Method and device for extracting forest land tree trunks, and method for training model

Abstract

The present disclosure provides a method and device for extracting forest tree trunks and a method for training a model. Through a trunk point cloud segmentation model based on deep learning, it can automatically learn the differences between trunk point clouds and non-trunk point clouds. It is a high-precision trunk point cloud detector that can effectively filter out the interference of shrubs and branches and leaves, and obtain high-precision trunk point clouds. At the same time, the present disclosure automatically detects whether it is a dense trunk point cloud. For dense trunk point clouds, through a pre-trained adjacent trunk recognition model, it can correct the misdetection of identifying adjacent trunks as a single trunk in the existing solutions, effectively reducing missed detections, and ensuring that even in dense trunk areas, the point clouds of each trunk can be effectively identified and extracted separately. Through the solution of the present disclosure, the trunks can be accurately separated from the point cloud data, improving the accuracy and efficiency of trunk extraction.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, particularly to the field of forestry resource management, and discloses a method and device for extracting tree trunks in forest land and a method for training a model. Background Art

[0002] In forestry management and forest resource assessment, it is crucial to accurately understand the location, quantity, species, and health status of trees. Currently, forestry resource surveys mainly adopt the methods of manual field measurement and sampling survey, which are labor-intensive, inefficient, and are easily restricted by the skills of measurers and topographical conditions.

[0003] Currently, using the data scanned by ground-based lidar for tree trunk recognition is a solution with high efficiency and low labor cost. However, the current technical solutions have the following defects: First, the existing solutions often cannot effectively remove the interference of shrubs and branches and leaves, which leads to inaccuracies in extracting tree trunk data. Since the shrubs and branches and leaves may be mixed with the point cloud data of the tree trunks, this will affect the accurate positioning and recognition of the tree trunks. Second, in the forest land scene, the distribution environment of the tree trunks is complex, and the existing solutions are prone to misidentifying multiple adjacent tree trunks as a single tree trunk when processing this data; this misidentification problem is particularly obvious in dense forest areas and will have a great impact on subsequent tree analysis and resource assessment. Summary of the Invention

[0004] The present disclosure provides at least a method and device for extracting tree trunks in forest land and a method for training a model to solve at least one of the above technical problems.

[0005] According to one aspect of the present disclosure, a method for extracting tree trunks in forest land is provided, including:

[0006] Obtaining the data collected by ground-based lidar in a forest area and preprocessing the data collected by the ground-based lidar;

[0007] Using a pre-trained tree trunk point cloud segmentation model to perform point cloud segmentation processing on the preprocessed data collected by the ground-based lidar to obtain the tree trunk point clouds of multiple tree trunks;

[0008] Using a preset confidence threshold, a preset starting height of the tree trunk point cloud, and a preset minimum number of points of the tree trunk point cloud to perform denoising processing on the tree trunk point cloud;

[0009] Performing layer slicing processing on the tree trunk point clouds of each tree trunk obtained after denoising in the Z-axis direction, and projecting the point clouds of each slice obtained by the slicing processing in the Z-axis direction to obtain a projection plane corresponding to each slice;

[0010] Determine the center points of the tree trunk contours corresponding to each projection plane respectively, and perform density clustering on the center points of the tree trunk contours in the Z-axis direction to determine the tree trunk marks of each tree trunk;

[0011] Use the tree trunk marks and a preset threshold to determine the non-dense tree trunk point cloud and the dense tree trunk point cloud, and use an adjacent tree trunk recognition model to determine the number of tree trunks corresponding to the dense tree trunk point cloud;

[0012] According to the number of tree trunks, perform clustering processing on the dense tree trunk point cloud to obtain new tree trunk point clouds corresponding to the number of tree trunks;

[0013] Perform post-processing operations on the new tree trunk point cloud and the non-dense tree trunk point cloud to obtain the target tree trunk point cloud.

[0014] In a possible implementation manner, the tree trunk point cloud segmentation model is trained using the following steps:

[0015] Obtain first sample data collected by a ground-based lidar, and preprocess the first sample data to obtain sample point cloud data;

[0016] For each measurement point, according to the coordinate value of the measurement point, determine the number of sample point clouds within the spherical domain corresponding to the measurement point, and determine the point cloud density feature of the measurement point according to the radius of the spherical domain and the number of sample point clouds; wherein, the measurement point is within the spherical domain;

[0017] For each measurement point, determine the average laser reflection intensity corresponding to the measurement point according to the laser reflection intensity at each point cloud within the spherical domain corresponding to the measurement point;

[0018] For each measurement point, determine the normal vector feature of the measurement point according to the average distance between the measurement point and K adjacent measurement points; wherein, K is a positive integer;

[0019] Train a tree trunk point cloud segmentation model using the point cloud density feature, the average laser reflection intensity, and the normal vector feature.

[0020] In a possible implementation manner, the denoising process of the tree trunk point cloud using a preset confidence threshold, a preset starting height of the tree trunk point cloud, and a preset minimum number of points of the tree trunk point cloud includes:

[0021] Eliminate the tree trunk point cloud data with a confidence less than the confidence threshold;

[0022] Eliminate the tree trunk point cloud data with a starting height of the tree trunk less than the preset starting height of the tree trunk point cloud;

[0023] Eliminate the tree trunk point cloud data with a point cloud number less than the preset minimum number of points of the tree trunk point cloud.

[0024] In a possible implementation, the hierarchical slicing process of the trunk point clouds of each denoised trunk in the Z-axis direction includes:

[0025] Perform hierarchical slicing on the trunk point clouds of each denoised trunk in the Z-axis direction according to the preset starting point height, slicing height, and ending point height respectively.

[0026] In a possible implementation, the process of respectively determining the center points of the trunk contours corresponding to each projection plane and performing density clustering on the center points of the trunk contours in the Z-axis direction to obtain the trunk labels of each trunk includes:

[0027] For each projection plane, assign the pixel value of the first preset color to the trunk point cloud in this projection plane, and assign the pixel value of the second preset color to the non-trunk point cloud in this projection plane;

[0028] Convert the assigned projection plane into an RGB image, and extract the trunk contour using the RGB image;

[0029] Determine the corresponding center points of the trunk contours using each trunk contour;

[0030] Perform density clustering on the coordinates of the center points of the trunk contours in the Z-axis direction to obtain the trunk labels of each trunk.

[0031] In a possible implementation, the adjacent trunk recognition model is trained using the following steps:

[0032] Obtain the second sample data collected by the ground-based lidar, and determine the sample trunk labels corresponding to each sample trunk in the second sample data;

[0033] Using each sample trunk label, locate the sample trunk point cloud corresponding to each sample trunk, and project the sample trunk point cloud of each sample trunk in the X-axis and Y-axis directions to obtain the projection images corresponding to each sample trunk;

[0034] Label the number of trunks in each projection image;

[0035] Using the labeled number of trunks in each projection image and each projection image, train to obtain an adjacent trunk recognition model.

[0036] In a possible implementation, the post-processing operation on the new trunk point cloud and the non-dense trunk point cloud to obtain the target trunk point cloud includes:

[0037] Merge the new trunk point cloud and the non-dense trunk point cloud to obtain a merged point cloud;

[0038] In the merged point cloud, fit the shape of the trunk to a cylinder;

[0039] Remove the trunk point clouds that cannot be fitted into a cylinder.

[0040] When the fitted cylinders intersect in space and the volume of the intersection is greater than a preset volume threshold, merge the intersecting trunk point clouds into the trunk point clouds of the same trunk.

[0041] Remove the trunk point clouds corresponding to the trunks with a height less than the preset height threshold of the cylinder to obtain the target trunk point clouds.

[0042] According to another aspect of the present disclosure, there is provided a method for training a trunk point cloud segmentation model, including:

[0043] Obtain the first sample data collected by the ground-based lidar and preprocess the first sample data to obtain sample point cloud data.

[0044] For each measurement point, according to the coordinate value of the measurement point, determine the number of sample point clouds within the spherical domain corresponding to the measurement point, and determine the point cloud density feature of the measurement point according to the radius of the spherical domain and the number of sample point clouds; wherein, the measurement point is within the spherical domain.

[0045] For each measurement point, determine the average laser reflection intensity corresponding to the measurement point according to the laser reflection intensity at each sample point cloud within the spherical domain corresponding to the measurement point.

[0046] For each measurement point, determine the normal vector feature of the measurement point according to the average distance between the measurement point and K adjacent measurement points; wherein, K is a positive integer.

[0047] Train a trunk point cloud segmentation model using the point cloud density feature, the average laser reflection intensity, and the normal vector feature.

[0048] According to another aspect of the present disclosure, there is provided a method for training an adjacent trunk recognition model, including:

[0049] Obtain the second sample data collected by the ground-based lidar and determine the sample trunk markers corresponding to each sample trunk of the second sample data.

[0050] Using each sample trunk marker, locate the sample trunk point clouds corresponding to each sample trunk, and project the sample trunk point clouds of each sample trunk on the X-axis and Y-axis to obtain the projection images corresponding to each sample trunk.

[0051] Label the number of trunks in each projection image.

[0052] Train an adjacent trunk recognition model using the labeled number of trunks in each projection image and each projection image.

[0053] According to another aspect of the present disclosure, there is provided a forest land tree trunk extraction device, including:

[0054] A data preprocessing module, configured to obtain ground-based lidar acquisition data of a forest area and preprocess the ground-based lidar acquisition data;

[0055] A tree trunk point cloud segmentation module, configured to perform point cloud segmentation processing on the preprocessed ground-based lidar acquisition data by using a pre-trained tree trunk point cloud segmentation model to obtain tree trunk point clouds of multiple tree trunks;

[0056] A denoising module, configured to perform denoising processing on the tree trunk point clouds by using a preset confidence threshold, a preset starting height of the tree trunk point clouds, and a preset minimum number of points of the tree trunk point clouds;

[0057] A slicing module, configured to perform hierarchical slicing processing on the tree trunk point clouds of each tree trunk obtained by denoising in the Z-axis direction, and project the point clouds of each slice obtained by the slicing processing in the Z-axis direction to obtain a projection plane corresponding to each slice;

[0058] A tree trunk marking determination module, configured to respectively determine the center points of the tree trunk contours corresponding to each projection plane, and perform density clustering on the center points of the tree trunk contours in the Z-axis direction to determine the tree trunk markings of each tree trunk;

[0059] A tree trunk number determination module, configured to determine non-dense tree trunk point clouds and dense tree trunk point clouds by using the tree trunk markings and a preset threshold, and determine the number of tree trunks corresponding to the dense tree trunk point clouds by using an adjacent tree recognition model;

[0060] A new tree trunk point cloud segmentation module, configured to perform clustering processing on the dense tree trunk point clouds according to the number of tree trunks to obtain new tree trunk point clouds corresponding to the number of tree trunks;

[0061] A point cloud post-processing module, configured to perform post-processing operations on the new tree trunk point clouds and the non-dense tree trunk point clouds to obtain target tree trunk point clouds

[0062] According to another aspect of the present disclosure, there is provided an electronic device, including a memory, a processor, and a computer program stored on the memory, where the processor implements the method described in any one of the above when executing the computer program.

[0063] According to another aspect of the present disclosure, there is provided a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and the computer program implements the method described in any one of the above when executed by a processor.

[0064] The method and device for extracting forest land tree trunks according to the present disclosure can automatically learn the differences between tree trunk point clouds and non-tree trunk point clouds through a deep learning-based tree trunk point cloud segmentation model. It is a high-precision tree trunk point cloud detector that can effectively filter out the interference of shrubs and branches and leaves, and obtain high-precision tree trunk point clouds. At the same time, the present disclosure automatically detects whether it is a dense tree trunk point cloud. For dense tree trunk point clouds, through a pre-trained adjacent tree trunk recognition model, the misdetection of identifying adjacent tree trunks as a single tree trunk in the existing solution can be corrected, effectively reducing the missed detection, ensuring that even in the dense tree trunk area, the point clouds of each tree trunk can be effectively identified and extracted separately. Through the solution of the present disclosure, the tree trunks can be accurately separated from the point cloud data, improving the accuracy and efficiency of tree trunk extraction.

[0065] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them:

[0067] Figure 1 is a flowchart of the method for extracting forest land tree trunks according to an embodiment of the present disclosure;

[0068] Figure 2A is one of the flowcharts of the method for extracting tree trunks according to another embodiment of the present disclosure;

[0069] Figure 2B is the second flowchart of the method for extracting tree trunks according to another embodiment of the present disclosure;

[0070] Figure 2C is the third flowchart of the method for extracting tree trunks according to another embodiment of the present disclosure;

[0071] Figure 3 is a flowchart of the training method of the tree trunk point cloud segmentation model according to an embodiment of the present disclosure;

[0072] Figure 4 is a flowchart of the training method of the adjacent tree trunk recognition model according to an embodiment of the present disclosure;

[0073] Figure 5 is a schematic structural diagram of the forest land tree trunk extraction device according to an embodiment of the present disclosure;

[0074] Figure 6 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0075] The exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, descriptions of well-known functions and structures are omitted in the following description for clarity and conciseness.

[0076] In view of the defects that the existing solutions cannot effectively remove the interference of shrubs and branches and leaves, and that multiple adjacent tree trunks are easily misidentified as a single tree trunk, the present disclosure provides a method and device for extracting tree trunks in a forest area, and a model training method. Through a tree trunk point cloud segmentation model based on deep learning, the present disclosure can automatically learn the differences between tree trunk point clouds and non-tree trunk point clouds. It is a high-precision tree trunk point cloud detector that can effectively filter out the interference of shrubs and branches and leaves and obtain high-precision tree trunk point clouds. At the same time, the present disclosure automatically detects whether it is a dense tree trunk point cloud. For dense tree trunk point clouds, through a pre-trained adjacent tree trunk recognition model, the misdetection of identifying adjacent tree trunks as a single tree trunk in the existing solutions can be corrected, effectively reducing the missed detection, ensuring that even in a dense tree trunk area, the point clouds of each tree trunk can be effectively identified and extracted separately. Through the solution of the present disclosure, tree trunks can be accurately separated from point cloud data, improving the accuracy and efficiency of tree trunk extraction.

[0077] The technical solutions of the present disclosure will be described below through specific embodiments.

[0078] As Figure 1 shown, it is a flowchart of the method for extracting tree trunks in a forest area in this embodiment. The execution subject of this embodiment is a computing device or component with data processing capabilities. Specifically, the method of this embodiment may include the following steps:

[0079] S110. Obtain the data collected by a terrestrial lidar in a forest area, and preprocess the data collected by the terrestrial lidar.

[0080] Terrestrial Laser Scanning (TLS) can provide high-resolution and high-precision three-dimensional structure data of trees.

[0081] Resample, denoise, classify ground points, and normalize according to ground points for the data collected by the terrestrial lidar in sequence, and finally obtain clean point cloud data.

[0082] S120. Use the pre-trained tree trunk point cloud segmentation model to perform point cloud segmentation processing on the preprocessed data collected by the terrestrial lidar, and obtain the tree trunk point clouds of multiple tree trunks.

[0083] The trunk point cloud segmentation model is pre-trained according to the point cloud density characteristics, average laser reflection intensity, and normal vector characteristics corresponding to each measurement point.

[0084] S130. Denoise the trunk point cloud by using a preset confidence threshold, a preset starting height of the trunk point cloud, and a preset minimum number of points of the trunk point cloud.

[0085] Generally speaking, most of the points with low confidence are noise points of non-trunk points and can be filtered out. The filtering of the starting point height of the trunk point cloud can correct the situation of misidentifying large branches as trunks because trunks do not start from the air. According to the minimum number of points of the trunk point cloud, some non-trunk point clouds such as low vegetation and branches and leaves can be removed because the number of points of a complete trunk will not be too small. Therefore, the specific denoising process can include the following steps:

[0086] Specifically, the trunk point cloud data with a confidence less than the confidence threshold can be removed; the trunk point cloud data with a starting height of the trunk less than the preset starting height of the trunk point cloud can be removed; the trunk point cloud data with a point cloud quantity less than the minimum number of points of the trunk point cloud can be removed.

[0087] S140. Perform hierarchical slicing processing on the trunk point cloud of each trunk obtained by denoising in the Z-axis direction, and project the point cloud of each slice obtained by the slicing processing in the Z-axis direction to obtain a projection plane corresponding to each slice.

[0088] Perform hierarchical slicing processing on the trunk point cloud of each trunk obtained by denoising in the Z-axis direction according to the preset starting point height, slice height, and end point height.

[0089] S150. Determine the center point of the trunk contour corresponding to each projection plane respectively, and perform density clustering on each center point of the trunk contour in the Z-axis direction to determine the trunk label of each trunk.

[0090] The following steps can be used to determine the center point of the trunk contour corresponding to each projection plane: for each projection plane, assign the pixel value of the first preset color to the trunk point cloud in the projection plane, and assign the pixel value of the second preset color to the non-trunk point cloud in the projection plane; convert the assigned projection plane into an RGB image, and extract the trunk contour by using the RGB image; determine the center point of the corresponding trunk contour by using each trunk contour.

[0091] The above first preset color can be blue, and the second preset color can be black.

[0092] Put the center points of all tree trunks of all layers, that is, the center points of the above-mentioned tree trunk contours together for density clustering. This density clustering will aggregate the tree trunks belonging to the same tree in different layers together. Only after clustering can a complete tree trunk be obtained and different trees can be distinguished. The tree trunk ID, that is, the above-mentioned tree trunk mark, is an attribute in the point cloud coordinate points. Each tree in the point cloud has a tree trunk ID, and the tree trunk IDs of all coordinate points of this tree are the same, while the tree trunk IDs of different trees are different. The tree trunk ID is an integer value.

[0093] S160. Use the tree trunk mark and a preset threshold to determine the non-dense tree trunk point cloud and the dense tree trunk point cloud, and use the adjacent tree trunk recognition model to determine the number of tree trunks corresponding to the dense tree trunk point cloud.

[0094] According to the tree trunk ID, locate the tree trunk point cloud corresponding to each tree trunk. Then calculate the average distance between each tree trunk point cloud and the adjacent tree trunk point cloud. If this average distance is less than the preset threshold, then this tree trunk point cloud is a dense tree trunk point cloud; otherwise, it is a non-dense tree trunk point cloud.

[0095] The adjacent tree trunk recognition model is trained based on the training samples with the number of tree trunks marked.

[0096] Using the adjacent tree trunk recognition model to determine the number of tree trunks corresponding to the dense tree trunk point cloud can specifically include the following steps:

[0097] Use the tree trunk marks corresponding to each tree trunk in the dense tree trunk point cloud to locate the tree trunk point cloud of each tree trunk in the dense tree trunk point cloud, and project the tree trunk point cloud of each tree trunk on the X-axis and Y-axis to obtain the projection image of each tree trunk; input the projection image of each tree trunk into the adjacent tree trunk recognition model. After the adjacent tree trunk recognition model processes the input projection image, output the number of tree trunks corresponding to the dense tree trunk point cloud.

[0098] S170. According to the number of tree trunks, perform clustering processing on the dense tree trunk point cloud to obtain the new tree trunk point cloud corresponding to the number of tree trunks.

[0099] Use the k-means clustering algorithm to cluster the dense tree trunk point cloud into the new tree trunk point cloud corresponding to the number of tree trunks, so that adjacent tree trunks are separated. K is the number of tree trunks recognized by the adjacent tree trunk recognition model.

[0100] S180. Perform post-processing operations on the new tree trunk point cloud and the non-dense tree trunk point cloud to obtain the target tree trunk point cloud.

[0101] The purpose of post-processing is to remove some wrongly segmented tree trunks and improve the accuracy of tree trunk extraction. Specifically, the following steps can be used for post-processing operations:

[0102] Merge the new tree trunk point cloud and the non-dense tree trunk point cloud to obtain a merged point cloud; in the merged point cloud, fit the shape of the tree trunk to a cylinder; remove the tree trunk point cloud that cannot be fitted to a cylinder; when the cylinders after fitting intersect in space and the intersecting volume is greater than a preset volume threshold, merge the intersecting tree trunk point clouds into the tree trunk point cloud of the same tree trunk; remove the tree trunk point cloud corresponding to the tree trunk with a height less than the preset height threshold to obtain the target tree trunk point cloud.

[0103] In some embodiments, the above-mentioned tree trunk point cloud segmentation model can be trained through the following steps:

[0104] Step 1: Obtain the first sample data collected by the ground-based lidar and preprocess the first sample data to obtain sample point cloud data.

[0105] Resample, denoise, classify ground points, and normalize according to ground points for the first sample data in sequence, and finally obtain clean point cloud data, that is, the above-mentioned sample point cloud data.

[0106] Step 2: For each measurement point, determine the number of sample point clouds within the spherical domain corresponding to the measurement point according to the coordinate value of the measurement point, and determine the point cloud density feature of the measurement point according to the radius of the spherical domain and the number of sample point clouds; wherein, the measurement point is within the spherical domain.

[0107] The radius of the spherical domain can be the measurement distance of the ground-based lidar.

[0108] The point cloud density feature characterizes the density feature of the point cloud, which refers to the number of point clouds per unit volume and can be used to distinguish sparse areas and dense areas. The calculation method of the point cloud density feature is:

[0109]

[0110] where r represents the radius of the spherical neighborhood of the measurement point P n and N is the number of sample point clouds in the spherical domain.

[0111] Step 3: For each measurement point, determine the average laser reflection intensity corresponding to the measurement point according to the laser reflection intensity at each point cloud within the spherical domain corresponding to the measurement point.

[0112] The average laser reflection intensity refers to the mean value of the laser reflection intensity at the point cloud within a given volume and can be used to distinguish different objects. The laser reflection intensity I at each point cloud n can be directly measured by the sensor, and the calculation formula for the average laser reflection intensity is:

[0113]

[0114] N represents the number of sample point clouds in the spherical domain corresponding to the measurement point P n

[0115] Step 4: For each measurement point, determine the normal vector feature of the measurement point according to the average distance between the measurement point and K adjacent measurement points; where K is a positive integer.

[0116] The normal vector feature represents the direction of the point surface and can be used to understand the shape and surface characteristics of geometric objects. The normal vector calculation steps are as follows:

[0117] 1. Calculate the covariance matrix C of the measurement point P n and K adjacent measurement points to it, and the formula is as follows:

[0118]

[0119] where: P n is the measurement point, p is the average distance between the point P n and K adjacent measurement points to it.

[0120] 2. Perform eigen - decomposition on the covariance matrix C to obtain three eigenvalues and eigenvectors.

[0121] 3. Select the eigenvector corresponding to the minimum eigenvalue as the normal vector d, that is, as the normal vector feature of the measurement point P n

[0122] Step 5: Use the point cloud density feature, the average laser reflection intensity, and the normal vector feature to train a trunk point cloud segmentation model.

[0123] In some embodiments, the above - mentioned adjacent trunk recognition model can be trained using the following steps:

[0124] Step 1: Obtain the second sample data collected by the ground - based lidar and determine the sample trunk markers corresponding to each sample trunk in the second sample data.

[0125] The steps for determining the sample trunk markers can refer to the above - mentioned steps S110 - S150 and will not be elaborated here.

[0126] Step 2: Use each sample trunk marker to locate the sample trunk point cloud corresponding to each sample trunk, and project the sample trunk point cloud of each sample trunk on the X - axis and Y - axis to obtain the projection image corresponding to each sample trunk.

[0127] Step 3: Mark the number of trunks in each projection image.

[0128] Step 4: Use the marked number of trunks in each projection image and each projection image to train an adjacent trunk recognition model.​​

[0129] The following further illustrates the method for extracting forest tree trunks of the present disclosure through an embodiment. Refer to Figures 2A - 2C .

[0130] Step 1: Segment the tree trunk point cloud to obtain the tree trunk point clouds of multiple tree trunks.

[0131] Step 1.1: Training of the tree trunk point cloud segmentation model: The model training steps mainly include preprocessing of sample data; extraction of point cloud density features, average laser reflection intensity, and normal vector features; and steps of model supervised training. For specific details, refer to the description in the above embodiment and will not be elaborated here.

[0132] Step 1.2: Obtain the point cloud data to be actually used for tree trunk extraction (corresponding to the data collected by the above ground-based lidar), and preprocess the point cloud data. Use the trained tree trunk point cloud segmentation model to perform tree trunk point cloud segmentation processing on the preprocessed data to obtain the tree trunk point clouds of multiple tree trunks.

[0133] Step 1.3: Denoise the tree trunk point cloud: Use a preset confidence threshold, a preset starting height of the tree trunk point cloud, and a preset minimum number of points of the tree trunk point cloud to perform denoising filtering processing on the tree trunk point cloud.

[0134] Step 2: Extract tree trunk markings.

[0135] Step 2.1: Perform hierarchical slicing processing on the tree trunk point clouds of each tree trunk obtained by denoising in Step 1.3 in the Z-axis direction according to a preset starting point height, slicing height, and ending point height, and project the point clouds of each slice obtained by the slicing processing in the Z-axis direction to obtain the projection planes corresponding to each slice.

[0136] Step 2.2: For each projection plane, assign the pixel value of blue to the tree trunk point cloud in the projection plane, and assign the pixel value of black to the non-tree trunk point cloud in the projection plane; convert the assigned projection plane into an RGB image, and use the RGB image to extract the tree trunk contour; use each tree trunk contour to determine the corresponding tree trunk contour center point.

[0137] Step 2.3: Perform density clustering on the coordinates of each tree trunk contour center point in the Z-axis direction to obtain the tree trunk markings of each tree trunk.

[0138] In the above Steps 1 and 2, a tree trunk point cloud segmentation model is trained through labeled samples. Use this model to accurately segment the tree trunk point cloud from the forest point cloud to be extracted, and then perform clustering through the multi-layer tree trunk contour center points to extract the tree trunk Id, that is, the above tree trunk marking.

[0139] Step 3: Identify adjacent tree trunks.

[0140] Step 3.1, Training of the adjacent tree trunk recognition model: The model training steps mainly include projecting the trunk point clouds corresponding to the sample data on the X-axis and Y-axis, and the steps of model supervised training. For specific details, refer to the description in the above embodiments and will not be elaborated here.

[0141] Step 3.1, According to the trunk markings determined in step 2.3, locate the trunk point clouds of each trunk.

[0142] Step 3.2, Dense trunk point cloud recognition.

[0143] Calculate the average distance between each trunk point cloud and its adjacent trunk point clouds. If this average distance is less than the preset threshold, then this trunk point cloud is a dense trunk point cloud; otherwise, it is a non-dense trunk point cloud.

[0144] Step 3.3, Determine the number of adjacent tree trunks: Use the adjacent tree trunk recognition model to determine the number of tree trunks corresponding to the dense trunk point clouds.

[0145] Using the trunk markings corresponding to the dense trunk point clouds, locate the trunk point clouds of each trunk in the dense trunk point clouds, and project the trunk point clouds of each trunk on the X-axis and Y-axis to obtain the projection images of each trunk; input the projection images of each trunk into the adjacent tree trunk recognition model. After the adjacent tree trunk recognition model processes the input projection images, output the number of tree trunks corresponding to the dense trunk point clouds.

[0146] Step 3.4, Adjacent trunk point cloud extraction: According to the number of tree trunks, perform clustering processing on the dense trunk point clouds to obtain the new trunk point clouds corresponding to the number of tree trunks.

[0147] Use the k-means clustering algorithm to cluster the dense trunk point clouds into new trunk point clouds corresponding to the number of tree trunks, so that adjacent tree trunks are separated. K is the number of tree trunks recognized by the adjacent tree trunk recognition model.

[0148] In the above step 3, after annotating the true number of tree trunks in the projection images of the sample data to obtain the training sample set, perform supervised training on the adjacent tree trunk recognition model based on deep learning to obtain the adjacent tree trunk recognition model. For the dense trunk point clouds, use the method based on mean clustering to complete the extraction of adjacent trunk point clouds to obtain the above new trunk point clouds.

[0149] Step 4, Post-processing of the trunk point clouds.

[0150] Merge the new tree trunk point cloud and the non-dense tree trunk point cloud to obtain a merged point cloud, and post-process the merged point cloud according to features such as geometric shape, intersection, and height. Specifically, the geometric shape of the tree trunk can be fitted to a cylinder. If there is a tree trunk point cloud that cannot be fitted to a cylinder, it will be regarded as a non-tree trunk point cloud; if the cylinders corresponding to multiple tree trunks intersect in space and the volume of the intersection is greater than a set threshold, then these multiple tree trunk point clouds will be merged and regarded as the same tree point cloud; for the fitted cylinder, if the height is lower than the set threshold, it may be low vegetation or some other non-tree trunk object, and such tree trunk point clouds will be regarded as non-tree trunk point clouds. Remove the above non-tree trunk point clouds to obtain the target tree trunk point cloud.

[0151] The present disclosure also provides a training method for a tree trunk point cloud segmentation model, as Figure 3 shown, including the following steps:

[0152] S310. Obtain the first sample data collected by the ground-based lidar, and preprocess the first sample data to obtain sample point cloud data.

[0153] S320. For each measurement point, determine the number of sample point clouds within the spherical domain corresponding to the measurement point according to the coordinate value of the measurement point, and determine the point cloud density feature of the measurement point according to the radius of the spherical domain and the number of sample point clouds; wherein, the measurement point is within the spherical domain.

[0154] S330. For each measurement point, determine the average laser reflection intensity corresponding to the measurement point according to the laser reflection intensity at each point cloud within the spherical domain corresponding to the measurement point.

[0155] S340. For each measurement point, determine the normal vector feature of the measurement point according to the average distance between the measurement point and K adjacent measurement points; where K is a positive integer.

[0156] S350. Use the point cloud density feature, the average laser reflection intensity, and the normal vector feature to train and obtain a tree trunk point cloud segmentation model.

[0157] The present disclosure also provides a training method for an adjacent tree trunk recognition model, as Figure 4 shown, including:

[0158] S410. Obtain the second sample data collected by the ground-based lidar, and determine the sample tree trunk labels corresponding to the second sample data for each sample tree trunk.

[0159] S420. Use the trunk markers of each sample to locate the sample trunk point cloud corresponding to each sample trunk, and project the sample trunk point cloud of each sample trunk on the X-axis and Y-axis to obtain the projection image corresponding to each sample trunk.

[0160] S430. Mark the number of tree trunks in each projection image.

[0161] S440. Use the marked number of tree trunks in each projection image and each projection image to train and obtain an adjacent tree trunk recognition model.

[0162] The steps of the above model training are the same as the relevant steps in the method for extracting tree trunks in the forest land of the embodiment, and will not be elaborated here.

[0163] Based on the same inventive concept, the present disclosure provides a device for extracting tree trunks in forest land. The steps performed by the components of the device are the same as or similar to those of the above method. Therefore, the similar parts will not be elaborated. As Figure 5 shown, the device for extracting tree trunks in forest land in this embodiment includes:

[0164] A data preprocessing module 510, configured to obtain the data collected by the ground lidar in the forest area and preprocess the data collected by the ground lidar.

[0165] A trunk point cloud segmentation module 520, configured to perform point cloud segmentation processing on the preprocessed data collected by the ground lidar by using a pre-trained trunk point cloud segmentation model to obtain the trunk point clouds of multiple tree trunks.

[0166] A denoising module 530, configured to perform denoising processing on the trunk point cloud by using a preset confidence threshold, a preset starting height of the trunk point cloud, and a preset minimum number of points of the trunk point cloud.

[0167] A slicing module 540, configured to perform hierarchical slicing processing on the trunk point cloud of each trunk obtained by denoising in the Z-axis direction, and project the point cloud of each slice obtained by the slicing processing in the Z-axis direction to obtain the projection plane corresponding to each slice.

[0168] A trunk marker determination module 550, configured to respectively determine the center points of the trunk contours corresponding to each projection plane, and perform density clustering on the center points of the trunk contours in the Z-axis direction to determine the trunk markers of each trunk.

[0169] A trunk number determination module 560, configured to determine the non-dense trunk point cloud and the dense trunk point cloud by using the trunk marker and a preset threshold, and determine the number of tree trunks corresponding to the dense trunk point cloud by using the adjacent tree trunk recognition model.

[0170] A new tree trunk point cloud segmentation module 570, which is used to perform clustering processing on the dense tree trunk point cloud according to the number of tree trunks, so as to obtain new tree trunk point clouds corresponding to the number of tree trunks.

[0171] A point cloud post-processing module 580, which is used to perform post-processing operations on the new tree trunk point cloud and the non-dense tree trunk point cloud to obtain target tree trunk point clouds.

[0172] Based on the same inventive concept, the present disclosure provides a training device for a tree trunk point cloud segmentation model, including:

[0173] A first sample acquisition module, which is used to acquire first sample data collected by a ground-based lidar and preprocess the first sample data to obtain sample point cloud data.

[0174] A feature processing module, which is used to, for each measurement point, determine the number of sample point clouds within the spherical domain corresponding to the measurement point according to the coordinate value of the measurement point, and determine the point cloud density feature of the measurement point according to the radius of the spherical domain and the number of sample point clouds; wherein, the measurement point is within the spherical domain; for each measurement point, determine the average laser reflection intensity corresponding to the measurement point according to the laser reflection intensity at each sample point cloud within the spherical domain corresponding to the measurement point; for each measurement point, determine the normal vector feature of the measurement point according to the average distance between the measurement point and K adjacent measurement points; wherein, K is a positive integer.

[0175] A first training module, which is used to train a tree trunk point cloud segmentation model by using the point cloud density feature, the average laser reflection intensity, and the normal vector feature.

[0176] Based on the same inventive concept, the present disclosure provides a training method for an adjacent tree trunk recognition model, including:

[0177] A second sample acquisition module, which is used to acquire second sample data collected by a ground-based lidar and determine sample tree trunk labels corresponding to each sample tree trunk in the second sample data.

[0178] A projection module, which is used to, by using each sample tree trunk label, locate the sample tree trunk point cloud corresponding to each sample tree trunk, and project the sample tree trunk point cloud of each sample tree trunk on the X-axis and Y-axis to obtain a projection image corresponding to each sample tree trunk.

[0179] A labeling module, which is used to label the number of tree trunks in each projection image.

[0180] A second training module, which is used to train an adjacent tree trunk recognition model by using the labeled number of tree trunks in each projection image and each projection image.

[0181] According to embodiments of the present disclosure, the present disclosure also provides an electronic device and a computer-readable storage medium.

[0182] Figure 6 FIG. shows a schematic block diagram of an exemplary electronic device 600 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, personal digital assistants, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0183] As Figure 6 shown, the device 600 includes a computing unit 610 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 620 or a computer program loaded from a storage unit 680 into a random access memory (RAM) 630. In the RAM 630, various programs and data required for the operation of the device 600 can also be stored. The computing unit 610, the ROM 620, and the RAM 630 are connected to each other via a bus 640. An input / output (I / O) interface 650 is also connected to the bus 640.

[0184] A plurality of components in the device 600 are connected to the I / O interface 650, including: an input unit 660, such as a keyboard, a mouse, etc.; an output unit 670, such as various types of displays, speakers, etc.; a storage unit 680, such as a magnetic disk, an optical disk, etc.; and a communication unit 690, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 690 allows the device 600 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0185] The computing unit 610 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 610 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 610 executes the various methods and processes described above. For example, in some embodiments, any of the above methods can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 680. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 600 via the ROM 620 and / or the communication unit 690. When the computer program is loaded into the RAM 630 and executed by the computing unit 610, one or more steps of any of the methods described above can be executed. Alternatively, in other embodiments, the computing unit 610 can be configured to execute any of the methods described above in any other suitable manner (e.g., by means of firmware).

[0186] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), system-on-a-chip systems (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor, receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0187] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0188] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0189] For purposes of providing an interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic, speech, or tactile input).

[0190] The systems and techniques described herein can be implemented in a computing system that includes a back-end component (e.g., as a data server), or a computing system that includes a middleware component (e.g., an application server), or a computing system that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of a communication network include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0191] A computer system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

[0192] It should be understood that the various forms of processes shown above can be used, with steps reordered, added or deleted. For example, the steps described in the present disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and no limitations are imposed herein.

[0193] The above specific embodiments do not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A method for extracting forest tree trunks, characterized in that, Including: Obtain the ground-based lidar acquisition data of the forest area, and preprocess the ground-based lidar acquisition data; Use the pre-trained trunk point cloud segmentation model to perform point cloud segmentation processing on the preprocessed ground-based lidar acquisition data to obtain the trunk point clouds of multiple trunks; Use the preset confidence threshold, the preset starting height of the trunk point cloud, and the preset minimum number of points of the trunk point cloud to perform denoising processing on the trunk point cloud; Perform layer slicing processing on the trunk point clouds of each trunk obtained by denoising in the Z-axis direction, and project the point clouds of each slice obtained by the slicing processing in the Z-axis direction to obtain the projection planes corresponding to each slice; Determine the trunk contour center points corresponding to each projection plane respectively, and perform density clustering on the trunk contour center points in the Z-axis direction to determine the trunk labels of each trunk; Use the trunk labels and the preset threshold to determine the non-dense trunk point clouds and the dense trunk point clouds, and use the adjacent trunk recognition model to determine the number of trunks corresponding to the dense trunk point clouds; According to the number of trunks, perform clustering processing on the dense trunk point clouds to obtain the new trunk point clouds corresponding to the number of trunks; Perform post-processing operations on the new trunk point clouds and the non-dense trunk point clouds to obtain the target trunk point clouds; Among them, the step of respectively determining the trunk contour center points corresponding to each projection plane and performing density clustering on the trunk contour center points in the Z-axis direction to obtain the trunk labels of each trunk includes: For each projection plane, assign the pixel value of the first preset color to the trunk point cloud in this projection plane, and assign the pixel value of the second preset color to the non-trunk point cloud in this projection plane; Convert the assigned projection plane into an RGB image, and extract the trunk contour using the RGB image; Use each trunk contour to determine the corresponding trunk contour center point; Perform density clustering on the coordinates of each trunk contour center point in the Z-axis direction to obtain the trunk labels of each trunk; Among them, the trunk point cloud segmentation model is trained using the following steps: Obtain the first sample data collected by the ground-based lidar, and preprocess the first sample data to obtain the sample point cloud data; For each measurement point, according to the coordinate value of this measurement point, determine the number of sample point clouds in the spherical neighborhood corresponding to this measurement point, and determine the point cloud density feature of this measurement point according to the radius of the spherical neighborhood and the number of sample point clouds; wherein, the measurement point is within the spherical neighborhood; For each measurement point, determine the average laser reflection intensity corresponding to this measurement point according to the laser reflection intensity at each sample point cloud in the spherical neighborhood corresponding to this measurement point; For each measurement point, determine the normal vector feature of this measurement point according to the average distance between this measurement point and K adjacent measurement points; where K is a positive integer; Use the point cloud density feature, the average laser reflection intensity, and the normal vector feature to train and obtain the trunk point cloud segmentation model.

2. The method according to claim 1, wherein The step of using the preset confidence threshold, the preset starting height of the trunk point cloud, and the preset minimum number of points of the trunk point cloud to perform denoising processing on the trunk point cloud includes: Eliminate the tree trunk point cloud data whose confidence level is less than the confidence level threshold; Eliminate the tree trunk point cloud data whose starting height of the tree trunk is less than the preset tree trunk point cloud starting height; The tree trunk point cloud data with a point cloud number less than the minimum number of tree trunk point cloud points are eliminated.

3. The method according to claim 1, wherein The step of performing layered slicing processing on the denoised tree trunk point cloud of each tree trunk in the Z-axis direction includes: According to the preset starting point height, slice height and end point height, the denoised tree trunk point cloud of each tree trunk is layered and sliced in the Z-axis direction.

4. The method according to claim 1, characterized in that The adjacent trunk recognition model is trained using the following steps: Acquire second sample data collected by the ground-based laser radar, and determine a sample trunk mark of each sample trunk corresponding to the second sample data; Using the markers of each sample trunk, locate the sample trunk point cloud corresponding to each sample trunk, and project the sample trunk point cloud of each sample trunk on the X-axis and the Y-axis to obtain the projection image corresponding to each sample trunk; Label the number of tree trunks in each projection image; The adjacent tree trunk recognition model is trained using the labeled number of tree trunks in each projection image and each projection image.

5. The method according to claim 1, wherein The post-processing operation is performed on the new trunk point cloud and the non-dense trunk point cloud to obtain the target trunk point cloud, including: Merging the new tree trunk point cloud and the non-dense tree trunk point cloud to obtain a merged point cloud; In the merged point cloud, the shape of the tree trunk is fitted as a cylinder; Remove the tree trunk point clouds that cannot be fitted into a cylindrical shape; When the fitted cylinders intersect in space and the intersecting volume is greater than a preset volume threshold, the intersecting trunk point clouds are merged into the trunk point cloud of the same trunk; The trunk point cloud corresponding to the tree trunk whose cylindrical height is less than the preset height threshold is removed to obtain the target trunk point cloud.

6. A training method for a trunk point cloud segmentation model, characterized in that, include: Acquire first sample data collected by a ground-based laser radar, and preprocess the first sample data to obtain sample point cloud data; wherein the sample point cloud data includes point cloud data corresponding to the trunks of trees in the forest; For each measurement point, the number of sample point clouds in a spherical neighborhood corresponding to the measurement point is determined according to the coordinate value of the measurement point, and the point cloud density feature of the measurement point is determined according to the radius of the spherical neighborhood and the number of sample point clouds; wherein the measurement point is in the spherical neighborhood; For each measurement point, determine the average laser reflection intensity corresponding to the measurement point according to the laser reflection intensity at each sample point cloud within the spherical neighborhood corresponding to the measurement point; For each measurement point, the normal vector feature of the measurement point is determined according to the average distance between the measurement point and its K neighboring measurement points; wherein K is a positive integer; The point cloud density feature, the average laser reflection intensity, and the normal vector feature are used to train a tree trunk point cloud segmentation model.

7. A device for extracting tree trunks in a forest land, characterized in that, include: A data preprocessing module, used to obtain ground-based laser radar data collected in the forest area and preprocess the ground-based laser radar data collected; A tree trunk point cloud segmentation module is used to perform point cloud segmentation processing on the pre-processed ground-based lidar data using a pre-trained tree trunk point cloud segmentation model to obtain tree trunk point clouds of multiple tree trunks; A denoising module for denoising the trunk point cloud by using a preset confidence threshold, a preset starting height of the trunk point cloud, and a preset minimum number of points of the trunk point cloud; A slicing module for performing hierarchical slicing on the trunk point cloud of each trunk obtained after denoising in the Z-axis direction, and projecting the point cloud of each slice obtained by the slicing process in the Z-axis direction to obtain a projection plane corresponding to each slice; A trunk marking determination module for respectively determining the center points of the trunk contours corresponding to each projection plane, and performing density clustering on the center points of each trunk contour in the Z-axis direction to determine the trunk markings of each trunk; A trunk number determination module for using the trunk markings and a preset threshold to determine non-dense trunk point clouds and dense trunk point clouds, and using an adjacent trunk recognition model to determine the number of trunks corresponding to the dense trunk point clouds; A new trunk point cloud segmentation module for clustering the dense trunk point clouds according to the number of trunks to obtain new trunk point clouds corresponding to the number of trunks; A point cloud post-processing module for performing post-processing operations on the new trunk point clouds and the non-dense trunk point clouds to obtain target trunk point clouds; Among them, when the trunk marking determination module respectively determines the center points of the trunk contours corresponding to each projection plane and performs density clustering on the center points of each trunk contour in the Z-axis direction to obtain the trunk markings of each trunk, it specifically is used for: For each projection plane, assign the pixel value of a first preset color to the trunk point cloud in this projection plane, and assign the pixel value of a second preset color to the non-trunk point cloud in this projection plane; Convert the assigned projection plane into an RGB image, and extract the trunk contour by using the RGB image; Use each trunk contour to determine the corresponding center point of the trunk contour; Perform density clustering on the coordinates of the center points of each trunk contour in the Z-axis direction to obtain the trunk markings of each trunk; Among them, the trunk point cloud segmentation model is trained by the following steps: Obtain first sample data collected by a ground-based lidar, and preprocess the first sample data to obtain sample point cloud data; For each measurement point, determine the number of sample point clouds within the spherical neighborhood corresponding to this measurement point according to the coordinate value of this measurement point, and determine the point cloud density feature of this measurement point according to the radius of the spherical neighborhood and the number of sample point clouds; wherein, the measurement point is within the spherical neighborhood; For each measurement point, determine the average laser reflection intensity corresponding to this measurement point according to the laser reflection intensity at each sample point cloud within the spherical neighborhood corresponding to this measurement point; For each measurement point, determine the normal vector feature of this measurement point according to the average distance between this measurement point and K adjacent measurement points; wherein, K is a positive integer; Train a trunk point cloud segmentation model by using the point cloud density feature, the average laser reflection intensity, and the normal vector feature.

Images