A method and system for extracting phenotypic information of three-dimensional structure of alfalfa population

Abstract

The application provides a kind of alfalfa population three-dimensional structure phenotype information extraction method and system. First, according to the ground and wall plane in the original point cloud, the original point cloud is corrected in coordinate;Then according to the reason and characteristics of noise generated by original point cloud, based on statistical filtering removes outlier, based on crown voxel filtering removes redundant point, based on moving least square method removes mixed point;Then according to the characteristics of planting and growth of potted alfalfa population, the alfalfa population is extracted by using straight-through filtering and grid division method, and the single plant is segmented, and then six kinds of structure parameters such as maximum plant height, average plant height, plant height quantile, canopy cover, projected leaf area and three-dimensional volume are extracted.The application realizes the lossless automatic extraction of alfalfa population structure phenotype parameters, and the application is also suitable for three-dimensional structure phenotype information extraction of other population growth plants.

Description

Technical Field

[0001] This invention belongs to the fields of remote sensing and botany, and specifically relates to a method and system for extracting three-dimensional structural phenotypic information of alfalfa populations, applicable to the extraction of phenotypic information of plants with various population types. Background Technology

[0002] Alfalfa is a globally cultivated forage crop, rich in nutrients and often referred to as the "King of Forage." However, my country's alfalfa self-sufficiency is low, failing to meet the needs of the livestock industry, resulting in a long-term heavy reliance on imports. Improving alfalfa self-sufficiency is urgently needed to screen high-quality alfalfa germplasm resources. A key challenge in this screening is obtaining accurate phenotypic data. Traditional methods for acquiring phenotypic information mainly rely on manual measurement, which often damages the plant's structure, is time-consuming and labor-intensive, difficult to conduct on a large scale, and cannot guarantee accuracy. Laser point cloud data provides detailed three-dimensional information, making it possible to quickly extract alfalfa structural phenotypic parameters. While reports on methods for extracting three-dimensional structural phenotypic information from plants using lidar are increasing, there are few reports on extracting three-dimensional structural phenotypic information from forage populations like alfalfa. This invention fills the gap in methods for extracting three-dimensional structural phenotypic information from alfalfa populations. Summary of the Invention

[0003] The technical problem this invention aims to solve is to overcome the shortcomings of existing technologies and provide a method and system for extracting three-dimensional structural phenotypic information of alfalfa populations. This includes correcting the coordinates of the original alfalfa point cloud; removing noise points caused by systematic or random errors; obtaining alfalfa population point cloud data and individual plant point cloud data through population and individual plant segmentation; and realizing the extraction of three-dimensional structural phenotypic information from alfalfa populations. This invention achieves non-destructive and automatic extraction of alfalfa population structural phenotypic parameters and is also applicable to the extraction of three-dimensional structural phenotypic information from other plant populations.

[0004] The technical solution adopted in this invention is a method for extracting three-dimensional structural phenotypic information of alfalfa populations, comprising the following steps:

[0005] Step (1): Calculate the appropriate rotation transformation angle based on the ground normal and wall direction of the original point cloud, perform point cloud coordinate correction, and obtain the point cloud data after coordinate correction;

[0006] Step (2): Analyze the causes and characteristics of point cloud noise points, and based on step (1), selectively remove various types of point cloud noise to obtain denoised point cloud data;

[0007] Step (3): Based on the denoised point cloud data obtained in step (2), and according to the characteristics of potted alfalfa planting, perform population extraction and single plant segmentation to obtain alfalfa population point cloud data and single plant point cloud data.

[0008] Step (4): Based on the alfalfa population point cloud data and single plant point cloud data obtained in step (3), the population structure phenotypic parameters are automatically extracted, and finally the three-dimensional structure phenotypic information of the alfalfa population is obtained.

[0009] The point cloud coordinate correction steps in step (1) are as follows:

[0010] Step (11): Use the random sampling consensus algorithm to detect the ground of the original point cloud and obtain the ground normal vector in the current coordinate system;

[0011] Step (12): Based on the original point cloud ground normal vector obtained in step (11), combined with the Z-axis direction... The rotation angle is calculated using (0,0,1);

[0012] Step (13): Based on the rotation angle obtained in step (12), ground correction is achieved using the Rodriguez rotation formula;

[0013] Step (14): Determine the angle between the wall and the X-axis based on the orientation of the wall in the original point cloud;

[0014] Step (15): Based on the angle obtained in step (14), plane correction is achieved by combining the Bursa seven-parameter model.

[0015] The point cloud denoising steps in step (2) are as follows:

[0016] Step (21): Statistical filtering is used to remove outlier noise points;

[0017] Step (22): Voxel filtering is used to remove redundant noise points;

[0018] Step (23): For the mixed noise points floating on the surface of the alfalfa object but not belonging to the alfalfa itself, the moving least squares method is used to smooth and remove them.

[0019] The steps for dividing the alfalfa dot cloud population and individual plants in step (3) are as follows:

[0020] Step (31): In order to remove invalid information such as basins and ground in the point cloud, the group segmentation is performed by using the pass-through filtering method based on the three-dimensional point cloud coordinates of the target area, and then the alfalfa group segmentation result is obtained.

[0021] Step (32): Based on the characteristics of potted alfalfa, the grid division method is used to extract individual plants. That is, the position of the first alfalfa plant in the group is automatically located according to the pre-deployed calibration column. The root of the individual plant is the center, and the planting distance is used to divide the plant into regular grids. Each grid is used as the current growth area of ​​the alfalfa plant, and the alfalfa plant segmentation result is obtained.

[0022] The steps for extracting the three-dimensional structural phenotypic information of the alfalfa population in step (4) are as follows:

[0023] Step (41): Based on the alfalfa plant segmentation results obtained in step (3), extract the three-dimensional structural phenotypic information of alfalfa plants, including plant height, canopy coverage, projected leaf area, and three-dimensional volume.

[0024] Step (42): Based on the results obtained in step (41), further calculate the maximum plant height, average plant height, plant height quantile, canopy coverage, projected leaf area and three-dimensional volume of the alfalfa population.

[0025] This invention also proposes a system for extracting three-dimensional structural phenotypic information of alfalfa populations, comprising:

[0026] Point cloud coordinate correction module: Based on the ground normal and wall direction of the original point cloud, calculate the rotation transformation angle of the alfalfa point cloud and perform coordinate correction of the original alfalfa point cloud;

[0027] Point cloud denoising module: Provides functions such as statistical filtering to remove outliers, subcoronal voxel filtering to remove redundant points, and moving least squares to remove mixed points;

[0028] Alfalfa population extraction and individual plant segmentation module: Provides pass-through filtering and grid division functions to extract alfalfa populations and segment individual plants from point cloud data;

[0029] Alfalfa population 3D structure phenotypic information extraction module: Automatically extracts six types of alfalfa population 3D structure phenotypic information, including maximum plant height, average plant height, height quantile, canopy cover, projected leaf area, and 3D volume.

[0030] Point cloud visualization module: Enables visualization processing and manipulation of alfalfa point cloud data on a computer screen.

[0031] The advantages of this invention compared to the prior art are:

[0032] (1) There are few reports on automatic extraction methods for three-dimensional structural phenotypic information of alfalfa populations. Compared with traditional alfalfa structural phenotypic information extraction methods, this invention fills the gap in automatic extraction of three-dimensional structural phenotypic information of alfalfa populations, greatly reduces labor input, and is conducive to long-term dynamic monitoring of alfalfa growth.

[0033] (2) In terms of redundancy removal of point clouds, traditional methods generally perform voxel filtering on the entire original point cloud. In this invention, the detailed information of the alfalfa canopy is preserved and voxel filtering is performed only on the part below the canopy. This preserves the canopy information and improves the efficiency of point cloud processing.

[0034] (3) The present invention provides a one-stop streamlined alfalfa three-dimensional structural phenotypic information extraction system. The system has a visual operation graphical interface, which simplifies the difficulty and operability of phenotypic information extraction and makes it easy for non-technical personnel to use. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the method and system implementation of the alfalfa population three-dimensional structural phenotypic information extraction method of the present invention;

[0036] Figure 2 This is the result of alfalfa population segmentation in this invention;

[0037] Figure 3 This is the result of alfalfa plant segmentation in this invention;

[0038] Figure 4 This is an example of the extraction results of plant height in the alfalfa population of the present invention. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0040] like Figure 1 As shown, the method for extracting three-dimensional structural phenotypic information of alfalfa populations according to the present invention includes the following main steps:

[0041] Step 1: Ground Correction. First, obtain the ground normal vector in the current point cloud coordinate system. Calculate the rotation angle and rotation matrix based on the angle between the normal vector and the Z-axis normal vector. This invention uses a random sample consensus algorithm to detect the ground and obtain the ground normal vector in the current coordinate system. Combined with the Z-axis normal vector The rotation angle θ is calculated from (0,0,1) using the following formula:

[0042]

[0043]

[0044] In the formula, m and n are respectively The length.

[0045] The rotation matrix is ​​obtained using the Rodriguez rotation formula, as follows:

[0046]

[0047]

[0048] In the formula, Let θ be a third-order identity matrix, and θ be the rotation angle. (d1,d2,d3) is The unit vector.

[0049] Planar correction. Based on the wall orientation in the original point cloud (the alfalfa planting pots are placed in the same direction as the wall), determine its angle with the X-axis. After transformation, the horizontal direction of the alfalfa point cloud is parallel to the X-axis. The transformation equation uses the Bursa seven-parameter model, as shown in the following formula:

[0050]

[0051] In the formula, ΔX, ΔY, and ΔZ are translational parameters, m is a scale parameter, and R... x R y R z Let be the rotation parameters, and α, β, and γ be the angles of rotation about the X, Y, and Z coordinate axes, respectively. The formulas for the rotation parameters are as follows:

[0052]

[0053] Step 2: The statistical filtering process for removing outliers is as follows:

[0054] 1) Establish spatial topological relationships of point clouds based on KD trees to facilitate point cloud search;

[0055] 2) Traverse the point cloud, calculate the average distance between each point and all points in its k-neighborhood, and establish a Gaussian distribution with the average distance D. aver Represented as:

[0056]

[0057] In the formula, k is the number of neighborhood points, and d i Let be the distance from a point to its i-th neighboring point;

[0058] 3) Set a standard range to determine whether a point belongs to outlier noise, using the following formula:

[0059] S=μ+σ*α (8)

[0060] In the formula, μ is the average distance between all points and all points in their k-neighborhood, σ is the standard deviation, and α is the standard deviation factor.

[0061] 4) Based on the average distance D aver Outlier removal is performed within the standard range S. A point is defined as an outlier and removed when its average distance exceeds the standard range; otherwise, it is retained.

[0062] The specific steps for voxel filtering to remove redundant points are as follows:

[0063] 1) Determine the maximum value x in the X, Y, and Z directions based on the point cloud spatial coordinates. max y max zmax and minimum value x min y min z min ;

[0064] 2) Determine the length, width, and height of the point cloud volume using the difference between the maximum and minimum values ​​in each dimension, as shown in the following formula:

[0065] x box =x max -x min

[0066] y box =y max -y min (9)

[0067] z box =z max -z min

[0068] 3) Divide the space volume into n parts with side length l x l y l z The formula for the small voxel V is as follows:

[0069]

[0070] 4) After uniformly dividing the point cloud space volume into voxels, determine whether the voxel belongs to the canopy. If it does, skip the voxel. Otherwise, use the centroid of all points in the voxel to approximate the other points in the voxel. That is, remove the canopy and only keep the center point of all points in the voxel. Filter out the remaining points. After processing all voxels, the filtered point cloud is obtained.

[0071] The specific steps for smoothing out redundant points using the moving least squares method are as follows:

[0072] 1) Establish spatial topological relationships of point clouds based on KD trees to facilitate point cloud search;

[0073] 2) Set the search radius. Establish a reference plane H based on a point a and all points in its neighborhood. Find the optimal plane model by minimizing the L2 norm, which minimizes the sum of the distances from all points in the neighborhood of the point to the plane. The formula is as follows:

[0074]

[0075] In the formula, N is the number of neighborhood points, p i Let be the i-th point in the neighborhood, n be the normal to the reference plane, q be the projection of point a onto the reference plane H, and θ(d) be the Gaussian weighted function, a non-negative and monotonic smoothly decreasing function, expressed as follows:

[0076]

[0077] In the formula, d is the distance from the i-th nearest neighbor to point a, and h is the Gaussian coefficient. The smaller the value, the smoother the point is.

[0078] 3) Perform local polynomial surface fitting based on the reference plane using the weighted least squares method, as shown in the following formula:

[0079]

[0080] In the formula, g is the locally fitted surface, (x i ,y i Let f be the projected coordinates of the i-th point in the neighborhood onto the reference plane. i =n*(p i -q) is the perpendicular distance from the i-th point in the neighborhood to the reference plane, q i It is a neighboring point p i Projection onto reference plane H;

[0081] 4) After the optimal polynomial surface fitting is completed, the projected coordinates of point a can be obtained. This process then smooths the point cloud. The operation is repeated for all points until a smoothed point cloud is obtained.

[0082] Step 3: Alfalfa Population Segmentation. Using a pass-through filter, a threshold range is set based on the position of the alfalfa point cloud in the X, Y, and Z dimensions. Points outside this range are removed, retaining only those within the range. For example... Figure 2 As shown in the figure, this figure represents the extraction effect of alfalfa population. This method effectively removes invalid information from the point cloud and retains only the alfalfa population, which is beneficial for long-term dynamic non-destructive monitoring of alfalfa growth changes.

[0083] Alfalfa plant segmentation. A grid-based method was used. Based on pre-positioned calibration posts, the location of the first alfalfa plant within the quadrat was automatically determined. A regular grid was created, centered on the root of each plant and based on plant spacing. Each grid was intended to represent the growth area of ​​the current alfalfa plant. The extraction results for individual plants were as follows: Figure 3 As shown, each cell represents a single alfalfa plant;

[0084] Step 4: Calculate the height of individual alfalfa plants, and calculate the maximum plant height, average plant height, height quantile, canopy coverage, projected leaf area, and three-dimensional volume of the alfalfa population.

[0085] Based on the point cloud segmentation results of alfalfa plants, the plant height is calculated by subtracting the lowest Z-value from the highest Z-value within the plant region, using the following formula:

[0086] H max =Z max -Z min (14)

[0087] In the formula, H max Z is the calculated maximum height of the plant. max The maximum value in the Z-axis direction, Z min This is the minimum value in the Z-axis direction.

[0088] Projected leaf area, i.e., the projected area of ​​the alfalfa canopy. First, the lidar points of the alfalfa sample are projected onto the XY plane. Then, the projected points are rasterized at a resolution of 1 times the average point spacing. Pixels containing projected points are marked as 1, and those without are marked as 0. The area of ​​pixels with a value of 1 is recorded as the estimated projected leaf area; the ratio of the number of pixels with a value of 1 to the total number of pixels in the alfalfa sample on the XY plane is the canopy coverage, calculated using the following formula:

[0089]

[0090]

[0091] In the formula, Cano is the canopy coverage, PLA is the projected leaf area, and n l n p These represent the number of pixels containing laser points and the total number of pixels, respectively; Δx and Δy are the grid boundary lengths of the projected points in the x and y directions, respectively. The average plant height is represented as the average height within the grid containing all projected points.

[0092] The three-dimensional volume was calculated using a three-dimensional voxel model. The three-dimensional point cloud was voxelized at a resolution of 1 times the average point spacing, and all points were assigned to their respective voxels. The three-dimensional volume of the plant was obtained by multiplying the number of voxels containing points by the size of a single voxel. The calculation formula is as follows:

[0093]

[0094] In the formula, Vol is the three-dimensional volume, and n l n p These represent the number of voxels containing the laser points and the total number of voxels, respectively; Δx, Δy, and Δz are the lengths of the voxel borders in the x, y, and z directions, respectively.

[0095] Plant height quantiles are the proportion of alfalfa plants with a height greater than 80%. They are determined by judging the height of lidar points on alfalfa samples and statistically analyzing the proportion of points with a height greater than 80% of the maximum height.

[0096] Figure 4 This paper demonstrates the changes in a single phenotypic trait (plant height) of an alfalfa population throughout its entire growth cycle, clearly showing the dynamic changes of the alfalfa population and individual plants throughout the entire growth cycle. This illustrates that the present invention can automatically extract alfalfa population structural phenotypic parameters without damage. Furthermore, the present invention is also applicable to the extraction of three-dimensional structural phenotypic information from other plant populations.

[0097] The above description is merely an embodiment of the method for extracting three-dimensional structural phenotypic information of alfalfa populations according to the present invention. The present invention is not limited to the above embodiments. This specification is for illustrative purposes only and does not limit the scope of the claims. It will be apparent to those skilled in the art that many substitutions, improvements, and variations are possible. All technical solutions formed by equivalent substitutions or equivalent transformations fall within the protection scope claimed by the present invention.

Claims

1. A method for extracting three-dimensional structural phenotypic information of alfalfa populations, characterized in that, Includes the following steps: Step 1: Calculate the rotation transformation angle of the alfalfa point cloud based on the ground normal and wall orientation in the original point cloud, and perform coordinate correction of the original alfalfa point cloud. Step 2: Use point cloud denoising methods to remove various types of point cloud noise, namely outliers, redundant points, and mixed points, from the point cloud data after coordinate correction in Step 1; in order to preserve canopy information, use statistical filtering to remove outliers, subcanopy voxel filtering to remove redundant points, and moving least squares method to remove redundant points to obtain denoised point cloud data. Step 3: Based on the denoised point cloud data, according to the regular planting method of alfalfa and the growth characteristics of branches and leaves crossing in the later stage of growth, making it difficult to completely and clearly separate individual plants, the alfalfa point cloud data is extracted from the population and segmented into individual plants using the direct filtering and grid division method to obtain the alfalfa population segmentation results and alfalfa individual plant segmentation results. Step 4: Based on the alfalfa population segmentation results and alfalfa individual plant segmentation results obtained in Step 3, automatically extract the three-dimensional structural phenotypic parameters of the alfalfa population, including maximum plant height, average plant height, plant height quantile, canopy cover, projected leaf area, and three-dimensional volume. In step 3, the alfalfa dot cloud population segmentation and individual plant segmentation steps are as follows: Step (31): In order to remove invalid information in the original point cloud of the pot and soil, the population is segmented by using the pass-through filtering method based on the three-dimensional point cloud coordinates of the target area, and then the alfalfa population segmentation result is obtained. Step (32): Based on the characteristics of alfalfa potted planting and growth, and based on the alfalfa population segmentation results obtained in step (31), the grid division method is used to segment individual plants. That is, the position of the first alfalfa plant in the population is automatically located according to the pre-set calibration post. The root of the individual plant is the center, and the planting distance is used to divide the plant into regular grids. Each grid is used as the current alfalfa plant growth area, and the alfalfa plant segmentation results are obtained.

2. The method for extracting three-dimensional structural phenotypic information of alfalfa populations according to claim 1, characterized in that: The coordinate correction steps in step 1 are as follows: Step (11): Use the random sampling consensus algorithm to detect the ground of the original alfalfa point cloud and obtain the ground normal vector in the original point cloud under the current coordinate system; Step (12): Based on the original point cloud ground normal vector obtained in step (11), combined with the Z-axis direction The ground rotation angle is calculated using (0,0,1); Step (13): Using the Rodriguez rotation formula, perform ground correction on the ground rotation angle obtained in step (12) to obtain ground-corrected point cloud data; Step (14): Based on the wall direction in the original point cloud, i.e. the placement direction of the alfalfa planting pot is consistent with the wall direction, determine the angle between the wall and the X-axis in the original point cloud data to obtain the plane rotation angle. Step (15): Using the Bursa seven-parameter model, perform plane correction based on the plane rotation angle obtained in step (14) to obtain point cloud data after coordinate correction.

3. The method for extracting three-dimensional structural phenotypic information of alfalfa populations according to claim 1, characterized in that: The point cloud denoising method in step 2 includes: Step (21): Statistical filtering is used to remove outlier noise points; Step (22): For redundant point clouds generated by repeated scanning of the same area by the lidar, i.e. redundant noise points, the crown voxel filter is used to remove redundant points; Step (23): For points that float on the surface of alfalfa but do not belong to alfalfa itself, i.e. mixed noise points, the moving least squares method is used to smooth and remove them.

4. The method for extracting three-dimensional structural phenotypic information of alfalfa populations according to claim 1, characterized in that: The automatically extracted three-dimensional structural phenotypic parameters of the alfalfa population in step 4 are as follows: Step (41): Based on the alfalfa plant segmentation results obtained in step (3), extract the three-dimensional structural phenotypic information of alfalfa plants, including plant height, canopy coverage, projected leaf area, and three-dimensional volume. Step (42): Based on the results obtained in step (41), further calculate the maximum plant height, average plant height, plant height quantile, canopy coverage, projected leaf area and three-dimensional volume of the alfalfa population.

5. A system for implementing the method for extracting three-dimensional structural phenotypic information of alfalfa populations according to any one of claims 1-4, characterized in that... include: The system includes a point cloud coordinate correction module, a point cloud denoising module, alfalfa population extraction and single plant segmentation module, alfalfa population structure and phenotypic information extraction module, and a point cloud visualization module. Alfalfa 3D laser point clouds are imported into the system through the visualization module, and the functions of the point cloud coordinate correction module, point cloud denoising module, alfalfa population extraction and single plant segmentation module, and alfalfa population structure and phenotypic information extraction module are executed in sequence. The running results of each module are displayed through the visualization module. Point cloud coordinate correction module: Based on the ground normal and wall direction of the original point cloud, calculate the rotation transformation angle of the alfalfa point cloud and perform coordinate correction of the original alfalfa point cloud; Point cloud denoising module: To address the noise points, namely outliers, redundant points, and mixed points, the module provides statistical filtering to remove outliers, under-canopy voxel filtering to remove redundant points, and moving least squares method to remove redundant points. Alfalfa population extraction and individual plant segmentation module: To achieve alfalfa population extraction and individual plant segmentation, it provides pass-through filtering and grid division functions to extract alfalfa populations and segment individual plants from point cloud data; Alfalfa population 3D structure phenotypic information extraction module: Automatically extracts six types of alfalfa population 3D structure phenotypic information: maximum plant height, average plant height, height quantile, canopy cover, projected leaf area, and 3D volume. Point cloud visualization module: Enables visualization processing and manipulation of alfalfa point cloud data on a computer screen.

Images