the target forest farm to obtain unmanned aerial vehicle laser radar point cloud data belonging to the ground; performing normalization processing on the unmanned aerial vehicle laser radar point cloud data of the target forest farm by adopting the unmanned aerial vehicle laser radar point cloud data belonging to the ground; and by taking the position coordinates of each to-be-segmented individual tree of the target forest farm as the position coordinates of the initial seed points, clustering the unmanned aerial vehicle laser radar point cloud data of the normalized target forest farm by adopting a clustering algorithm based on point cloud distance discrimination to realize individual tree segmentation. And the method has the characteristics of high precision, high robustness and high adaptability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of single tree segmentation, and in particular to a single tree segmentation method and device based on an unmanned aerial vehicle laser radar, equipment, medium and product. BACKGROUND

[0002] Accurate investigation of forest resources is an important basis for forest genetic breeding decision-making. However, the traditional forestry investigation method mainly relies on manual field measurement, which not only consumes a large amount of manpower, material resources and time, but also has great difficulty in implementing in complex terrain and dense vegetation areas, and is difficult to meet the efficiency and accuracy requirements of large-area forest resource investigation.

[0003] At present, the genetic determination of major coniferous plantations in the world mainly adopts the offspring determination method. In order to improve the genetic evaluation efficiency of the number of parents, the offspring determination forest is usually established in a dense afforestation mode with a row spacing of no more than 4 meters, and rarely thinned, resulting in a generally high canopy density of more than 0.8. In such high-density stands, it is difficult for traditional laser height meters to accurately obtain the tree height vertex information of each tree, and large errors are easily generated, which seriously affects the accurate determination of single tree traits, and further limits the extraction of genetic parameters and the identification of excellent families.

[0004] However, in high-density stands, due to the serious overlap of tree crowns and the blurred boundaries between tree individuals, the accuracy of traditional single tree segmentation based on unmanned aerial vehicle laser radar is significantly reduced, and the phenomena of "over-segmentation" or "under-segmentation" often occur, which seriously affects the accuracy of subsequent parameter extraction.

[0005] Single tree segmentation is a prerequisite for obtaining single tree attributes such as tree height, diameter at breast height, crown width, etc. In high-density breeding forests, if high-precision single tree segmentation cannot be achieved, it will directly affect the quality and reliability of genetic determination data. The accuracy of some existing single tree segmentation algorithms is generally less than 80% in high-density stands, which cannot meet the actual needs of fine forestry resource investigation and breeding decision-making. Therefore, it is urgent to develop a single tree segmentation method suitable for high-density stand environment, with high precision, high robustness and high adaptability, to improve the efficiency and accuracy of phenotypic data collection of forest resources, and to provide strong support for forest breeding research. SUMMARY

[0006] The purpose of the present application is to provide a single tree segmentation method, device, equipment, medium and product based on an unmanned aerial vehicle laser radar, which can be suitable for high-density stand environment and has the characteristics of high precision, high robustness and high adaptability.

[0007] To achieve the above-mentioned purpose, the present application provides the following solutions:

[0008] In a first aspect, the present application provides a single tree segmentation method based on an unmanned aerial vehicle laser radar, comprising:

[0009] Obtain UAV LiDAR point cloud data of the target forest; the target forest includes all the individual trees to be segmented;

[0010] The carrier phase dynamic real-time differential positioning technology is used to obtain the position coordinates of each single tree to be split in the target forest;

[0011] Process the UAV LiDAR point cloud data of the target forest farm to obtain the UAV LiDAR point cloud data belonging to the ground;

[0012] The UAV lidar point cloud data belonging to the ground is used to normalize the UAV lidar point cloud data of the target forest farm to obtain the normalized UAV lidar point cloud data of the target forest farm;

[0013] Taking the position coordinates of each single tree to be segmented in the target forest as the position coordinates of the initial seed point, a clustering algorithm based on point cloud distance discrimination is used to cluster the normalized UAV lidar point cloud data of the target forest to achieve single tree segmentation.

[0014] In a second aspect, the present application provides a single wood splitting device based on a UAV laser radar, comprising:

[0015] The point cloud acquisition module is used to obtain the UAV lidar point cloud data of the target forest; the target forest includes all the individual trees to be segmented;

[0016] The positioning module is used to obtain the position coordinates of each single tree to be segmented in the target forest using carrier phase dynamic real-time differential positioning technology;

[0017] The ground point cloud extraction module is used to process the UAV lidar point cloud data of the target forest farm to obtain the UAV lidar point cloud data belonging to the ground;

[0018] A normalization module is used to use the UAV lidar point cloud data belonging to the ground to normalize the UAV lidar point cloud data of the target forest farm, thereby obtaining normalized UAV lidar point cloud data of the target forest farm;

[0019] The segmentation module is used to cluster the normalized UAV lidar point cloud data of the target forest farm using the position coordinates of each single tree to be segmented in the target forest farm as the position coordinates of the initial seed point, and adopt a clustering algorithm based on point cloud distance discrimination to achieve single tree segmentation.

[0020] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned single tree segmentation method based on drone lidar.

[0021] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the single tree segmentation method based on the unmanned aerial vehicle laser radar.

[0022] In a fifth aspect, the present application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the single tree segmentation method based on the unmanned aerial vehicle laser radar.

[0023] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0024] The present application provides a single tree segmentation method, device, equipment, medium and product based on an unmanned aerial vehicle laser radar. In the process of processing, the position coordinates of the single tree to be segmented obtained by using the carrier phase dynamic real-time difference (RTK) positioning technology are used as the position coordinates of the initial seed point, which significantly improves the recognition ability of the target tree in the high-density stand and is suitable for the high-density stand environment. The present application combines the real position coordinates of the single tree to be segmented obtained by the RTK positioning technology and the clustering algorithm based on the point cloud distance discrimination, has the advantages of high precision, strong adaptability and good universality, has the characteristics of high precision, high robustness and high adaptability, and can effectively solve the problem that the single tree is difficult to accurately segment in the complex stand, and is especially suitable for the forest environment with high density and complex forest structure. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0026] Figure 1 A flowchart of a single tree segmentation method based on an unmanned aerial vehicle laser radar provided by an embodiment of the present application is shown in the figure.

[0027] Figure 2 A flowchart of a single tree segmentation method based on an unmanned aerial vehicle laser radar provided by another embodiment of the present application is shown in the figure.

[0028] Figure 3 CHM maps generated by different methods are shown in the figures.

[0029] Figure 4 An effect diagram of single tree segmentation based on seed points generated based on CHM is shown in the figure.

[0030] Figure 5 An effect diagram of single tree segmentation based on real positioning points is shown in the figure.

[0031] Figure 6 Regression fitting diagram of true tree height and extracted tree height for the optimal method combination;

[0032] Figure 7 A schematic diagram of the functional modules of a single wood splitting device based on a UAV laser radar provided in another embodiment of the present application;

[0033] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0035] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0036] In an exemplary embodiment, Figure 1 As shown, a single tree segmentation method based on UAV laser radar is provided, including:

[0037] Step 201: Obtain UAV lidar point cloud data of the target forest; the target forest includes all individual trees to be segmented.

[0038] Step 202: Carrier phase dynamic real-time differential positioning technology is used to obtain the position coordinates of each individual tree to be segmented in the target forest. Specifically, RTK positioning technology is used to collect the position coordinates of each individual tree to be segmented in the target forest in the same coordinate system as the UAV lidar point cloud data of the target forest.

[0039] Step 203: Process the UAV lidar point cloud data of the target forest farm to obtain UAV lidar point cloud data belonging to the ground.

[0040] Step 204: Use the UAV lidar point cloud data belonging to the ground to normalize the UAV lidar point cloud data of the target forest farm to obtain normalized UAV lidar point cloud data of the target forest farm.

[0041] Step 205: Using the position coordinates of each individual tree to be segmented in the target forest as the position coordinates of the initial seed point, a clustering algorithm based on point cloud distance discrimination is used to cluster the normalized UAV lidar point cloud data of the target forest to achieve individual tree segmentation.

[0042] In another exemplary embodiment of the present application, the UAV lidar point cloud data of the target forest farm is processed to obtain the UAV lidar point cloud data belonging to the ground, specifically including:

[0043] The improved Progressive TINDensification (IPTD) algorithm is used to extract ground points from the UAV lidar point cloud data of the target forest farm, and the UAV lidar point cloud data belonging to the ground is obtained.

[0044] In another exemplary embodiment of the present application, an improved progressively encrypted triangulated network filtering algorithm is used to extract ground points from the UAV lidar point cloud data of the target forest farm to obtain the UAV lidar point cloud data belonging to the ground, specifically including:

[0045] Determine the stand boundary according to the position coordinates of each single tree to be divided in the target forest;

[0046] Determine the study area within the target forest farm based on the stand boundaries;

[0047] The UAV LiDAR point cloud data of the target forest farm was cropped to obtain the UAV LiDAR point cloud data of the study area;

[0048] The UAV LiDAR point cloud data for the study area was sequentially processed to remove outliers, resample, and remove noise points, resulting in the pre-processed UAV LiDAR point cloud data for the study area. Specifically, aerial points significantly above the ground (aerial noise points) and abnormally low points significantly below the ground (ground error points) were manually removed. Abnormally high and low points were manually removed to ensure data quality. The point cloud data after outlier removal was then resampled and noise removed to reduce data redundancy and remove noise points.

[0049] An improved progressively encrypted triangulated network filtering algorithm is used to extract ground points from the pre-processed UAV LiDAR point cloud data of the study area, obtaining UAV LiDAR point cloud data belonging to the ground. An improved progressively encrypted triangulated network filtering algorithm is used to perform ground filtering and extract ground points from the point cloud data to distinguish between ground points and non-ground points.

[0050] In another exemplary embodiment of the present application, the UAV LiDAR point cloud data belonging to the ground is used to normalize the UAV LiDAR point cloud data of the target forest farm to obtain the normalized UAV LiDAR point cloud data of the target forest farm, specifically including:

[0051] A digital elevation model (DEM) was constructed based on ground-based UAV LiDAR point cloud data. Specifically, a digital elevation model (DEM) with a spatial resolution of 0.3 m was generated using the Kriging interpolation method (K).

[0052] The UAV lidar point cloud data of the target forest farm is normalized using the digital elevation model to obtain the normalized UAV lidar point cloud data of the target forest farm.

[0053] In another exemplary embodiment of the present application, the UAV LiDAR point cloud data of the target forest farm is normalized using a digital elevation model to obtain normalized UAV LiDAR point cloud data of the target forest farm. Specifically, the UAV LiDAR point cloud data of the target forest farm is subtracted from the digital elevation model to obtain normalized UAV LiDAR point cloud data of the target forest farm. Specifically, the Z coordinate of the UAV LiDAR point cloud data of the target forest farm corresponding to the (x, y) coordinate point is subtracted from the elevation value of the digital elevation model corresponding to the (x, y) coordinate point to eliminate the influence of the terrain, thereby obtaining normalized point cloud data, thereby eliminating the interference of terrain undulations on the point cloud single tree segmentation.

[0054] In another exemplary embodiment of the present application, the position coordinates of each individual tree to be segmented in the target forest are used as the position coordinates of the initial seed point, and a clustering algorithm based on point cloud distance discrimination is used to cluster the normalized UAV lidar point cloud data of the target forest to achieve individual tree segmentation, specifically including:

[0055] For any tree to be segmented, the coordinates of the tree to be segmented are used as the initial seed point, and the coordinates of the target seed point are determined based on the initial seed point. Specifically, a local maximum point is searched near the initial seed point to be used as the seed point for clustering, i.e., the target seed point.

[0056] Determine whether the distance between the target point and the target seed point is less than a preset spatial spacing threshold to obtain a first judgment result; the target point is the point cloud data other than the target seed point in the normalized UAV lidar point cloud data of the target forest farm.

[0057] If the first judgment result is yes, it is determined whether the distance between the target point and the target seed point meets the minimum spacing rule to obtain a second judgment result.

[0058] If the second judgment result is yes, the target point and the target seed point are determined to belong to the same tree to be segmented, and tree segmentation is achieved. Specifically, point clouds whose distance from the target seed point meets the minimum spacing rule are grouped into the point cloud cluster of the tree, and point clouds that exceed the minimum spacing rule are removed from the cluster, thereby achieving accurate segmentation of the tree point cloud.

[0059] In practical applications, a preset spatial spacing threshold serves as the initial screening criterion for determining whether a point belongs to a target tree. Its value must be approximately equal to the crown radius. If the minimum distance between a point and an existing point on the target tree (the target seed point) exceeds this threshold, the point is directly classified as a non-target tree. If the distance is less than or equal to the threshold, further judgment is made based on the minimum spacing principle. This serves to initially filter out points that clearly do not belong to the target tree, reducing subsequent computational effort while also preventing points that are too far apart from each other from being mistakenly classified as belonging to the same tree.

[0060] In practical applications, the minimum distance principle is a refined classification rule after the preset spatial distance threshold is filtered, which is used to determine the belonging of points that are closer to the target tree. Specifically, it compares the minimum distance (d min1 ) and the minimum distance to the existing points in the non-target tree (d min2 ):If d min1 ≤d min2 , the point is classified as a target tree; otherwise, it is classified as a non-target tree. Its function is to solve the problem of assigning points within the threshold range. It is particularly suitable for areas with dense or overlapping trees. By comparing the distance, it ensures that the point is assigned to the closest tree cluster.

[0061] This application uses the local highest point of the single tree positioning point search as the target seed point, performs top-down clustering growth based on the point cloud distance discrimination clustering algorithm, and judges whether they belong to the same single tree based on the preset spatial spacing threshold and minimum spacing rule, thereby achieving high-precision single tree point cloud segmentation.

[0062] This application uses the point cloud data obtained by the UAV lidar and the position information of each tree in the same coordinate system. By identifying the forest stand boundary, the study area is extracted in the point cloud, and the aerial points that are obviously higher than the ground objects and the low points that are obviously lower than the ground surface are manually removed. On this basis, the steps of resampling, denoising, ground filtering, generation of digital elevation model, point cloud normalization, etc. are carried out to complete the data preprocessing. Then, the single tree positioning is used as the initial seed point, combined with the point cloud distance discrimination clustering algorithm to complete the single tree segmentation. In response to the demand for higher accuracy of single tree segmentation in breeding forests, the single tree positioning information is added in the single tree segmentation process to improve the single tree segmentation accuracy in forest stands with high canopy density.

[0063] This application introduces a directional optimization strategy in the cluster segmentation process (the position coordinates of the individual trees to be segmented are obtained using carrier phase dynamic real-time differential positioning technology), which effectively improves the ability to segment adjacent individual trees in high-density forest stands.

[0064] This application introduces a directional optimization strategy in the key link of single tree segmentation, which significantly improves the ability to identify target trees in high-density forest stands.

[0065] This application combines the real positioning information of individual trees with a clustering algorithm based on spatial distance discrimination. It has the advantages of high precision, strong adaptability, and good versatility. It is especially suitable for scenarios with compact tree species layout and complex structure in breeding forests. It can not only achieve high-precision segmentation of individual trees, but also provide a reliable basis for the extraction of individual traits such as diameter at breast height, tree height, and canopy structure. It also lays a data foundation for subsequent genetic parameter analysis and selection of excellent families.

[0066] In order to verify the applicability and accuracy of the single tree segmentation method based on UAV laser radar provided in the above embodiment of this application, this application also provides an embodiment, such as Figure 2 As shown, the general steps are:

[0067] By identifying the forest stand boundary, the region of interest (ROI) is extracted from the point cloud, and aerial points that are obviously above the ground and points that are obviously below the ground are manually removed.

[0068] On this basis, resampling, denoising, ground filtering (separating ground points), generating digital elevation models, point cloud normalization, and generating canopy height models (CHM) are performed to complete data preprocessing.

[0069] Then, using CHM and RTK positioning, we obtained two types of seed points: CHM seed points and RTK real-world positioning seed points. We then combined three algorithms, including marker-controlled watershed segmentation (MWS), region growing (SRG), and point cloud-based cluster segmentation (PCS), to complete the segmentation of individual trees and evaluate their accuracy. The CHM seed point identification method uses a fixed moving window of 2.95m, detecting local maxima in the image as tree vertices, and setting the minimum tree height to 10m. RTK real-world positioning seed points are obtained from the actual measured tree positions. More detailed implementation steps include:

[0070] Step 1: Collection of original data of experimental forest.

[0071] A 37-year-old Japanese larch experimental forest was selected from a pre-set forest farm as the experimental plot for data collection.

[0072] (1) UAV lidar data collection

[0073] During the leaf-fall period of the experimental forest, a DJI M600Pro drone equipped with a RIEGLVUX-1UAV22 lidar sensor was used to collect lidar point cloud data, covering 11.78 hm2 of the study area. 2 .

[0074] The operating wavelength of the laser radar is near-infrared 1550nm, and the collection adopts two laser pulse repetition frequencies of 380kHz and 550kHz respectively. The flight design plan is a "well"-shaped route layout, with two sorties of collection in the forward and reverse directions respectively, for a total of four flights.

[0075] The flight parameters are: flight altitude: 300m; flight speed: 8m / s; route spacing: about 70m; lateral overlap rate: 50%; laser scanning angle: ±45°.

[0076] (2) Point cloud data solution and fusion.

[0077] Point cloud data processing includes two stages: UAV flight trajectory calculation and raw laser data matching:

[0078] First, POSPac software is used to jointly calculate GNSS (Global Navigation Satellite System) and IMU (Inertial Measurement Unit) data to obtain a high-precision flight trajectory.

[0079] Then, the original point cloud is registered and fused with the above trajectory using RiPROCESS software.

[0080] Finally, three types of fused point cloud data are generated and output in standard LAS format, as shown in Table 1, including:

[0081] 380kHz forward and reverse data fusion (denoted as 380);

[0082] 550kHz forward and reverse data fusion (denoted as 550);

[0083] 380kHz and 550kHz mixed frequency data fusion (recorded as 380550).

[0084] Table 1 LiDAR point cloud data

[0085] data Maximum number of echoes Maximum point cloud density Point cloud density average 380 7 3148 741 550 3 2222 301 380550 7 4467 1041

[0086] (3) Field investigation and accuracy verification data collection.

[0087] A detailed growth factor survey was conducted on the 3,486 trees retained within the experimental forest. Diameter at breast height (DBH) (cm) was measured using a DBH ruler. Tree height (H) (m) was measured using a Haglof Vertex III laser altimeter. Crown width in the east-west and north-south directions (m) was measured using a tape measure. Individual tree positioning was determined using a Huace M9 RTK receiver using carrier phase kinematic real-time differential (RTK) technology. Specific values ​​are shown in Table 2.

[0088] All collected data are uniformly projected to the WGS 1984 UTM Zone 51N coordinate system.

[0089] Table 2 Descriptive statistics of the experimental forest

[0090] Statistical indicators Tree height / m Diameter at breast height / cm North-south crown width / m East-west crown width / cm Minimum 13.00 10.80 0.70 0.30 average value 26.48 24.41 3.68 3.71 Maximum 33.90 44.80 10.10 9.30

[0091] Step 2: Preprocessing of original point cloud data.

[0092] To achieve high-precision tree identification and parameter extraction, the raw UAV lidar point cloud data requires standardized preprocessing. First, noise removal is performed on the three fused point cloud data sets (380, 550, and 380550 pixels) to remove isolated points and outliers, ensuring the overall quality and spatial consistency of the point cloud data.

[0093] After noise removal, the point cloud data is subjected to terrain filtering using an improved progressively encrypted triangulated network filtering algorithm. This method constructs a progressively encrypted irregular triangulated network (TIN) and gradually classifies ground points from non-ground points based on terrain features, thereby achieving ground feature separation.

[0094] Based on the ground point cloud extracted after filtering, the Kriging interpolation method was used to construct a digital elevation model with a spatial resolution of 0.3 m. The original point cloud height value was subtracted from the digital elevation model to obtain normalized point cloud data without the influence of terrain undulation.

[0095] On this basis, three interpolation methods are used to generate the canopy height model (CHM), as follows:

[0096] Triangular Irregular Network (TIN): This method constructs discrete point cloud data into a triangular mesh structure and uses the geometric relationship of each triangle plane to interpolate the height of the target forest. It is suitable for high-precision simulation in complex terrain environments.

[0097] Kriging interpolation: A geostatistical interpolation method based on spatial autocorrelation. It calculates the variation function between sample points to obtain the optimal weight coefficient, and then estimates the unknown point value. It has high spatial prediction accuracy.

[0098] Inverse Distance Weighting (IDW): Based on the principle of "close similarity", it uses the inverse of the distance from the known point to the estimated point as the weight for weighted averaging. The interpolation process is simple and suitable for scenarios with evenly distributed point cloud data.

[0099] The above three methods all generate CHM models with a spatial resolution of 0.3m based on normalized point clouds. Figure 3 As shown, Figure 3 Part (a), (b), and (c) represent the CHM models obtained by using TIN based on 380, 550, and 80550, respectively. Figure 3 Part (d), (e) and (f) represent the CHM model obtained by using K at 380, 550 and 80550 respectively. Figure 3 Parts (g), (h), and (i) represent the CHM models obtained using IDW on the basis of 380, 550, and 80550, respectively.

[0100] Step 3: Comparison of various single tree segmentation methods.

[0101] To verify the effectiveness of the proposed single tree segmentation method, a total of 10 single tree segmentation methods were evaluated, which were divided into two categories:

[0102] The first category: based on the combination of CHM seed point generation and three algorithms (MWS, SRG and PCS), there are 9 methods in total: TIN-MWS, TIN-SRG, TIN-PCS, K-MWS, K-SRG, K-PCS, IDW-MWS, IDW-SRG, and IDW-PCS.

[0103] The second category: Based on the combination of RTK real positioning seed points and PCS, a method RTK-PCS is formed, which is the method proposed in this application.

[0104] (1) A dual matching strategy based on distance and height difference is used to pair the single trees extracted by the drone with the measured single trees to evaluate the segmentation accuracy:

[0105] The distance threshold is set to 3m;

[0106] The allowable range of tree height difference ΔH is classified into three levels:

[0107] Level I: H≤15m, ΔH≤3m;

[0108] Level II: 15m <H≤20m,ΔH≤4m;

[0109] Level III: H>20m, ΔH≤5m.

[0110] The matching logic is as follows: Using the measured tree as the reference, the closest extracted tree is selected for matching within the distance and height difference thresholds. If an extracted tree corresponds to multiple measured trees, the closest matching one is considered a correct segmentation, and the others are considered missed segmentations. Extracted trees that are not matched are considered over-segmented, and measured trees that are not matched are considered missed segmentations.

[0111] According to the above matching method, the recall rate r, precision rate p and overall accuracy F were calculated. The results were verified on all measured individual trees (a total of 3486 trees).

[0112]

[0113] Where r represents the ratio of correctly segmented trees to the total number of measured trees, p represents the ratio of correctly segmented trees to the total number of extracted trees, F is the weighted harmonic mean calculated by comprehensively considering r and p, TP represents the number of correctly segmented trees, FN represents the number of missed trees, and FP represents the number of over-segmented trees.

[0114] In addition, to verify the accuracy of tree height extraction, the linear regression method was used to verify the measured and extracted values ​​of tree height, and the determination coefficient (R 2 The model accuracy is evaluated using four indicators: root mean square error (RMSE), relative root mean square error (rRMSE) and mean absolute error (MAE).

[0115] The formula is as follows:

[0116]

[0117]

[0118] Where y i is the measured value of the i-th tree height, Extract the value for the i-th tree height, is the mean of the measured tree heights, and n is the number of samples.

[0119] (2) Results and analysis.

[0120] Result 1: Evaluation of segmentation effect based on CHM automatic detection of seed points.

[0121] In the absence of real tree location information, a regional maximum detection method based on CHM was used to generate seed points. Nine single tree segmentation algorithms (including TIN-MWS, TIN-SRG, and TIN-PCS based on triangulation interpolation, K-MWS, K-SRG, and K-PCS based on Kriging interpolation, and IDW-MWS, IDW-SRG, and IDW-PCS based on IDW interpolation) were used. Experiments were conducted under three lidar PRF combinations (380 kHz, 550 kHz, and their fusion of 380 and 550 kHz). A total of 27 sets of segmentation results were obtained, as shown in Table 3.

[0122] Table 3 Segmentation results

[0123]

[0124]

[0125]

[0126] As can be seen from Table 3, the segmentation effect of 380kHz data is better than other combinations. Among all the combinations, the F-value of the four combinations of 380-IDW-MWS, 380-K-MWS, 380-IDW-PCS and 380-K-PCS is not less than 0.82, the recall rate (r) ≥ 0.86, and the precision rate (p) ≥ 0.78, indicating the best segmentation effect.

[0127] The accuracy of 550kHz data is relatively low, and the F values ​​of the three algorithms, K-MWS, IDW-MWS and TIN-MWS, do not exceed 0.67 under 550kHz data.

[0128] Comprehensive analysis showed that the average F value of 380 data was 0.80, significantly higher than that of 550 (the average F value was 0.74), while the average F value of 380550 data was 0.79, which had no significant difference from 380 (P < 0.05), but was better than 550 (see Figure 4 and Table 3), Figure 4 Parts (a1), (a2), (a3), (a4), (a5), (a6), (a7), (a8), and (a9) are schematic diagrams of the effects of single tree splitting under the 380-TIN-MWS combination, 380-K-MWS combination, 380-IDW-MWS combination, 380-TIN-SRG combination, 380-K-SRG combination, 380-IDW-SRG combination, 380-TIN-PCS combination, 380-K-PCS combination, and 380-IDW-PCS combination, respectively. Figure 4Parts (b1), (b2), (b3), (b4), (b5), (b6), (b7), (b8), and (b9) are schematic diagrams of the effects of single tree splitting under the 550-TIN-MWS combination, 550-K-MWS combination, 550-IDW-MWS combination, 550-TIN-SRG combination, 550-K-SRG combination, 550-IDW-SRG combination, 550-TIN-PCS combination, 550-K-PCS combination, and 550-IDW-PCS combination, respectively. Part (c1), part (c2), part (c3), part (c4), part (c5), part (c6), part (c7), part (c8), and part (c9) are schematic diagrams of the effects of single tree splitting under the 380550-TIN-MWS combination, the 380550-K-MWS combination, the 380550-IDW-MWS combination, the 380550-TIN-SRG combination, the 380550-K-SRG combination, the 380550-IDW-SRG combination, the 380550-TIN-PCS combination, the 380550-K-PCS combination, and the 380550-IDW-PCS combination, respectively.

[0129] Result 2: Analysis of seed point segmentation results based on real-time positioning points (RTK).

[0130] To further improve the segmentation accuracy, this example introduces the single tree positioning of RTK measured trees as high-precision seed points, combines it with the high-precision PCS single tree segmentation algorithm, and constructs RTK-PCS segmentation schemes on three laser data combinations of 380, 550, and 380550. A total of three groups of results are obtained (see Table 4 and Figure 5 ) Figure 5 Parts (a), (b), and (c) represent the segmentation results obtained by constructing RTK-PCS segmentation schemes on three laser data combinations of 380, 550, and 380550, respectively.

[0131] Table 4 RTK-PCS segmentation results

[0132]

[0133] The F-values ​​of all RTK-PCS combinations are ≥0.93, among which the F-value, recall rate r, and precision rate p of the 380-RTK-PCS and 380550-RTK-PCS combinations all reach 0.96, significantly outperforming other solutions.

[0134] Compared with the CHM detection seed point method, using RTK positioning points as seed points can improve the segmentation accuracy F value from a maximum of 0.83 to 0.96, an improvement of 15.66%.

[0135] Especially in high-density natural forests, RTK seed points can effectively alleviate the problems of missed segmentation and over-segmentation caused by crown overlap in the CHM method, ensuring the accurate matching of the spatial positions of individual trees.

[0136] Result 3: The impact of segmentation results on tree height extraction accuracy.

[0137] Based on the segmentation results, tree heights were extracted from each combination and compared with ground-based tree height measurements to evaluate the impact of different segmentation methods on the accuracy of tree parameter extraction. The top five combinations with the highest segmentation accuracy in the 380 dataset (380-IDW-MWS, 380-K-MWS, 380-IDW-PCS, 380-K-PCS, and 380-RTK-PCS) were selected for tree height accuracy analysis. The results are shown in Table 5.

[0138] Table 5 Tree high-precision analysis results

[0139]

[0140] All combinations showed high consistency in tree height extraction, with a determination coefficient R 2 All are ≥0.83, RMSE is between 1.14-1.17m, relative RMSE (rRMSE) is about 4.25%-4.38%, and MAE is stable at 0.79-0.81m.

[0141] Among them, the 380-RTK-PCS combination is still the best, achieving high tree height extraction accuracy while maintaining high segmentation accuracy.

[0142] The linear fitting diagram is as follows Figure 6 As shown, Figure 6 It shows that the predicted values ​​of each combination fit well with the measured values, and the regression slope is close to 1, which verifies the effectiveness and practicality of the segmentation results. Figure 6 Part (A), (B), (C), (D), and (E) represent the linear fitting graphs under the 380-K-MWS combination, the 380-IDW-MWS combination, the 380-K-PCS combination, the 380-IDW-PCS combination, and the 380-RTK-PCS combination, respectively.

[0143] This example shows that when the LiDAR PRF is 380kHz, the point cloud quality is high and suitable for high-precision single tree segmentation. In the absence of actual measurement sites, using CHM to detect seed points and combining it with the PCS / MWS algorithm of IDW or K interpolation can achieve better segmentation results. In the presence of actual RTK positioning points, using them as seed points and combining them with the PCS algorithm can significantly improve the segmentation accuracy, which is currently the best single tree segmentation method. Using the RTK seed point solution can increase the overall accuracy F value from 0.83 to 0.96, an increase of 15.66%, significantly enhancing the accuracy and stability of tree parameter extraction.

[0144] Compared with existing methods, this application has the following significant technical advantages: Integration of high-precision single tree positioning information as seed points: By using the measured single tree position coordinates as segmentation seed points, it effectively guides the spatial division of single tree boundaries, significantly reduces the missed segmentation and over-segmentation phenomena caused by problems such as crown overlap and point cloud occlusion, and significantly improves the segmentation accuracy, which is especially suitable for forest stand environments with high density and complex spatial structure.

[0145] Build a high-quality point cloud data foundation: This application effectively improves the spatial consistency and quality of point cloud data through a systematic data preprocessing process, including boundary identification, removal of aerial points and surface outliers, resampling, denoising, ground point filtering, DEM construction and point cloud normalization, providing a solid foundation for subsequent segmentation accuracy.

[0146] A distance-discriminant clustering algorithm is used to achieve fine segmentation: During the segmentation stage, the spatial geometric features of the point cloud and the seed point information are combined, and a clustering algorithm based on spatial distance threshold is used to segment the point cloud into individual trees, which improves the algorithm's adaptability to changes in the internal structure of the forest stand.

[0147] Meeting the needs of high-precision individual identification in tree breeding: This application method is particularly suitable for applications such as tree breeding and precision forestry surveys that require extremely high accuracy of single tree structural parameters. It has high applicability, versatility and promotion value, and provides an effective solution for the refined identification of single trees under complex forest stand conditions.

[0148] In summary, compared with the existing technology, this application has significant advantages in single tree identification accuracy, processing process systematization and adaptability to complex forest stands, and has good prospects for scientific research and industrial application.

[0149] Based on the same inventive concept, embodiments of the present application also provide a UAV-based laser radar-based tree segmentation device for implementing the aforementioned UAV-based laser radar-based tree segmentation method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more UAV-based laser radar-based tree segmentation device embodiments provided below can be found in the above-described limitations of the UAV-based laser radar-based tree segmentation method and will not be further elaborated here.

[0150] In an exemplary embodiment, Figure 7 As shown, a single tree segmentation device based on a UAV laser radar is provided, comprising: a point cloud acquisition module for acquiring UAV laser radar point cloud data of a target forest; the target forest includes all single trees to be segmented;

[0151] The positioning module is used to obtain the position coordinates of each single tree to be segmented in the target forest using carrier phase dynamic real-time differential positioning technology;

[0152] The ground point cloud extraction module is used to process the UAV lidar point cloud data of the target forest farm to obtain the UAV lidar point cloud data belonging to the ground;

[0153] A normalization module is used to use the UAV lidar point cloud data belonging to the ground to normalize the UAV lidar point cloud data of the target forest farm, thereby obtaining normalized UAV lidar point cloud data of the target forest farm;

[0154] The segmentation module is used to cluster the normalized UAV lidar point cloud data of the target forest farm using the position coordinates of each single tree to be segmented in the target forest farm as the position coordinates of the initial seed point, and adopt a clustering algorithm based on point cloud distance discrimination to achieve single tree segmentation.

[0155] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store single tree segmentation data based on the UAV laser radar. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a single tree segmentation method based on the UAV laser radar is implemented.

[0156] Those skilled in the art will understand that Figure 8 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the above-mentioned method embodiments when executing the computer program.

[0157] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the above-mentioned method embodiments when executed by a processor.

[0158] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the above method embodiments are implemented.

[0159] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0160] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0161] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0162] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0163] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A single tree segmentation method based on UAV laser radar, characterized in that: The single tree segmentation method based on UAV laser radar includes: Obtain UAV LiDAR point cloud data of the target forest; the target forest includes all the individual trees to be segmented; The carrier phase dynamic real-time differential positioning technology is used to obtain the position coordinates of each single tree to be split in the target forest; Process the UAV LiDAR point cloud data of the target forest farm to obtain the UAV LiDAR point cloud data belonging to the ground; The UAV lidar point cloud data belonging to the ground is used to normalize the UAV lidar point cloud data of the target forest farm to obtain the normalized UAV lidar point cloud data of the target forest farm; Taking the position coordinates of each single tree to be segmented in the target forest as the position coordinates of the initial seed point, a clustering algorithm based on point cloud distance discrimination is used to cluster the normalized UAV lidar point cloud data of the target forest to achieve single tree segmentation.

2. The single tree segmentation method based on UAV laser radar according to claim 1 is characterized in that: The UAV LiDAR point cloud data of the target forest farm is processed to obtain the UAV LiDAR point cloud data belonging to the ground, including: The improved progressive encryption triangulation filtering algorithm is used to extract ground points from the UAV lidar point cloud data of the target forest farm to obtain the UAV lidar point cloud data belonging to the ground.

3. The single tree segmentation method based on drone laser radar according to claim 2 is characterized in that: The improved progressive encryption triangulation filtering algorithm is used to extract ground points from the UAV lidar point cloud data of the target forest farm, and the UAV lidar point cloud data belonging to the ground is obtained, including: Determine the stand boundary according to the position coordinates of each single tree to be divided in the target forest; Determine the study area within the target forest farm based on the stand boundaries; The UAV LiDAR point cloud data of the target forest farm was cropped to obtain the UAV LiDAR point cloud data of the study area; The UAV lidar point cloud data of the study area is subjected to outlier removal, resampling, and noise point removal operations in sequence to obtain the pre-processed UAV lidar point cloud data of the study area; The improved progressive encryption triangulation filtering algorithm is used to extract ground points from the UAV lidar point cloud data after preprocessing in the study area to obtain the UAV lidar point cloud data belonging to the ground.

4. The single tree segmentation method based on UAV laser radar according to claim 1 is characterized in that: The UAV LiDAR point cloud data belonging to the ground is used to normalize the UAV LiDAR point cloud data of the target forest farm to obtain the normalized UAV LiDAR point cloud data of the target forest farm, specifically including: Build a digital elevation model based on the ground-based UAV lidar point cloud data; The UAV lidar point cloud data of the target forest farm is normalized using the digital elevation model to obtain the normalized UAV lidar point cloud data of the target forest farm.

5. The single tree segmentation method based on UAV laser radar according to claim 4 is characterized in that: The digital elevation model is used to normalize the UAV lidar point cloud data of the target forest farm to obtain the normalized UAV lidar point cloud data of the target forest farm. Specifically, the digital elevation model is subtracted from the UAV lidar point cloud data of the target forest farm to obtain the normalized UAV lidar point cloud data of the target forest farm.

6. The single tree segmentation method based on UAV laser radar according to claim 1 is characterized in that: The position coordinates of each single tree to be segmented in the target forest are used as the position coordinates of the initial seed points. A clustering algorithm based on point cloud distance discrimination is used to cluster the normalized UAV lidar point cloud data of the target forest to achieve single tree segmentation. Specifically, the algorithm includes: For any single tree to be segmented, the position coordinates of the single tree to be segmented are used as the position coordinates of the initial seed point, and the position coordinates of the target seed point are determined according to the position coordinates of the initial seed point; Determine whether the distance between the target point and the target seed point is less than a preset spatial spacing threshold to obtain a first judgment result; the target point is point cloud data other than the target seed point in the normalized UAV lidar point cloud data of the target forest farm; If the first judgment result is yes, then determining whether the distance between the target point and the target seed point meets the minimum spacing rule to obtain a second judgment result; If the second judgment result is yes, it is determined that the target point and the target seed point belong to the same single tree to be segmented, thereby achieving single tree segmentation.

7. A single wood splitting device based on drone laser radar, characterized in that: The single wood splitting device based on the UAV laser radar includes: The point cloud acquisition module is used to obtain the UAV lidar point cloud data of the target forest; the target forest includes all the individual trees to be segmented; The positioning module is used to obtain the position coordinates of each single tree to be segmented in the target forest using carrier phase dynamic real-time differential positioning technology; The ground point cloud extraction module is used to process the UAV lidar point cloud data of the target forest farm to obtain the UAV lidar point cloud data belonging to the ground; A normalization module is used to use the UAV lidar point cloud data belonging to the ground to normalize the UAV lidar point cloud data of the target forest farm, thereby obtaining normalized UAV lidar point cloud data of the target forest farm; The segmentation module is used to cluster the normalized UAV lidar point cloud data of the target forest farm using the position coordinates of each single tree to be segmented in the target forest farm as the position coordinates of the initial seed point, and adopt a clustering algorithm based on point cloud distance discrimination to achieve single tree segmentation.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the single-tree segmentation method based on a UAV laser radar 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, the single tree segmentation method based on UAV laser radar described in any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the single tree segmentation method based on UAV laser radar described in any one of claims 1 to 6 is implemented.

Images