CN114743008A - A method, device and computer equipment for segmenting point cloud data of single plant vegetation - Google Patents

Abstract

The invention relates to the technical field of point cloud data processing, and discloses a method, device and computer equipment for segmenting point cloud data of a single plant vegetation. The vegetation crown point of the extreme point, and then the tree trunk is extracted by the combination of the top-view point cloud and the side-view point cloud using the tensor voting method, and the center point of the trunk is used as the new extreme point. Correction processing is performed to obtain the corrected vegetation crown points, and then according to the predetermined growth restriction conditions based on the actual situation of the data, the crown growth processing is performed on the corrected vegetation crown points to obtain the growth used for constructing the crown boundary. Finally, based on the contour of a single tree canopy obtained by edge detection, through coordinate affine transformation and point cloud data extraction, the point cloud data of a single vegetation is finally obtained by segmentation, which can ensure the accuracy of the segmentation results.

Description

A method, device and computer equipment for segmenting point cloud data of single plant vegetation

technical field

The invention belongs to the technical field of point cloud data processing, and in particular relates to a method, device and computer equipment for segmenting point cloud data of a single plant vegetation.

Background technique

The detection and identification of a single vegetation (such as a single tree) is of great significance for applications in forests or cities. For example, based on the accurate identification of individual trees, the precise total number of trees can be obtained, and parameters such as tree height, DBH, crown and volume can be calculated from it, which can improve the accuracy of estimating forest biomass or tree age distribution. . So how to get from LiDAR point cloud data (each point contains three-dimensional coordinate information - namely X, Y and Z three elements, sometimes also contains color information, reflection intensity information and/or number of echoes It has important application value for forest resource inventory and urban planning and so on.

At present, the detection and segmentation of single vegetation point cloud data based on LiDAR point cloud data mainly has the following problems: (1) Due to the complex shape of trees, it is difficult to determine the extreme point, which leads to the easy occurrence of various points during segmentation. Various segmentation problems, such as under-segmentation caused by the sharing of extreme points between two adjacent trees, mis-segmentation/over-segmentation caused by irregular tree shapes, etc.; (2) Canopy boundary recognition on 2D images When the tree crown boundary is not clear, the segmentation result is not accurate enough, for example, two adjacent trees cross seriously or there are small trees under the big tree.

SUMMARY OF THE INVENTION

In order to solve the problem of inaccurate segmentation results in the existing detection and segmentation methods of single plant vegetation point cloud data based on lidar point cloud data, the present invention aims to provide a single plant vegetation point cloud data segmentation method, device and computer equipment.

In a first aspect, the present invention provides a method for segmenting point cloud data of a single vegetation, including:

Input the original point cloud data to be segmented into a vegetation point recognition model based on a binary classification network that has completed training, and output the vegetation point recognition results, wherein the original point cloud data is a fusion of top-down point clouds and side-view points. Cloud multi-source point cloud data;

Interpolate the point cloud data of the identified vegetation points in the vegetation point identification result into rasterized canopy height model CHM image data;

According to the CHM image data, applying mathematical morphology to determine the vegetation crown point used as the local extreme point;

According to the original point cloud data, the tensor voting method is applied to extract the tree trunk;

Feeding back the center point of the trunk as a new extreme point to the CHM image data to perform correction processing on the local extreme point to obtain a corrected vegetation top point;

According to the pre-determined growth restriction conditions based on the actual situation of the data, the canopy growth processing is performed on the corrected vegetation top and crown points to obtain the growth points;

Mark the growth point and the corrected vegetation top point with the same mark, and write the mark as a pixel value into the new image data, wherein the mark refers to the value obtained from the pixel value Selecting a numerical number for marking the growth point and the corrected vegetation top point within the range, and the new image data and the CHM image data have the same grid;

Perform edge detection processing on the new image data to obtain a single tree crown outline;

Affine transformation of the row and column numbers of the single tree canopy outline into real geographic coordinates to obtain the target segmentation space;

The point cloud data located in the target segmentation space is extracted from the original point cloud data to obtain the point cloud data of a single plant.

Based on the above-mentioned content of the invention, a new solution is provided that can accurately detect and segment the point cloud data of a single plant from the multi-source point cloud data. After the original point cloud data, the vegetation crown points used as local extreme points are determined through vegetation point identification processing, interpolation processing and mathematical morphology, and then the tensor voting method is used by combining the top-view point cloud and the side-view point cloud. Extract the trunk, and use the center point of the trunk as a new extreme point to perform correction processing on the local extreme point to obtain the corrected vegetation crown point, and then according to the actual situation of the data. Pre-determined growth limit Condition, the crown growth process is performed on the corrected vegetation crown points to obtain the growth points used to construct the crown boundary, and finally the single tree crown outline obtained based on edge detection is extracted through coordinate affine transformation and point cloud data, The final segmentation obtains the point cloud data of a single plant, which ensures the accuracy of the segmentation results, and has important application value for forest resource inventory and urban planning, which is convenient for practical application and promotion.

In a possible design, the binary classification network adopts a RandLA-Net network or an improved structure based on the RandLA-Net network.

In a possible design, based on the CHM image data, mathematical morphology is applied to determine the vegetation crown points used as local extreme points, including:

Perform image noise removal processing on the CHM image data by using the median filter method to obtain new CHM image data;

According to the new CHM image data, mathematical morphology is applied to determine the vegetation crown points used as local extreme points.

In a possible design, based on the CHM image data, mathematical morphology is applied to determine the vegetation crown points used as local extreme points, including:

Perform morphological erosion operation processing on the CHM image data to obtain image data erosion results;

performing a subtraction process of subtracting the corrosion result of the image data on the CHM image data to obtain a subtraction result of the image data;

According to the image data subtraction result, determine the maximum value located in the positive value area;

A point corresponding to the maximum value is determined as a vegetation crown point for use as a local extremum point.

In a possible design, according to the original point cloud data, the tensor voting method is applied to extract the tree trunk, including:

，先构建对应的方程：

，式中，

表示棒张量，

表示处于以所述某一原始点云

为中心的指定半径范围内的点云总数，

表示自然数，

表示所述某一原始点云

与在所述指定半径范围内的第

个邻域点云

之间的特征向量且有

，

表示棒张量对邻域点作用的张量算子且有

，然后求解所述方程得到所述棒张量

的至少一个特征值，最后若判定所述至少一个特征值中的最大特征值大于预设的特征阈值，则确定所述某一原始点云

为树干特征点；Traverse each original point cloud in the original point cloud data: for a certain original point cloud

, first construct the corresponding equation:

, where,

represents a stick tensor,

Indicates that it is in the original point cloud with the

is the total number of point clouds within the specified radius from the center,

represents a natural number,

represents one of the original point clouds

with the first radius within the specified radius

neighborhood point cloud

eigenvectors between and have

,

is a tensor operator representing the action of a stick tensor on a neighborhood point and has

, then solve the equation to get the stick tensor

At least one eigenvalue of , and finally if it is determined that the largest eigenvalue in the at least one eigenvalue is greater than the preset eigenvalue, then determine the certain original point cloud

is the trunk feature point;

All tree trunk feature points in the original point cloud data are extracted to obtain the tree trunk.

In a possible design, the growth constraints include tree height constraints, growth range constraints and/or adjacent tree competition point constraints, wherein the tree height constraints include the tree height H of the seed point. The minimum value is 1.5 meters, the growth range constraints include that the maximum distance R between the obtained growth point and the seed point is 3 meters, and the adjacent tree competition point constraints include that the growth point competing with the adjacent tree is higher than Seed point tree height H is 0.7 times.

In a possible design, the original point cloud data is a multi-source fusion obtained by combining and registering the vehicle lidar point cloud data, the airborne lidar point cloud data and the ground station lidar point cloud data. Point cloud data, the interpolation processing adopts the triangulation interpolation method, and the edge detection processing adopts the openCV edge detection function.

In a second aspect, the present invention provides a single plant vegetation point cloud data segmentation device, including a vegetation point identification unit, an interpolation processing unit, an extreme point determination unit, a trunk extraction unit, an extreme point correction unit, a canopy growth unit, a token processing unit, an edge detection unit, an affine transformation unit and a data extraction unit;

The vegetation point identification unit is used for inputting the original point cloud data to be segmented into a vegetation point identification model based on a binary classification network and having completed training, and outputting a vegetation point identification result, wherein the original point cloud data It is multi-source point cloud data fused with top-view point cloud and side-view point cloud;

The interpolation processing unit, which is communicatively connected to the vegetation point identification unit, is used to interpolate the point cloud data of the identified vegetation points in the vegetation point identification result into rasterized canopy height model CHM image data;

The extremum point determining unit is connected to the interpolation processing unit in communication, and is configured to apply mathematical morphology to determine the vegetation crest points used as local extremum points according to the CHM image data;

The tree trunk extraction unit is used for extracting tree trunks by applying tensor voting method according to the original point cloud data;

The extremum point correction unit is respectively connected to the extremum point determination unit and the tree trunk extraction unit in communication, and is used for feeding back the center point of the tree trunk as a new extremum point to the CHM image data. The local extreme point is corrected to obtain the corrected vegetation top point;

The canopy growth unit is connected to the extreme point correction unit in communication, and is configured to perform a canopy growth process on the corrected vegetation top and crown points according to a predetermined growth restriction condition based on the actual situation of the data to obtain a growth point;

The marking processing unit is connected to the tree canopy growth unit in communication, and is used to mark the growth point and the corrected vegetation crown point with the same mark, and write the mark as a pixel value into a new image In the data, the mark refers to a numerical number selected from the range of pixel values for marking the growth point and the corrected vegetation top point, and the new image data is the same as the CHM image data has the same raster;

The edge detection unit is communicatively connected to the mark processing unit, and is used to perform edge detection processing on the new image data to obtain a single tree canopy outline;

The affine transformation unit is connected to the edge detection unit in communication, and is used for affine transformation of the row and column numbers of the outline of the single tree canopy into real geographic coordinates to obtain the target segmentation space;

The data extraction unit is connected to the affine transformation unit in communication, and is used for extracting point cloud data located in the target segmentation space from the original point cloud data to obtain single vegetation point cloud data.

In a third aspect, the present invention provides a computer device, comprising a memory, a processor and a transceiver connected in sequence in communication, wherein the memory is used for storing a computer program, the transceiver is used for sending and receiving messages, and the processor is used for sending and receiving messages. After reading the computer program, execute the method for segmenting point cloud data of a single plant vegetation as described in the first aspect or any possible design of the first aspect.

In a fourth aspect, the present invention provides a computer-readable storage medium, where instructions are stored on the computer-readable storage medium, and when the instructions are executed on a computer, the first aspect or any possibility in the first aspect is executed. Design the described segmentation method of single vegetation point cloud data.

In a fifth aspect, the present invention provides a computer program product comprising instructions that, when the instructions are run on a computer, cause the computer to execute the single plant vegetation described in the first aspect or any possible design of the first aspect Point cloud data segmentation method.

Beneficial effects of the present invention:

(1) The invention provides a new solution that can accurately detect and segment the point cloud data of a single plant from the multi-source point cloud data. After the original point cloud data of the viewpoint cloud, the vegetation crest points used as local extreme points are determined through vegetation point identification processing, interpolation processing and mathematical morphology. The tree trunk is extracted by the quantitative voting method, and the center point of the trunk is used as the new extreme point to correct the local extreme point to obtain the corrected vegetation crown point, which is then pre-determined based on the actual situation of the data. According to the growth restriction conditions, the crown growth processing is performed on the corrected vegetation crown points to obtain the growth points used to construct the crown boundary, and finally the single tree crown outline obtained based on edge detection is obtained through coordinate affine transformation and point cloud. Data extraction, and finally segmentation to obtain the point cloud data of a single plant, which can ensure the accuracy of the segmentation results, and has important application value for forest resource inventory and urban planning, which is convenient for practical application and promotion.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

FIG. 1 is a schematic flowchart of a method for segmenting point cloud data of a single plant vegetation provided by the present invention.

FIG. 2 is a schematic structural diagram of a device for segmenting point cloud data of a single plant of vegetation provided by the present invention.

FIG. 3 is a schematic structural diagram of a computer device provided by the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be noted here that, although the description of these embodiments is for helping understanding of the present invention, it does not constitute a limitation of the present invention. Specific structural and functional details disclosed herein are merely intended to describe exemplary embodiments of the present invention. The present invention, however, may be embodied in many alternative forms and should not be construed as limited to the embodiments set forth herein.

It should be understood that although the terms first and second, etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, a first object could be referred to as a second object, and similarly a second object could be referred to as a first object, without departing from the scope of example embodiments of this invention.

It should be understood that the term "and/or" that may appear in this document is only an association relationship for describing associated objects, indicating that there may be three kinds of relationships, for example, A and/or B, which may indicate: the existence of A, There are three situations such as B alone or A and B at the same time; for the term "/and" that may appear in this article, it is to describe another related object relationship, indicating that there can be two relationships, for example, A/ and B, It can be expressed as: the existence of A alone or the existence of A and B at the same time; in addition, for the character "/" that may appear in this article, it generally means that the related objects before and after are an "or" relationship.

As shown in FIG. 1 , the method for segmenting point cloud data of a single plant vegetation provided in the first aspect of this embodiment may be executed by, but not limited to, a computer device with certain computing resources, such as a platform server, a personal computer (Personal Computer, PC , refers to a multi-purpose computer of size, price and performance suitable for personal use; desktops, notebooks to small notebooks and tablets, and ultrabooks are all personal computers), smartphones, Personal Digital Assistants (Personal Digital Assistants) , PDA) or electronic devices such as wearable devices. As shown in FIG. 1 , the method for segmenting point cloud data of a single plant of vegetation may, but is not limited to, include the following steps S1 to S10.

S1. Input the original point cloud data to be segmented into a vegetation point recognition model based on a binary classification network and have completed training, and output a vegetation point recognition result, wherein the original point cloud data is a combination of a bird's-eye view point cloud and a Multi-source point cloud data for side view point clouds.

In the step S1, in order to ensure that the original point cloud data has a top-view point cloud and a side-view point cloud, the original point cloud data is preferably the point cloud data of the vehicle-mounted lidar, the point cloud of the airborne lidar and the ground station. The multi-source point cloud data obtained by combining and registering the lidar point cloud data. At the same time, the binary classification network can preferably use the RandLA-Net network or an improved structure based on the RandLA-Net network. Since the RandLA-Net network is an existing network structure that can perform large-scale point cloud semantic segmentation, it can The entire point cloud is used as input without preprocessing and post-processing operations such as splitting or merging, so the vegetation point recognition model that can identify vegetation points or non-vegetation points based on input point cloud data can be obtained through conventional training. In addition, the input form of the original point cloud data may be, but not limited to, a text document format containing XYZ coordinate information and RGB color information.

S2. Interpolate point cloud data of identified vegetation points in the vegetation point identification result into rasterized canopy height model CHM image data.

In the step S2, the canopy height model (Canopy Height Model, CHM) is an extremely practical model that can be obtained by lidar in the forest area, and it is an expression of the height of the forest canopy on the ground , which reflects the height change of the forest canopy in the vertical direction and the distribution state in the horizontal direction. Through CHM, important forest vegetation parameters (such as single tree parameters, stand parameters, stock volume and biomass) can be extracted in various forestry investigations. Wait). Specifically, the interpolation processing may adopt the triangulation interpolation method, or may adopt the B-spline interpolation method (B-Spline), the ordinary kriging interpolation method (OK), the inverse distance weighted interpolation method (IDW), etc. The interpolation methods are all existing methods, and the rasterized CHM image data can be obtained by normal modification and interpolation.

S3. According to the CHM image data, apply mathematical morphology to determine the vegetation crest points used as local extreme points.

In the step S3, the mathematical morphology (Mathematical Morphology) is an image analysis subject based on lattice theory and topology, and is the basic theory of mathematical morphology image processing; its basic operations include: Erosion and dilation, opening and closing operations, skeleton extraction, limit erosion, hit-miss transform, morphological gradient, top-hat transform, particle analysis, and watershed transform, etc. Since each pixel point in the CHM image data represents the true height of the vegetation point, the vegetation crown point can be regarded as the local extreme point of the image, and the method of mathematical morphology is used to determine, that is, specifically, According to the CHM image data, applying mathematical morphology to determine the vegetation crown points used as local extreme points, including but not limited to the following steps S31 to S34: S31. Perform morphological erosion operation processing on the CHM image data , obtain the image data corrosion result; S32. Perform a subtraction process of subtracting the image data corrosion result on the CHM image data to obtain the image data subtraction result; S33. According to the image data subtraction result, determine that the image data is located in the positive The maximum value in the value area; S34. Determine the point corresponding to the maximum value as the vegetation crown point used as the local extreme value point. In the aforementioned step S31, since the erosion operation is to convolve the image (or a small area of the image) with the kernel (the kernel size can be set to a 5*5 square), the smooth terrain surface will not be affected , and the protruding parts will be cut off due to corrosion. Therefore, through the aforementioned subtraction processing, the protruding vegetation crown part (ie the positive area) can be retained, and then the largest part located in the positive area can be compared with The point corresponding to the value is determined as the vegetation crown point used as the local extreme point. In addition, in order to remove the influence noise in advance, preferably, according to the CHM image data, mathematical morphology is applied to determine the vegetation crown point used as the local extreme point, which also includes but is not limited to: first adopt the median filtering method to The CHM image data is subjected to image noise removal processing to obtain new CHM image data; and then, according to the new CHM image data, mathematical morphology is applied to determine the vegetation crown points used as local extreme points.

S4. According to the original point cloud data, the tensor voting method is applied to extract the tree trunk.

，先构建对应的方程：

，式中，

表示棒张量，

表示处于以所述某一原始点云

为中心的指定半径范围内的点云总数，

表示自然数，

表示所述某一原始点云

与在所述指定半径范围内的第

个邻域点云

之间的特征向量且有

（此处的

表示从所述某一原始点云

至所述第

个邻域点云

的有向线段），

表示棒张量对邻域点作用的张量算子且有

，然后求解所述方程得到所述棒张量

的至少一个特征值，最后若判定所述至少一个特征值中的最大特征值大于预设的特征阈值，则确定所述某一原始点云

为树干特征点；S42.提取所述原始点云数据中的所有树干特征点，得到树干。在前述步骤S41中，所述特征阈值可举例为0.9。In the step S4, the tensor voting method is a method for extracting image features. It uses the strong robustness of tensor to extract the characteristics of point, line and surface features in the image, and eliminate the isolation in the image. Point saliency, highlight the extracted line and surface features, and reconstruct the image, so as to achieve the purpose of removing noise and highlighting edges. Since the original point cloud data has the characteristics of multi-source data, the tree trunk can be extracted by using the tensor voting method by combining the top-view point cloud and the side-view point cloud. Specifically, according to the original point cloud data, tensor voting is applied. The tree trunk is extracted by the method, including but not limited to the following steps S41-S42: S41. Traverse each original point cloud in the original point cloud data: for a certain original point cloud

, first construct the corresponding equation:

, where,

represents a stick tensor,

Indicates that it is in the original point cloud with the

is the total number of point clouds within the specified radius from the center,

represents a natural number,

represents one of the original point clouds

with the first radius within the specified radius

neighborhood point cloud

eigenvectors between and have

(here's

represents one of the original point clouds from the

to the

neighborhood point cloud

directed line segment),

is a tensor operator representing the action of a stick tensor on a neighborhood point and has

, then solve the equation to get the stick tensor

At least one eigenvalue of , and finally if it is determined that the largest eigenvalue in the at least one eigenvalue is greater than the preset eigenvalue, then determine the certain original point cloud

is the trunk feature point; S42. Extract all the trunk feature points in the original point cloud data to obtain the trunk. In the foregoing step S41, the feature threshold may be 0.9, for example.

S5. Feed back the center point of the tree trunk as a new extreme value point to the CHM image data to perform correction processing on the local extreme value point to obtain a corrected vegetation crown point.

In the step S5, the specific method of the correction processing is an existing conventional method, for example, the intermediate point between the new extreme point and the corresponding local extreme point is used as the corrected vegetation crown point .

S6. According to the pre-determined growth restriction conditions based on the actual situation of the data, perform tree crown growth processing on the corrected vegetation crown points to obtain growth points.

In the step S6, the canopy growth process refers to the process of extending the tree crown boundary around the vegetation top crown point as the center, so the growth point is the new tree crown boundary point obtained after the tree crown growth process. Specifically, the growth constraints include, but are not limited to, tree height constraints, growth range constraints, and/or adjacent tree competition point constraints, etc., wherein the tree height constraints include but are not limited to having seed points ( That is, the corrected vegetation crown point, because the crown growth process refers to the process of extending the tree crown boundary around the vegetation crown point as the center, so it is regarded as the seed point of crown growth here) The height of the tree is H. The minimum value is 1.5 meters, etc., the growth range constraints include but are not limited to the maximum distance R between the obtained growth point and the seed point is 3 meters, etc., and the adjacent tree competition point constraints include but are not limited to adjacent trees. The growth point of the tree competition is higher than 0.7 times the tree height H of the seed point, etc.

S7. Mark the growth point and the corrected vegetation top point with the same mark, and write the mark as a pixel value into the new image data, wherein the mark refers to the value from the pixel value. Within the range of values, a numerical number for marking the growth point and the corrected vegetation top point is selected, and the new image data and the CHM image data have the same grid.

In the step S7, when the value range is [0, 255], a value of 128 can be selected as an example to uniquely mark the growth point and the corrected vegetation top point.

S8. Perform edge detection processing on the new image data to obtain a single tree crown outline.

In the step S8, since the corrected vegetation crown point and its corresponding growth point are marked with the same sign, the edge detection process can be used to determine the corrected vegetation crown point corresponding to the Single tree canopy silhouette. Specifically, the edge detection processing may be, but not limited to, use of the openCV edge detection function.

S9. Affine transform the row and column numbers of the individual tree canopy outlines into real geographic coordinates to obtain a target segmentation space.

In the step S9, since the new image data and the CHM image data have the same grid, the corresponding real images can be obtained by conventional affine transformation for the row and column numbers of the individual tree canopy contours. geographic coordinates, and then obtain the target segmented space enclosed by all the real geographic coordinates.

S10. Extract point cloud data located in the target segmentation space from the original point cloud data to obtain single vegetation point cloud data.

In the step S10, specifically, for each original point cloud in the original point cloud data, a judgment is made about the positional relationship between the point and the polygon outline (that is, the boundary of the target segmentation space). The point cloud data in the contour is used as the point cloud data of the single vegetation obtained by the final segmentation.

Therefore, based on the method for segmenting point cloud data of single plant vegetation described in the foregoing steps S1 to S10, a new solution that can accurately detect and segment point cloud data of single plant vegetation from multi-source point cloud data is provided. After the original point cloud data to be segmented and fused with the top-view point cloud and the side-view point cloud, the vegetation crest points used as local extreme points are determined through vegetation point identification processing, interpolation processing and mathematical morphology. The tree trunk is extracted by combining the top-view point cloud and the side-view point cloud using the tensor voting method, and the center point of the tree trunk is used as a new extreme point to correct the local extreme point to obtain the corrected vegetation crown Then, according to the growth restriction conditions pre-determined based on the actual situation of the data, the crown growth processing is performed on the corrected vegetation crown points to obtain the growth points used to construct the crown boundary. Tree canopy contour, through coordinate affine transformation and point cloud data extraction, and finally segmented to obtain single plant vegetation point cloud data, which can ensure the accuracy of segmentation results, and has important application value for forest resource inventory and urban planning, etc. It is convenient for practical application and promotion.

As shown in FIG. 2 , a second aspect of this embodiment provides a virtual device for implementing the method for segmenting point cloud data of a single plant vegetation as described in the first aspect, including a vegetation point identification unit, an interpolation processing unit, and an extreme point determination unit. unit, trunk extraction unit, extreme point correction unit, crown growth unit, sign processing unit, edge detection unit, affine transformation unit and data extraction unit;

The vegetation point identification unit is used for inputting the original point cloud data to be segmented into a vegetation point identification model based on a binary classification network and having completed training, and outputting a vegetation point identification result, wherein the original point cloud data It is multi-source point cloud data fused with top-view point cloud and side-view point cloud;

The interpolation processing unit, which is communicatively connected to the vegetation point identification unit, is used to interpolate the point cloud data of the identified vegetation points in the vegetation point identification result into rasterized canopy height model CHM image data;

The extremum point determining unit is connected to the interpolation processing unit in communication, and is configured to apply mathematical morphology to determine the vegetation crest points used as local extremum points according to the CHM image data;

The tree trunk extraction unit is used for extracting tree trunks by applying tensor voting method according to the original point cloud data;

The extremum point correction unit is respectively connected to the extremum point determination unit and the tree trunk extraction unit in communication, and is used for feeding back the center point of the tree trunk as a new extremum point to the CHM image data. The local extreme point is corrected to obtain the corrected vegetation top point;

The canopy growth unit is connected to the extreme point correction unit in communication, and is configured to perform a canopy growth process on the corrected vegetation top and crown points according to a predetermined growth restriction condition based on the actual situation of the data to obtain a growth point;

The marking processing unit is connected to the tree canopy growth unit in communication, and is used to mark the growth point and the corrected vegetation crown point with the same mark, and write the mark as a pixel value into a new image In the data, the mark refers to a numerical number selected from the range of pixel values for marking the growth point and the corrected vegetation top point, and the new image data is the same as the CHM image data has the same raster;

The edge detection unit is communicatively connected to the mark processing unit, and is used to perform edge detection processing on the new image data to obtain a single tree canopy outline;

The affine transformation unit is connected to the edge detection unit in communication, and is used for affine transformation of the row and column numbers of the outline of the single tree canopy into real geographic coordinates to obtain the target segmentation space;

The data extraction unit is connected to the affine transformation unit in communication, and is used for extracting point cloud data located in the target segmentation space from the original point cloud data to obtain single vegetation point cloud data.

For the working process, working details and technical effects of the aforementioned device provided in the second aspect of this embodiment, reference may be made to the method for segmenting point cloud data of a single plant vegetation described in the first aspect, and details are not described herein again.

As shown in FIG. 3 , a third aspect of this embodiment provides a computer device for executing the method for segmenting point cloud data of a single plant vegetation as described in the first aspect, including a memory, a processor and a transceiver that are sequentially connected in communication, Wherein, the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program, and execute the method for segmenting point cloud data of a single plant vegetation according to the first aspect. For example, the memory may include, but is not limited to, random-access memory (Random-Access Memory, RAM), read-only memory (Read-Only Memory, ROM), flash memory (Flash Memory), first-in-first-out memory (First Input First Output, FIFO) and/or First Input Last Output (FILO), etc.; the processor may be, but not limited to, a microprocessor of the STM32F105 series. In addition, the computer equipment may also include, but is not limited to, a power module, a display screen and other necessary components.

For the working process, working details, and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment, reference may be made to the method for segmenting point cloud data of a single vegetation plant described in the first aspect, which will not be repeated here.

A fourth aspect of this embodiment provides a computer-readable storage medium for storing instructions including the method for segmenting point cloud data of a single plant of vegetation as described in the first aspect, that is, the computer-readable storage medium stores instructions, when When the instructions are run on a computer, the method for segmenting point cloud data of a single plant vegetation as described in the first aspect is executed. The computer-readable storage medium refers to a carrier for storing data, which may include, but is not limited to, computer-readable storage media such as a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and/or a Memory Stick. Is a general purpose computer, special purpose computer, computer network, or other programmable device.

For the working process, working details, and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment, reference may be made to the method for segmenting point cloud data of a single plant vegetation as described in the first aspect, which will not be repeated here.

A fifth aspect of this embodiment provides a computer program product including instructions, when the instructions are run on a computer, the computer is made to execute the method for segmenting point cloud data of a single plant vegetation as described in the first aspect. Wherein, the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

Finally, it should be noted that the present invention is not limited to the above-mentioned optional embodiments, and anyone can obtain other various forms of products under the inspiration of the present invention. The above specific embodiments should not be construed as limiting the protection scope of the present invention, which should be defined in the claims, and the description can be used to interpret the claims.

Claims (10)

1. A single plant vegetation point cloud data segmentation method is characterized by comprising the following steps:

inputting original point cloud data to be segmented into a vegetation point identification model which is based on a binary classification network and is trained, and outputting to obtain a vegetation point identification result, wherein the original point cloud data is multi-source point cloud data fused with top view point cloud and side view point cloud;

performing interpolation processing on the point cloud data of the vegetation points identified in the vegetation point identification result to obtain rasterized canopy height model CHM image data;

determining vegetation top crown points serving as local extreme points by applying mathematical morphology according to the CHM image data;

extracting a trunk by applying a tensor voting method according to the original point cloud data;

feeding back the central point of the trunk as a new extreme point to the CHM image data to correct the local extreme point to obtain a corrected vegetation top crown point;

performing crown growth processing on the corrected vegetation top crown point according to a growth limiting condition predetermined based on the actual data condition to obtain a growth point;

marking the same mark on the growing point and the corrected vegetation crown point, and writing the mark into new image data as a pixel value, wherein the mark is a numerical value number selected from a value range of the pixel value and used for marking the growing point and the corrected vegetation crown point, and the new image data and the CHM image data have the same grid;

carrying out edge detection processing on the new image data to obtain the outline of the single tree crown;

affine transforming the row number and the column number of the outline of the single tree crown into real geographic coordinates to obtain a target segmentation space;

and extracting point cloud data in the target partition space from the original point cloud data to obtain point cloud data of single plant vegetation.

2. The method for segmenting point cloud data of individual vegetation according to claim 1, wherein the binary classification network adopts a RandLA-Net network or an improved structure based on the RandLA-Net network.

3. The method of segmenting point cloud data of individual vegetation according to claim 1, wherein the step of applying mathematical morphology to determine vegetation cap points as local extreme points from the CHM image data comprises:

performing image noise removal processing on the CHM image data by adopting a median filtering method to obtain new CHM image data;

and determining vegetation crown points serving as local extreme points by applying mathematical morphology according to the new CHM image data.

4. The method of segmenting point cloud data of individual vegetation according to claim 1, wherein the step of applying mathematical morphology to determine vegetation cap points as local extreme points from the CHM image data comprises:

performing morphological corrosion operation processing on the CHM image data to obtain an image data corrosion result;

subtracting the image data corrosion result from the CHM image data to obtain an image data subtraction result;

determining the maximum value in a positive value area according to the image data subtraction result;

and determining the point corresponding to the maximum value as a vegetation crown point serving as a local extreme point.

5. The point cloud data segmentation method for vegetation cover plants as claimed in claim 1, wherein the step of extracting a trunk by applying a tensor voting method according to the original point cloud data comprises the following steps:

traversing each of the original point clouds in the original point cloud data: for a certain original point cloud

Firstly, constructing a corresponding equation:

in the formula (I), wherein,

the tensor of the rod is expressed,

the representation is in the certain original point cloud

The total number of point clouds within a specified radius of the center,

a natural number is represented by a number of characters,

representing the certain original point cloud

And the second within the specified radius

Personal neighborhood point cloud

A feature vector in between and having

，

A tensor operator representing the effect of the rod tensor on the neighborhood point and having

Then solving the equation to obtain the rod tensor

And finally, if the maximum characteristic value in the at least one characteristic value is judged to be larger than a preset characteristic threshold value, determining the certain original point cloud

The characteristic points of the trunk are taken;

and extracting all trunk characteristic points in the original point cloud data to obtain a trunk.

6. The point cloud data partitioning method for single vegetation as set forth in claim 1, wherein the growth restriction includes a tree height restriction, a growth range restriction and/or an adjacent tree competition point restriction, wherein the tree height restriction includes a minimum value of the tree height H of the seed points being 1.5 m, the growth range restriction includes a maximum distance R between the obtained growth point and the seed point being 3 m, and the adjacent tree competition point restriction includes a growth point competing with the adjacent tree being 0.7 times higher than the tree height H of the seed point.

7. The method for segmenting the point cloud data of the individual vegetation according to claim 1, wherein the original point cloud data is multi-source point cloud data obtained by combining, registering and fusing vehicle-mounted laser radar point cloud data, vehicle-mounted laser radar point cloud data and ground station laser radar point cloud data, the interpolation processing adopts a triangulation network interpolation method, and the edge detection processing adopts an openCV edge detection function.

8. A single plant vegetation point cloud data segmentation device is characterized by comprising a vegetation point identification unit, an interpolation processing unit, an extreme point determination unit, a trunk extraction unit, an extreme point correction unit, a crown growth unit, a mark processing unit, an edge detection unit, an affine transformation unit and a data extraction unit;

the vegetation point identification unit is used for inputting original point cloud data to be segmented into a vegetation point identification model which is based on a binary classification network and is trained, and outputting to obtain a vegetation point identification result, wherein the original point cloud data is multi-source point cloud data fused with top view point cloud and side view point cloud;

the interpolation processing unit is in communication connection with the vegetation point identification unit and is used for interpolating the point cloud data of the identified vegetation points in the vegetation point identification result into rasterized canopy height model CHM image data;

the extreme point determining unit is in communication connection with the interpolation processing unit and is used for determining vegetation crown points serving as local extreme points by applying mathematical morphology according to the CHM image data;

the trunk extraction unit is used for extracting a trunk by applying a tensor voting method according to the original point cloud data;

the extreme point correcting unit is respectively in communication connection with the extreme point determining unit and the trunk extracting unit, and is used for feeding back the central point of the trunk serving as a new extreme point to the CHM image data to correct the local extreme point to obtain a corrected vegetation crown point;

the crown growing unit is in communication connection with the extreme point correcting unit and is used for performing crown growing processing on the corrected vegetation top crown point according to a growth limiting condition predetermined based on the actual data situation to obtain a growing point;

the mark processing unit is in communication connection with the crown growing unit and is used for marking the same mark on the growing point and the corrected vegetation crown point and writing the mark into new image data as a pixel value, wherein the mark is a numerical value number which is selected from a value range of the pixel value and used for marking the growing point and the corrected vegetation crown point, and the new image data and the CHM image data have the same grid;

the edge detection unit is in communication connection with the mark processing unit and is used for carrying out edge detection processing on the new image data to obtain the outline of the single tree crown;

the affine transformation unit is in communication connection with the edge detection unit and is used for affine transforming the row number and the column number of the outline of the single tree crown into real geographic coordinates to obtain a target segmentation space;

the data extraction unit is in communication connection with the affine transformation unit and is used for extracting point cloud data located in the target segmentation space from the original point cloud data to obtain single plant vegetation point cloud data.

9. A computer device, comprising a memory, a processor and a transceiver which are sequentially connected in communication, wherein the memory is used for storing a computer program, the transceiver is used for transmitting and receiving messages, and the processor is used for reading the computer program and executing the individual vegetation point cloud data segmentation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, which when executed on a computer perform the method of point cloud data segmentation of individual vegetation as claimed in any one of claims 1 to 7.

Images