Method, system, device and storage medium for identifying single understory saplings

Abstract

The present invention discloses a method, system, device and storage medium for identifying single young trees under forest, relating to the technical field of lidar point cloud processing. The method includes performing single tree segmentation on the initial airborne lidar point cloud, determining the airborne lidar point cloud determined as large trees based on the initial airborne lidar point cloud, and determining the initial structural parameters of the large trees according to the airborne lidar point cloud of each large tree; determining the initial reference single tree point set based on the airborne lidar point cloud of all large trees, and performing secondary segmentation on the remaining airborne lidar point cloud to obtain the secondary segmentation lidar point cloud; performing iterative clustering screening on the secondary segmentation lidar point cloud to obtain the second reference single tree point set and the remaining single tree point set; screening again to obtain the updated structural parameters; obtaining the airborne lidar point cloud to be segmented based on the initial airborne lidar point cloud and obtaining the segmentation parameters of the young trees; and determining the young trees in the current forest according to the airborne lidar point cloud to be segmented and the segmentation parameters. The present invention realizes the identification of young trees in the forest.

Description

Technical Field

[0001] The present invention relates to the technical field of lidar point cloud processing, and particularly to a method, system, device and storage medium for identifying individual young trees under forest canopy. Background Art

[0002] Young trees under forest canopy play a crucial role in promoting the growth, regeneration and succession of forest communities, maintaining forest species diversity, and balancing and realizing the functions of forest ecosystems. Clarifying the spatial distribution and structural characteristics of young trees under forest canopy is of great significance for studying the impact of canopy structural characteristics and dynamic changes in understory light environment on the growth of young trees, and further exploring the internal mechanism of understory plant regeneration and succession. Airborne Laser Scanning (ALS) can obtain large-scale lidar point clouds, and its high penetrability can accurately depict the vertical structure of forests. The individual tree segmentation of ALS point clouds is the basis for extracting the spatial structural characteristics of individual trees, so as to accurately evaluate the growth trend and ecological functions of forest stands. However, limited by the relatively sparse point cloud density of ALS and the high canopy closure of forests, existing methods only focus on the study of the single-tree structure in the upper layer of forests, pay less attention to the structural characteristics of young trees under forest canopy, and lack the identification of young trees under forest canopy. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, system, device and storage medium for identifying individual young trees under forest canopy, which realizes the identification of young trees under forest canopy.

[0004] To achieve the above purpose, the present invention provides the following solutions:

[0005] A method for identifying individual young trees under forest canopy, the method comprising:

[0006] Obtain the initial airborne lidar point cloud of each individual tree in the current forest;

[0007] Use the spectral clustering method to perform individual tree segmentation on the initial airborne lidar point cloud, determine the airborne lidar point cloud with a height greater than 5m as the airborne lidar point cloud of each big tree, and determine the initial structural parameters of each big tree according to the airborne lidar point cloud of each big tree; the structural parameters include: tree height, crown width, position, centroid, lowest point height of the tree crown, and the number of single-tree point clouds;

[0008] Determine the airborne lidar point clouds of all big trees that meet the preset single-tree conditions as the initial reference single-tree point set, and use the bidirectional projection method to perform secondary segmentation on the airborne lidar point clouds of big trees that do not meet the preset single-tree conditions to obtain the secondary segmentation lidar point cloud; the preset single-tree conditions are determined according to the initial structural parameters;

[0009] Using the mean shift clustering algorithm, iteratively cluster the secondary segmented laser point cloud to obtain a secondary segmented single-tree point set with class labels, where the secondary segmented single-tree point set with class labels includes multiple single-tree points with class labels;

[0010] Based on the preset single-tree conditions and preset distance conditions, screen the secondary segmented single-tree point set with class labels, and add the points in the secondary segmented single-tree point set with class labels that meet the preset single-tree conditions or the preset distance conditions to the initial reference single-tree point set to obtain a second reference single-tree point set, and determine the secondary segmented single-tree point set that does not meet the preset single-tree conditions and the preset distance conditions as the remaining single-tree point set;

[0011] Based on the second reference single-tree point set and the remaining single-tree point set, construct the local K-nearest neighbor point set of each remaining single-tree point in the remaining single-tree point set, determine the proportion of the number of points in the second reference single-tree point set in the local K-nearest neighbor point set to the total number of points in the local K-nearest neighbor point set, and add the remaining single-tree points corresponding to the local K-nearest neighbor point sets with a proportion greater than 0.5 to the second reference single-tree point set to obtain a third reference single-tree point set, and determine the updated structural parameters of each big tree based on the airborne laser point cloud of each big tree in the third reference single-tree point set;

[0012] Remove the airborne laser point cloud with the updated lowest crown height of each big tree greater than 5m from the initial airborne laser point cloud to obtain a secondarily removed airborne laser point cloud;

[0013] Remove the airborne laser point cloud in the preset cylindrical buffer area from the secondarily removed airborne laser point cloud to obtain the airborne laser point cloud to be segmented; the bottom surface of the preset cylindrical buffer area is a circle with the updated position as the center and a radius of 0.8m, and the height of the preset cylindrical buffer area is the updated lowest crown height;

[0014] Adopt the Delaunay triangulation interpolation algorithm to generate a young tree crown layer height model according to the airborne laser point cloud to be segmented, and adopt a fixed window to detect the local maximum value in the young tree crown layer height model to obtain the segmentation parameters of the young trees, where the segmentation parameters include: position, local plant density, and crown width;

[0015] Using the mean shift algorithm, determine the young trees in the current forest according to the airborne laser point cloud to be segmented and the segmentation parameters.

[0016] Optionally, the preset single-tree conditions include:

[0017] The tree height is greater than 5m;

[0018] The range of the ratio of the east-west radius to the north-south radius of the crown width is: 0.7 - 1.3;

[0019] The centroid is within the crown width range;

[0020] The number of the individual tree point clouds is greater than half of the average value of the number of all individual tree point clouds.

[0021] Optionally, the preset distance condition includes:

[0022] The distance between the individual tree composed of the points with the same class label in the secondarily segmented individual tree point cloud with class labels and the first reference individual tree is less than the crown width of the first reference individual tree, and the first reference individual tree is the individual tree composed of the points in the initial reference individual tree point set.

[0023] An identification system for understory saplings, the system includes:

[0024] An initial point cloud acquisition module, configured to acquire the initial airborne laser point cloud of each individual tree in the current forest;

[0025] A first segmentation module, configured to perform individual tree segmentation processing on the initial airborne laser point cloud by using the spectral clustering method, determine the airborne laser point cloud with a height greater than 5m as the airborne laser point cloud of each big tree, and determine the initial structural parameters of each big tree according to the airborne laser point cloud of each big tree; the structural parameters include: tree height, crown width, position, centroid, the height of the lowest point of the tree crown, and the number of individual tree point clouds;

[0026] A second segmentation module, configured to determine the airborne laser point cloud of all big trees that meet the preset individual tree conditions as the initial reference individual tree point set, and adopt the bidirectional projection method to perform secondary segmentation on the airborne laser point cloud of the big trees that do not meet the preset individual tree conditions to obtain the secondarily segmented laser point cloud; the preset individual tree conditions are determined according to the initial structural parameters;

[0027] A clustering module, configured to perform iterative clustering on the secondarily segmented laser point cloud by using the mean shift clustering algorithm to obtain a secondarily segmented individual tree point cloud with class labels, and the secondarily segmented individual tree point cloud with class labels includes a plurality of individual tree points with class labels;

[0028] A screening module, configured to screen the secondarily segmented individual tree point cloud with class labels based on the preset individual tree conditions and the preset distance condition, add the points in the secondarily segmented individual tree point cloud with class labels that meet the preset individual tree conditions or the preset distance condition to the initial reference individual tree point set to obtain a second reference individual tree point set, and determine the secondarily segmented individual tree point cloud that does not meet the preset individual tree conditions and the preset distance condition as the remaining individual tree point set;

[0029] A structure parameter update module, which is used to construct a local K-nearest neighbor point set for each remaining individual tree point in the remaining individual tree point set based on the second reference individual tree point set and the remaining individual tree point set, determine the proportion of the number of points in the second reference individual tree point set in the local K-nearest neighbor point set to the total number of points in the local K-nearest neighbor point set, add the remaining individual tree points corresponding to the local K-nearest neighbor point sets with a proportion greater than 0.5 to the second reference individual tree point set to obtain a third reference individual tree point set, and determine the updated structure parameters of each big tree based on the airborne laser point cloud of each big tree in the third reference individual tree point set;

[0030] A secondary removal of airborne laser point cloud determination module, which is used to remove the airborne laser point cloud with the updated lowest crown height of each big tree greater than 5m from the initial airborne laser point cloud to obtain a secondary removed airborne laser point cloud;

[0031] A to-be-segmented airborne laser point cloud determination module, which is used to remove the airborne laser point cloud in the preset cylindrical buffer area from the secondary removed airborne laser point cloud to obtain a to-be-segmented airborne laser point cloud; the bottom surface of the preset cylindrical buffer area is a circle with the updated position as the center and a radius of 0.8m, and the height of the preset cylindrical buffer area is the updated lowest crown height;

[0032] A segmentation parameter determination module, which is used to adopt the Delaunay triangulation interpolation algorithm to generate a young tree crown layer height model according to the to-be-segmented airborne laser point cloud, and adopt a fixed window to detect local maxima in the young tree crown layer height model to obtain segmentation parameters of young trees, where the segmentation parameters include: position, local plant density, and crown width;

[0033] A young tree determination module, which is used to use the mean shift algorithm to determine the young trees in the current forest according to the to-be-segmented airborne laser point cloud and the segmentation parameters.

[0034] A device, comprising:

[0035] One or more processors;

[0036] A storage device, on which one or more programs are stored;

[0037] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0038] A storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the method as described above.

[0039] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] The present invention discloses a method, system, device and storage medium for identifying single understory saplings. First, the NSC segmentation results are optimized through multiple screenings, solving the problems of over-segmentation, under-segmentation and mis-segmentation in the existing segmentation results, achieving an accurate division of the upper-layer big tree canopy, and providing a good data basis for the accurate segmentation of the lower-layer saplings. Secondly, considering the cross-planting strategy of the upper-layer big trees and the lower-layer saplings, and using the spatial distribution difference between the two, the point clouds of the upper-layer big trees existing in the point clouds of the lower-layer saplings are removed to obtain the airborne lidar point clouds after secondary removal. Finally, the saplings are determined according to the airborne lidar point clouds after secondary removal, realizing the accurate extraction of the saplings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0042] Figure 1 It is a schematic flowchart of the method for identifying single understory saplings provided in Embodiment 1 of the present invention;

[0043] Figure 2 It is a schematic structural diagram of the system for identifying single understory saplings provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0045] The purpose of the present invention is to provide a method, system, device and storage medium for identifying single understory saplings, aiming to realize the identification of understory saplings.

[0046] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0047] Embodiment 1

[0048] Figure 1 It is a schematic flowchart of the method for identifying single understory saplings provided in Embodiment 1 of the present invention. As Figure 1 shown, the method for identifying single understory saplings in this embodiment includes:

[0049] Step 101: Obtain the initial airborne laser point cloud of each individual tree in the current forest.

[0050] After step 101, it further includes: classifying the initial airborne laser point cloud using a cloth filtering algorithm to determine ground points and non-ground points, then generating a digital surface model of the forest area by Delaunay triangulation interpolation based on the ground points, and normalizing the initial laser point cloud based on this model.

[0051] Step 102: Based on the initial airborne laser point cloud, use the spectral clustering method to determine the airborne laser point cloud and initial structural parameters of each big tree.

[0052] Step 102 specifically includes: using the spectral clustering method to perform individual tree segmentation on the initial airborne laser point cloud, determining the airborne laser point cloud of each big tree with a height greater than 5m, and determining the initial structural parameters of each big tree according to the airborne laser point cloud of each big tree; the structural parameters include: tree height, crown width, position, centroid, height of the lowest point of the crown, and the number of individual tree point clouds.

[0053] Step 103: Based on the airborne laser point clouds of all big trees, determine the initial reference individual tree point set and the secondarily segmented laser point cloud.

[0054] Step 103 specifically includes: determining the airborne laser point clouds of all big trees that meet the preset individual tree conditions as the initial reference individual tree point set, and using the bidirectional projection method to secondarily segment the airborne laser point clouds of big trees that do not meet the preset individual tree conditions to obtain the secondarily segmented laser point cloud; the preset individual tree conditions are determined according to the initial structural parameters.

[0055] Specifically, the preset individual tree conditions include:

[0056] Tree height is greater than 5m;

[0057] The ratio range of the east-west radius to the north-south radius of the crown width is: 0.7 - 1.3.

[0058] The centroid is within the crown width.

[0059] The number of individual tree point clouds is greater than half of the average value of the number of all individual tree point clouds.

[0060] Step 104: Perform iterative clustering on the secondarily segmented laser point cloud to obtain the secondarily segmented individual tree point set with class labels.

[0061] Step 104 specifically includes: using the mean shift clustering algorithm to perform iterative clustering on the secondarily segmented laser point cloud to obtain the secondarily segmented individual tree point set with class labels, and the secondarily segmented individual tree point set with class labels includes multiple individual tree points with class labels.

[0062] Among them, the points with the same class label in the secondary segmented individual tree point set with class labels are regarded as the same individual tree.

[0063] Step 105: Based on the secondary segmented individual tree point set with class labels, determine the second reference individual tree point set and the remaining individual tree point set.

[0064] Step 105 specifically includes: screening the secondary segmented individual tree point set with class labels based on a preset individual tree condition and a preset distance condition, adding the points in the secondary segmented individual tree point set with class labels that meet the preset individual tree condition or the preset distance condition to the initial reference individual tree point set to obtain the second reference individual tree point set, and determining the remaining individual tree point set as the secondary segmented individual tree point set that does not meet the preset individual tree condition and the preset distance condition.

[0065] Specifically, the preset distance condition includes:

[0066] The distance between the individual tree composed of the points with the same class label in the secondary segmented individual tree point set with class labels and the first reference individual tree is less than the crown width of the first reference individual tree, and the first reference individual tree is the individual tree composed of the points in the initial reference individual tree point set.

[0067] Step 106: Based on the second reference individual tree point set and the remaining individual tree point set, determine the third reference individual tree point set and the updated structural parameters of each big tree.

[0068] Step 106 specifically includes: constructing the local K-nearest neighbor point set of each remaining individual tree point in the remaining individual tree point set based on the second reference individual tree point set and the remaining individual tree point set, determining the proportion of the number of points in the second reference individual tree point set in the local K-nearest neighbor point set to the total number of points in the local K-nearest neighbor point set, adding the remaining individual tree points corresponding to the local K-nearest neighbor point sets with a proportion greater than 0.5 to the second reference individual tree point set to obtain the third reference individual tree point set, and determining the updated structural parameters of each big tree based on the airborne laser point cloud of each big tree in the third reference individual tree point set.

[0069] Step 107: Remove the airborne laser point cloud with the updated lowest crown point height of each big tree greater than 5m from the initial airborne laser point cloud to obtain the second removed airborne laser point cloud.

[0070] Step 108: Determine the airborne laser point cloud to be segmented based on the second removed airborne laser point cloud.

[0071] Step 108 specifically includes: removing the airborne laser point cloud in the preset cylindrical buffer zone from the second removed airborne laser point cloud to obtain the airborne laser point cloud to be segmented; the bottom surface of the preset cylindrical buffer zone is a circle with the updated position as the center and a radius of 0.8m, and the height of the preset cylindrical buffer zone is the updated lowest crown point height.

[0072] Step 109: Determine the segmentation parameters of saplings based on the airborne lidar point cloud to be segmented.

[0073] Step 109 specifically includes: Using the Delaunay triangulation interpolation algorithm, generating a sapling canopy height model according to the airborne lidar point cloud to be segmented, and using a fixed window to detect the local maximum values in the sapling canopy height model to obtain the segmentation parameters of the saplings. The segmentation parameters include: position, local plant density, and crown width.

[0074] The sapling canopy height model is a surface model expressing the height of saplings from the ground, which can provide the horizontal and vertical distribution of the sapling canopy.

[0075] Step 110: Use the mean shift algorithm to determine the saplings in the current forest according to the airborne lidar point cloud to be segmented and the segmentation parameters.

[0076] The method in the present invention specifically includes two major steps: precise segmentation of the upper canopy and extraction of individual saplings under the forest.

[0077] Among them, the precise segmentation of the upper canopy includes:

[0078] The segmentation of the upper single-tree canopy adopts the -based spectral clustering ( NSC) method. This method can efficiently complete the single-tree segmentation of large-scale ALS point clouds and estimate single-tree parameters. However, for complex forest stand conditions, there are still problems such as mis-segmentation caused by the too-close horizontal or vertical distance between adjacent points that do not belong to the same single tree in the segmentation results. These segmentation errors will further affect the extraction accuracy of the lower-layer saplings. Therefore, the present invention proposes a refined post-processing method for the segmentation results to further improve the accuracy of the upper canopy segmentation. This method mainly includes two parts: preliminary determination of the reference single-tree point set and hierarchical iterative extraction of single-tree canopies. The second part is completed in three steps, including hierarchical fusion screening, hierarchical clustering screening, and backward selection processing of the discrete remaining point set.

[0079] (1) Preliminary determination of the reference single-tree point set

[0080] First, the NSC algorithm processes the high-density airborne lidar point cloud to obtain the preliminarily segmented single trees. A single tree refers to the lidar point cloud of a single tree obtained by the NSC single-tree segmentation algorithm and the obtained single-tree parameters (including tree height and crown width). Set the single-tree shape factor to screen the results (that is, the results obtained by using the NSC algorithm for single-tree segmentation (that is, the lidar point cloud of a single tree)). Those that meet the screening conditions are the reference single-tree point set. For the single trees that do not meet the conditions, further segmentation processing is carried out using the bidirectional projection method. The specific process is as follows:

[0081] Project each non - compliant single tree onto the horizontal plane and construct a two - dimensional minimum bounding box from this. In this way, the direction lines L1 and L2 with the maximum and minimum crown widths of the single tree can be obtained. L1 and L2 respectively form two projection elevations in directions orthogonal to the horizontal plane (L1 forms one projection elevation and L2 forms one projection elevation); Project the three - dimensional single - tree point cloud (the "three - dimensional laser point cloud" refers to the three - dimensional laser point cloud of a single tree, which is obtained by segmenting the original laser point cloud using the NSC single - tree segmentation algorithm. The processing steps here are for the three - dimensional laser point cloud of each single tree obtained after NSC single - tree segmentation) onto the two elevations respectively to form two sets of secondary segmentation reference points P1 and P2 for the single tree. Apply the Alpha Shape boundary detection algorithm combined with threshold conditions to P1 and P2 respectively to obtain the canopy boundary points of the two point sets. At this time, the iterative adaptive point algorithm can be used to detect inflection points for each boundary point set. If there are no inflection points, directly classify this single tree into the remaining single - tree point set; if there are inflection points, which are the secondary segmentation projection points of the single tree in the two elevation directions (the single tree is projected onto the two elevations respectively, and the detected inflection points form the projection points of the secondary segmentation), pass through each inflection point (and form a segmentation plane with the plane orthogonal to the projection elevation. Based on these segmentation planes, the three - dimensional single - tree point cloud can be processed for secondary segmentation in two directions to obtain the laser point cloud of each single tree after processing).

[0082] At this time, use the same screening rules to screen the results of the secondary segmentation. Add the single trees that meet the screening conditions to the reference single - tree point set, and those that do not meet the screening conditions are the remaining single - tree point set. In the present invention, the same screening rules are adopted for all screening processes involving the reference single - tree point set, specifically including the following four conditions:

[0083] 1) The height of the single tree is greater than 5m;

[0084] 2) The ratio of the east - west crown width to the north - south crown width radius of the single tree is greater than scale1 and less than scale2, where the empirical values of scale1 and scale2 are 0.7 and 1.3 respectively;

[0085] 3) The centroid of the single tree is within the crown width range;

[0086] 4) The number of single - tree points is greater than num_mean / 2, where num_mean is the average number of single - tree points, calculated from the reference single - tree point set obtained in the previous step of screening.

[0087] (2) Hierarchical iterative extraction of the single - tree crown

[0088] After determining the initial reference single - tree point set, the remaining single - tree point set can be processed based on this point set. It specifically includes two iterative processing processes, namely hierarchical fusion screening and hierarchical clustering screening.

[0089] A. Hierarchical fusion screening:

[0090] In the specific calculation process, taking the positions of individual trees in the reference individual tree point set as references and the crown radius of each individual tree as the distance condition, calculate the distances from each individual tree in the remaining individual tree point set to each reference individual tree. Merge the individual trees within the crown radius of the reference individual tree (merge the laser point clouds of the individual trees within the crown radius with the laser point cloud of the reference individual tree onto the reference individual tree), and then screen and update the reference individual trees based on the screening criteria in part (1). Iterate continuously until there are no remaining individual trees that can be merged, and the iteration terminates.

[0091] B. Hierarchical clustering screening:

[0092] If there are still remaining individual trees that cannot be merged after hierarchical fusion screening, the hierarchical clustering screening method is used for further clustering. Since most of the qualified individual trees have been obtained through the process of hierarchical fusion screening. If there are remaining individual trees at this time, the distances between individual trees are relatively far and the data volume is small. Therefore, the mean shift algorithm is directly used for clustering (the purpose of clustering is to further process the remaining individual trees that cannot be merged in the above steps to obtain the three-dimensional laser point cloud of each remaining individual tree) and combined with the screening rules for processing, and the qualified reference individual trees can be obtained relatively quickly and added to the reference individual tree point set.

[0093] C. Backward selection of discrete remaining points:

[0094] Since relatively strict fusion rules are set in the process of fusing the remaining points with the reference individual trees, after the above steps of processing, there may still be individual trees or discrete points that cannot be merged.

[0095] At this time, construct the local K-nearest neighbor point set P of each discrete point in the remaining point set (for each discrete point, select the K points closest to it from all candidate individual tree points to form the point set P), and judge the probability of each point belonging to each candidate individual tree point set through the distance criterion, so as to reassign it: (a) If the point set P contains reference individual trees (one or more), then count the proportion of each reference individual tree point in P. If the number of reference individual tree points with the highest proportion accounts for more than 50% of the point set P, then merge this point onto this reference individual tree; otherwise, do not merge; (b) If the point set P does not contain reference individual trees, it is regarded as a noise point and not classified. (Classify the individual trees that do not meet the above various requirements as noise points and remove them).

[0096] The extraction of understory saplings includes:

[0097] Based on the accurate segmentation results of the upper tree crowns, combined with the height of the lowest point of the tree crown obtained through calculation, the tree crown point clouds of each large tree are removed from the original high-density point cloud (the original high-density point cloud is the initial input data); then, combined with the single-tree position coordinates of each large tree obtained from the upper-layer single-tree segmentation, a corresponding horizontal buffer zone is set (the buffer threshold is set to 0.8 m according to experience), and the large-tree trunk point clouds mixed in the sapling point cloud (the most original input "high-density airborne laser point cloud" includes large-tree point clouds and sapling point clouds (i.e., all the point clouds in the study area)) are removed. Although the large-tree laser point clouds in the upper part and inside of the saplings are basically removed by the above method, due to the inaccuracy of the trunk position caused by the single-tree segmentation error and uneven crowns of the upper tree crowns, as well as the influence of understory shrubs and grasses, there are still some discrete points, which will cause great interference to the extraction of saplings. Therefore, the present invention further removes the outlier points by using the point cloud filtering function of the LastTools software.

[0098] The high-density point cloud with the upper tree crown point clouds removed can be used to generate the CHM (Canopy Height Model) of the saplings. However, when there are ground and tree crown edges in the laser point cloud, invalid values (invalid values are abnormal values, and the latter half of the sentence is the explanation of the invalid values) usually occur, which will cause unnatural black holes or gray holes in the tree crown area of the original CHM. These invalid values not only affect the accuracy of single-tree extraction but also lead to unreliable single-tree parameter estimation. For this reason, an invalid value filling method based on morphological tree crown control is further used to optimize the quality of the original CHM. This method can effectively distinguish the holes and tree crown gaps in the CHM, overcoming the problem of a large amount of information loss caused by the previous global smoothing method, and specifically includes three steps: First, the Laplace operator is applied to the original CHM to determine the possible invalid values; then, based on the assumption that the tree crown is approximately circular in the vertical view (generally speaking, looking down from above, the tree crown is generally circular), a morphological closing operator is proposed to restore the tree crown coverage; finally, by comprehensively considering the above results, corresponding threshold conditions are set according to the definition of the invalid values and the crown base height to determine the invalid values in the CHM, and the corresponding values in the CHM obtained by the median filtering method are used to replace the invalid values (by processing the canopy height model (CHM) of the saplings using the median filtering algorithm, the values at the abnormal value positions can be calculated, and using these values to replace the invalid values completes the filling of the abnormal values).

[0099] Due to the relatively large planting spacing of the understory saplings and the small range of the point cloud of each individual tree, the mean-shift algorithm is used to cluster the sapling point cloud on the basis of removing the big trees and outliers, so as to obtain the point cloud of each individual tree and calculate the corresponding structural parameters. However, the uneven sapling density and different crown sizes make it difficult to achieve an ideal segmentation effect with a fixed Gaussian kernel bandwidth. Therefore, the present invention intends to consider the spatial structure characteristics of individual trees and comprehensively consider the position and crown size of individual trees to obtain a locally adaptive kernel bandwidth from the optimized CHM. First, all local maxima are detected in the optimized CHM using a 3×3 fixed window as the reference positions of individual trees. For each reference tree, the average distance between it and the nearest K trees can be calculated to represent its local density. Secondly, based on the reference tree positions, the boundary points of each reference tree are detected from the optimized CHM, and the circumcircle of the tree is obtained based on the boundary points. Finally, the radius of the circumcircle represents the crown size at the position of the reference tree. Theoretically, the local kernel bandwidth should be proportional to the crown size of each sapling and inversely proportional to the local density. Therefore, the present invention defines the adaptive kernel bandwidth as:

[0100]

[0101] In the formula, is the local kernel bandwidth, radius m represents the crown size at the position of the m-th reference tree, sd m is the local density at the position of the m-th reference tree. and μ are the control factors of radius m and sd m respectively, and empirical values can be obtained by adjusting according to the actual situation.

[0102] In the process of clustering the point cloud using the mean-shift algorithm, the tree closest to the clustering center of each tree is selected from the candidate tree point set, and the kernel bandwidth at the reference tree position is added to the Gaussian kernel for clustering, and finally an accurate segmentation result of the understory sapling trees is obtained.

[0103] Embodiment 2

[0104] Figure 2 is a schematic structural diagram of the system for identifying individual understory saplings provided in Embodiment 2 of the present invention. As Figure 2 shown, the system for identifying individual understory saplings in this embodiment includes:

[0105] An initial point cloud acquisition module 201, configured to acquire the initial airborne laser point cloud of each tree in the current forest.

[0106] The first segmentation module 202 is configured to perform single-tree segmentation on the initial airborne lidar point cloud using spectral clustering method, determine the airborne lidar point cloud with a height greater than 5m as the airborne lidar point cloud of each big tree, and determine the initial structural parameters of each big tree according to the airborne lidar point cloud of each big tree; the structural parameters include: tree height, crown width, position, centroid, the lowest point height of the tree crown, and the number of single-tree point clouds.

[0107] The second segmentation module 203 is configured to determine the airborne lidar point cloud of all big trees that meet the preset single-tree conditions as the initial reference single-tree point set, and use the bidirectional projection method to perform secondary segmentation on the airborne lidar point cloud of big trees that do not meet the preset single-tree conditions to obtain the secondary segmentation lidar point cloud; the preset single-tree conditions are determined according to the initial structural parameters.

[0108] The clustering module 204 is configured to perform iterative clustering on the secondary segmentation lidar point cloud using the mean shift clustering algorithm to obtain a secondary segmentation single-tree point set with class labels, and the secondary segmentation single-tree point set with class labels includes multiple single-tree points with class labels.

[0109] The screening module 205 is configured to screen the secondary segmentation single-tree point set with class labels based on the preset single-tree conditions and the preset distance conditions, add the points in the secondary segmentation single-tree point set with class labels that meet the preset single-tree conditions or the preset distance conditions to the initial reference single-tree point set to obtain the second reference single-tree point set, and determine the secondary segmentation single-tree point set that does not meet the preset single-tree conditions and the preset distance conditions as the remaining single-tree point set.

[0110] The structural parameter update module 206 is configured to construct a local K-nearest neighbor point set of each remaining single-tree point in the remaining single-tree point set based on the second reference single-tree point set and the remaining single-tree point set, determine the proportion of the number of points in the second reference single-tree point set in the local K-nearest neighbor point set to the total number of points in the local K-nearest neighbor point set, add the remaining single-tree points corresponding to the local K-nearest neighbor point sets with a proportion greater than 0.5 to the second reference single-tree point set to obtain the third reference single-tree point set, and determine the updated structural parameters of each big tree based on the airborne lidar point cloud of each big tree in the third reference single-tree point set.

[0111] The secondary removal airborne lidar point cloud determination module 207 is configured to remove the airborne lidar point cloud with the updated lowest point height of the tree crown of each big tree greater than 5m from the initial airborne lidar point cloud to obtain the secondary removal airborne lidar point cloud.

[0112] The airborne laser point cloud to be segmented determination module 208 is used to remove the airborne laser point cloud in the preset cylindrical buffer area from the secondary removed airborne laser point cloud, so as to obtain the airborne laser point cloud to be segmented; the bottom surface of the preset cylindrical buffer area is a circle with the updated position as the center and a radius of 0.8 m, and the height of the preset cylindrical buffer area is the lowest height of the updated tree crown.

[0113] The segmentation parameter determination module 209 is used to generate a young tree crown layer height model according to the airborne laser point cloud to be segmented by using the Delaunay triangulation interpolation algorithm, and use a fixed window to detect the local maximum value in the young tree crown layer height model to obtain the segmentation parameters of the young trees. The segmentation parameters include: position, local plant density, and crown width.

[0114] The young tree determination module 210 is used to determine the young trees in the current forest by using the mean shift algorithm according to the airborne laser point cloud to be segmented and the segmentation parameters.

[0115] Embodiment 3

[0116] A device includes:

[0117] One or more processors.

[0118] A storage device on which one or more programs are stored.

[0119] When one or more programs are executed by one or more processors, the one or more processors are caused to implement the method as in Embodiment 1.

[0120] Embodiment 4

[0121] A storage medium on which a computer program is stored, wherein when the computer program is executed by a processor, the method as in Embodiment 1 is implemented.

[0122] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0123] In this article, specific examples are used to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A method for identifying single young trees under forest, characterized in that, The method includes: Obtaining the initial airborne laser point cloud of each individual tree in the current forest; Using the spectral clustering method to perform individual tree segmentation on the initial airborne laser point cloud, determining the airborne laser point cloud with a height greater than 5m as the airborne laser point cloud of each large tree, and determining the initial structural parameters of each large tree according to the airborne laser point cloud of each large tree; the structural parameters include: tree height, crown width, position, centroid, the height of the lowest point of the tree crown, and the number of individual tree point clouds; Determining the airborne laser point clouds of all large trees that meet the preset individual tree conditions as the initial reference individual tree point set, and using the bidirectional projection method to perform secondary segmentation on the airborne laser point clouds of large trees that do not meet the preset individual tree conditions to obtain the secondary segmented laser point cloud; the preset individual tree conditions are determined according to the initial structural parameters; Using the mean shift clustering algorithm to perform iterative clustering on the secondary segmented laser point cloud to obtain a secondary segmented individual tree point set with class labels, and the secondary segmented individual tree point set with class labels includes multiple individual tree points with class labels; Based on the preset individual tree conditions and the preset distance conditions, screening the secondary segmented individual tree point set with class labels, adding the points in the secondary segmented individual tree point set with class labels that meet the preset individual tree conditions or the preset distance conditions to the initial reference individual tree point set to obtain a second reference individual tree point set, and determining the secondary segmented individual tree point set that does not meet the preset individual tree conditions and the preset distance conditions as the remaining individual tree point set; Based on the second reference individual tree point set and the remaining individual tree point set, constructing the local K-nearest neighbor point set of each remaining individual tree in the remaining individual tree point set, determining the proportion of the number of points in the second reference individual tree point set in the local K-nearest neighbor point set to the total number of points in the local K-nearest neighbor point set, adding the remaining individual trees corresponding to the local K-nearest neighbor point sets with a proportion greater than 0.5 to the second reference individual tree point set to obtain a third reference individual tree point set, and determining the updated structural parameters of each large tree based on the airborne laser point cloud of each large tree in the third reference individual tree point set; Removing the airborne laser point cloud with the updated height of the lowest point of the tree crown greater than 5m of each large tree from the initial airborne laser point cloud to obtain the second removed airborne laser point cloud; Removing the airborne laser point cloud in the preset cylindrical buffer area from the second removed airborne laser point cloud to obtain the airborne laser point cloud to be segmented; the bottom surface of the preset cylindrical buffer area is a circle with the updated position as the center and a radius of 0.8m, and the height of the preset cylindrical buffer area is the updated height of the lowest point of the tree crown; Using the Delaunay triangulation interpolation algorithm to generate a young tree crown layer height model according to the airborne laser point cloud to be segmented, and using a fixed window to detect the local maximum value in the young tree crown layer height model to obtain the segmentation parameters of the young trees, the segmentation parameters include: position, local plant density, and crown width; Using the mean shift algorithm to determine the young trees in the current forest according to the airborne laser point cloud to be segmented and the segmentation parameters.

2. The identification method of single young trees under forest as claimed in claim 1, wherein The preset individual tree conditions include: The tree height is greater than 5m; The range of the ratio of the east-west radius to the north-south radius of the crown width is: 0.7 to 1.3; The centroid is within the range of the crown width; The number of single-tree point clouds is greater than half of the average value of the number of all single-tree point clouds.

3. The identification method of single understory saplings according to claim 1, characterized in that, The preset distance condition includes: The distance between the single tree composed of the points with the same class label in the secondarily segmented single-tree point set with class labels and the first reference single tree is less than the crown width of the first reference single tree, and the first reference single tree is the single tree composed of the points in the initial reference single-tree point set.

4. An identification system for single understory saplings, characterized in that, The system includes: An initial point cloud acquisition module for acquiring the initial airborne laser point cloud of each single tree in the current forest; A first segmentation module for performing single-tree segmentation processing on the initial airborne laser point cloud by using the spectral clustering method, determining the airborne laser point cloud with a height greater than 5m as the airborne laser point cloud of each big tree, and determining the initial structural parameters of each big tree according to the airborne laser point cloud of each big tree; the structural parameters include: tree height, crown width, position, centroid, the height of the lowest point of the tree crown, and the number of single-tree point clouds; A second segmentation module for determining the airborne laser point clouds of all big trees that meet the preset single-tree conditions as the initial reference single-tree point set, and using the bidirectional projection method to secondarily segment the airborne laser point clouds of the big trees that do not meet the preset single-tree conditions to obtain secondarily segmented laser point clouds; the preset single-tree conditions are determined according to the initial structural parameters; A clustering module for iteratively clustering the secondarily segmented laser point clouds by using the mean shift clustering algorithm to obtain a secondarily segmented single-tree point set with class labels, and the secondarily segmented single-tree point set with class labels includes multiple single-tree points with class labels; A screening module for screening the secondarily segmented single-tree point set with class labels based on the preset single-tree conditions and the preset distance conditions, adding the points in the secondarily segmented single-tree point set with class labels that meet the preset single-tree conditions or the preset distance conditions to the initial reference single-tree point set to obtain a second reference single-tree point set, and determining the secondarily segmented single-tree point set that does not meet the preset single-tree conditions and the preset distance conditions as the remaining single-tree point set; A structural parameter update module for constructing a local K-nearest neighbor point set of each remaining single-tree point in the remaining single-tree point set based on the second reference single-tree point set and the remaining single-tree point set, determining the proportion of the number of points in the second reference single-tree point set in the local K-nearest neighbor point set to the total number of points in the local K-nearest neighbor point set, adding the remaining single-tree points corresponding to the local K-nearest neighbor point sets with a proportion greater than 0.5 to the second reference single-tree point set to obtain a third reference single-tree point set, and determining the updated structural parameters of each big tree based on the airborne laser point cloud of each big tree in the third reference single-tree point set; A second airborne laser point cloud removal determination module for removing the airborne laser point cloud with the updated height of the lowest point of the tree crown of each big tree greater than 5m from the initial airborne laser point cloud to obtain a second removed airborne laser point cloud; The airborne laser point cloud to be segmented determination module is used to remove the airborne laser point cloud in the preset cylindrical buffer area from the secondary removed airborne laser point cloud to obtain the airborne laser point cloud to be segmented; the bottom surface of the preset cylindrical buffer area is a circle with the updated position as the center and a radius of 0.8 m, and the height of the preset cylindrical buffer area is the lowest height of the updated tree crown; The segmentation parameter determination module is used to generate a young tree crown layer height model according to the airborne laser point cloud to be segmented by using the Delaunay triangulation interpolation algorithm, and detect the local maximum value in the young tree crown layer height model by using a fixed window to obtain the segmentation parameters of the young trees, and the segmentation parameters include: position, local plant density, and crown width; The young tree determination module is used to determine the young trees in the current forest by using the mean shift algorithm according to the airborne laser point cloud to be segmented and the segmentation parameters.

5. A device, characterized in that, Comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 3.

6. A storage medium, characterized in that, On which a computer program is stored, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images