A high-throughput stem-leaf separation method for field corn using multi-platform laser radar

Point cloud data of Datian corn is obtained through multi-platform lidar, and the stem and leaf separation is separated using benchmark cylinders and optimized cylinder models. Combined with statistical outlier removal and singular value decomposition, the problem of stem and leaf separation of Datian corn is solved, achieving high Extraction of maize phenotypic parameters and screening of ideal plant traits provide scientific support for corn breeding and promote the large-scale development of breeding.

CN114283163BActive Publication Date: 2024-12-03CHANGAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210017232.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-07
Publication Date
2024-12-03
Estimated Expiration
2042-01-07

AI Technical Summary

Technical Problem

The existing stem and leaf separation methods are mainly suitable for forest or single-plant corn, making it difficult to achieve high-throughput stem and leaf separation in field corn, resulting in accurate and high-throughput corn phenotype parameter extraction and ideal plant trait screening difficult to achieve. .

Method used

Multi-platform lidar is used to obtain point cloud data of Datian corn. By establishing a benchmark cylinder and optimizing cylindrical model, the point cloud data is preprocessed and initially separated, and combined with statistical outlier removal and singular value decomposition methods, the separation results are optimized. Achieve accurate separation of stems and leaves of corn in the field.

Benefits of technology

The high-throughput stem and leaf separation of large fields corn has been achieved, providing accurate phenotypic parameter extraction and ideal plant trait screening for corn breeding, promoting the large-scale development of corn breeding, reducing costs and improving breeding efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114283163B_ABST
    Figure CN114283163B_ABST
Patent Text Reader

Abstract

The invention discloses a multi-platform laser radar field corn high-throughput stem-leaf separation method, including using laser radar scanning to obtain field corn point cloud data; pre-processing the obtained point cloud data; using a corn stem-leaf separation model to preliminarily separate the corn stems and leaves; optimizing the separation results, and the specific optimization steps are: using statistical outlier removal SOR filtering to remove noise points in the stems, and adding the noise points to the leaves; using an optimized cylinder to traverse the separated stems, counting the difference between the highest point and the lowest point in each optimized cylinder, setting a threshold, and if the difference is less than the threshold, the points in the optimized cylinder are separated from the stems and added to the leaf points. The invention realizes high-precision stem-leaf separation of field corn, provides a basis for accurate and high-throughput extraction of corn phenotypic parameters, provides scientific methods and technical support for the screening of ideal plant type traits of corn, and is of great significance to accelerating the corn breeding process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of high-throughput stem-leaf separation of field corn, and specifically to a multi-platform laser radar high-throughput stem-leaf separation method for field corn. Background Art

[0002] Accurate and high-throughput extraction of crop phenotypic parameters is the key to breeding. The premise of using LiDAR to extract accurate phenotypic parameters is stem-leaf separation. However, due to the discrete, irregular, and non-topological characteristics of crop point cloud data obtained by LiDAR and the difficulty of processing massive data, the segmentation and recognition of corn point cloud data is still quite challenging, making it difficult to separate stems and leaves and extract phenotypic parameters.

[0003] Existing stem-leaf separation methods mainly include feature-based, model-based and learning-based methods. The feature-based method obtains the intensity, shape and waveform of the point cloud and separates the stems and leaves according to the different characteristics of the stems and leaves; the model-based method uses a variety of primitives to fit corn, such as fitting the stems with cylinders, and then combines the region growing algorithm to separate the stems and leaves; the learning-based method obtains local eigenvalues, then establishes training samples, uses machine learning algorithms such as support vector machines and random forests to achieve stem-leaf separation, and uses deep learning methods to automatically extract features from massive data for stem-leaf separation. However, the existing methods mainly achieve the stem-leaf separation of forest or single-plant scale corn, and there are almost no methods for stem-leaf separation of field corn. High-throughput stem-leaf separation of field corn can provide a basis for accurate and high-throughput extraction of corn phenotypic parameters, and provide scientific methods and technical support for the screening of ideal plant traits of corn. Therefore, it is necessary to achieve high-throughput stem-leaf separation of field corn. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-platform laser radar high-throughput stem and leaf separation method for field corn. In view of the fact that there is currently almost no research focus on the separation of stems and leaves of field corn, the present invention aims to achieve high-throughput stem and leaf separation of field corn using multi-platform laser radar, provide a basis for accurate and high-throughput extraction of corn phenotypic parameters, provide scientific methods and technical support for the screening of ideal plant traits of corn, and promote the large-scale development of corn breeding.

[0005] The present invention provides a multi-platform laser radar high-throughput stem and leaf separation method for field corn, which is characterized by comprising the following steps:

[0006] Use LiDAR scanning to obtain point cloud data of corn in the field;

[0007] Preprocessing the acquired point cloud data;

[0008] The corn stem and leaf separation model is used to perform preliminary separation of corn stems and leaves. The specific separation steps are as follows:

[0009] Establish multiple reference cylinders, use multiple reference cylinders to traverse all points of the preprocessed point cloud data, calculate the multidimensional eigenvalues ​​and angles of the traversed points in the reference cylinders, and realize the preliminary separation of corn stems and leaves according to the differences in multidimensional eigenvalues ​​and angles of corn stalks and leaves;

[0010] Optimize the separation results. The specific optimization steps are:

[0011] Statistical outlier removal (SOR) filtering was used to remove noise points in the stems and add noise points to the leaves;

[0012] An optimized cylinder is used to traverse the separated stems, and the difference between the highest point and the lowest point in each optimized cylinder is counted. A threshold is set. If the difference is less than the threshold, the points in the optimized cylinder are separated from the stems and added to the leaf points. The diameter of the optimized cylinder is smaller than the diameter of the reference cylinder, and its height is greater than the height of the reference cylinder.

[0013] Preferably, the steps of obtaining the multi-dimensional eigenvalues ​​and angles of the points traversed in the reference cylinder are as follows:

[0014] A plane is fitted for all points in each of the reference cylinders, the sum of distances from the fitting plane to all points in the reference cylinder is minimized, and multi-dimensional eigenvalues ​​and angles of the fitting plane are obtained.

[0015] Preferably, the constraint conditions of the fitting plane are:

[0016] a 2 +b 2 +c 2 =1

[0017] The fitting plane equation is:

[0018] ax+by+cz=d

[0019] The three-dimensional coordinates of all points in the reference cylinder are (x i ,y i ,z i ), i=1,2,3...n;

[0020] Substitute the three-dimensional coordinates of all points in the reference cylinder into the fitting plane equation to obtain

[0021] ax i +by i +cz i =d, i=1,2,3...n (1)

[0022] Let the average coordinates of all points in the reference cylinder be And put it into the fitting plane equation, we get

[0023]

[0024] Subtracting equation (1) from equation (2), we get

[0025]

[0026] Let matrix

[0027]

[0028] Column Matrix

[0029]

[0030] Then formula (3) is equivalent to AX = 0,

[0031] Performing singular value decomposition on the matrix A can obtain the eigenvalues ​​and eigenvectors of the fitting plane, that is, the multi-dimensional eigenvalues ​​and angles of the fitting plane.

[0032] Preferably, the method for obtaining point cloud data of field corn by laser radar includes obtaining point cloud data of field corn by ground-based laser radar and obtaining point cloud data of field corn by backpack laser radar;

[0033] The ground-based laser radar obtains the point cloud data of field corn, including setting different laser radar scanning parameters for corn in different growth stages, and obtaining the point cloud data of field corn through a multi-site scanning method;

[0034] The backpack laser radar acquires the point cloud data of field corn by planning a scanning route according to the actual field environment, and adopts a figure-eight method to eliminate the inertial navigation error before and after scanning.

[0035] Preferably, the preprocessing of the acquired point cloud data includes:

[0036] Solving and stitching point cloud data obtained by ground-based LiDAR, and solving point cloud data obtained by backpack LiDAR;

[0037] De-noising of point cloud data, filtering of ground point cloud data, and segmentation of point cloud data of field corn.

[0038] Preferably, the scanning parameters of the laser radar include scanning resolution / quality, horizontal scanning angle, and vertical scanning angle.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The invention proposes a high-throughput stem-leaf separation method for field corn using a multi-platform laser radar. The method uses the point cloud data of field corn obtained by ground-based laser radar and backpack laser radar platforms, adopts a corn stem-leaf separation model to perform preliminary separation of corn stems and leaves, and uses statistical outlier removal (SOR), filtering and cylindrical traversal to calculate height difference to optimize the separation result. The invention realizes the separation of stems and leaves of field corn, provides a basis for accurate and high-throughput extraction of corn phenotypic parameters, provides scientific methods and technical support for the screening of ideal plant traits of corn, and is of great significance to accelerating the process of corn breeding.

[0041] As the price of LiDAR sensors continues to drop, this method can serve corn breeding in a low-cost, efficient and accurate manner, allowing breeding experts to quickly obtain field corn phenotypic information, and then conduct gene association analysis based on the phenotypic information, accelerate the corn breeding process, and improve breeding efficiency and level. It has good application prospects and huge market value. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0043] Figure 1 It is a flowchart of the present invention;

[0044] Figure 2 is a schematic diagram of multi-dimensional characteristic values ​​of corn leaves and stalks in the present invention;

[0045] Figure 3 is a schematic diagram of the angles of corn leaves and stalks in the present invention;

[0046] Figure 4 This is a diagram showing the separation results of stems and leaves of different varieties of corn when the planting density is 45,000 plants / hectare in an embodiment of the present invention;

[0047] Figure 5 This is a diagram showing the separation results of corn stems and leaves at different planting densities, different growth periods and different platforms in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention, but it should be understood that the protection scope of the present invention is not limited by the specific implementation methods.

[0049] Example

[0050] like Figure 1-5 As shown, a multi-platform laser radar high-throughput stem and leaf separation method for field corn includes the following steps:

[0051] Step 1, use laser radar to scan and obtain the point cloud data of field corn. The method of laser radar to obtain the point cloud data of field corn includes ground-based laser radar to obtain the point cloud data of field corn and backpack laser radar to obtain the point cloud data of field corn; the ground-based laser radar to obtain the point cloud data of field corn includes setting different laser radar scanning parameters for corn in different growth periods; the laser radar scanning parameters include scanning resolution / quality, horizontal scanning angle, and vertical scanning angle. The point cloud data of field corn is obtained by multi-site scanning. The backpack laser radar to obtain the point cloud data of field corn includes planning the scanning route according to the actual field environment, and eliminating the inertial guidance error by the eight-shaped method before and after scanning. When collecting point cloud data, try to keep horizontal and uniform speed.

[0052] Step 2: preprocess the acquired point cloud data. The preprocessing of the acquired point cloud data includes: solving and splicing the point cloud data acquired by the ground-based lidar, solving the point cloud data acquired by the backpack lidar, denoising the point cloud data, filtering the ground point cloud data, and performing cell segmentation on the acquired point cloud data of field corn.

[0053] Step 3: Use the corn stem and leaf separation model to perform preliminary separation of corn stems and leaves. The specific separation steps are as follows:

[0054] Step 3.1, establish multiple reference cylinders;

[0055] Step 3.2, using multiple reference cylinders to traverse all points of the pre-processed point cloud data;

[0056] Step 3.3, calculate the multi-dimensional eigenvalues ​​and angles of the points traversed in the reference cylinder, including:

[0057] Step 3.3.1, fit a plane to all points in each reference cylinder, minimize the sum of distances from the fitted plane to all points in the reference cylinder, and obtain the multi-dimensional eigenvalues ​​and angles of the fitted plane;

[0058] The constraints for fitting the plane are:

[0059] a 2 +b 2 +c 2 =1

[0060] The equation of the fitted plane is:

[0061] ax+by+cz=d

[0062] The three-dimensional coordinates of all points on the reference cylinder are (x i ,y i ,z i), i=1,2,3...n;

[0063] Step 3.3.2: Substitute the three-dimensional coordinates of all points in the reference cylinder into the fitting plane equation to obtain:

[0064] ax i +by i +cz i =d, i=1,2,3...n (1)

[0065] Step 3.3.3: Let the average coordinates of all points in the reference cylinder be And put it into the fitting plane equation, we get:

[0066]

[0067] Subtracting equation (1) from equation (2), we get:

[0068]

[0069] Step 3.3.4, set the matrix:

[0070]

[0071] Column Matrix

[0072]

[0073] Then formula (3) is equivalent to AX = 0,

[0074] Step 3.3.5: Perform singular value decomposition on matrix A to obtain the eigenvalues ​​and eigenvectors of the fitting plane, that is, the multi-dimensional eigenvalues ​​and angles of the fitting plane.

[0075] Step 3.4: Based on the differences in multi-dimensional feature values ​​and angles of corn stalks and leaves, the corn stalks and leaves are initially separated.

[0076] Step 4: Optimize separation results;

[0077] There are two problems with the initial stem-leaf separation results. Very few leaf points, such as leaf tips and relatively narrow leaf points, are classified into the stems, and there are some inaccurately segmented points at the intersection of leaves and stems. The separation results need to be optimized. The specific optimization steps are:

[0078] Step 4.1, using statistical outlier removal SOR filtering to remove noise points in the stems and add noise points to the leaves;

[0079] Step 4.2: Use the optimized cylinder to traverse the separated stems, count the difference between the highest point and the lowest point in each optimized cylinder, set a threshold, and if the difference is less than the threshold, the point in the optimized cylinder is separated from the stem and added to the leaf point; the diameter of the optimized cylinder is smaller than the diameter of the reference cylinder, and its height is greater than the height of the reference cylinder.

[0080] The present invention analyzes the structural characteristics of corn stalks and leaves and separates the stems and leaves according to the different structural characteristics, which are:

[0081] Eigenvalue: The local points of corn stalks are similar to one-dimensional distribution, i.e., lines, while the local points of leaves are similar to two-dimensional distribution, i.e., surfaces. This feature can be used to separate the stems and leaves of corn.

[0082] Angle: The angle between the stem and the vertical direction is very small or almost zero, while the angle between the leaves and the vertical direction is a certain angle.

[0083] Since the tip of the leaf is narrow and similar to the stem, it is not possible to separate the corn stem and leaf well by using only the eigenvalue, while the angle feature can make up for this problem well. However, the upper leaves of most varieties of corn grow vertically upward, with a small angle with the vertical direction, and have the same characteristics as the stem, so only using the angle cannot separate the upper leaves. Combining these two parameters makes up for the problems of a single parameter and can separate the stem and leaf of corn with higher accuracy.

[0084] Machine inspection training of the corn stem-leaf separation model includes:

[0085] 70% of the corn in the experimental field was selected, the stalks were marked as 0, and the leaves were marked as 1, and the multi-dimensional characteristic values ​​and angles of the stalks and leaves of 70% of the corn in the experimental field were calculated respectively, so as to achieve the separation of the stems and leaves of 70% of the corn in the experimental field;

[0086] Calculate the multidimensional characteristic values ​​and angles of the remaining 30% of corn stalks and leaves in the experimental field;

[0087] 70% of the corn in the experimental field was used as training samples, and the remaining 30% of the corn in the experimental field was used as prediction samples. The random forest model was used to predict the prediction samples based on the training samples to achieve preliminary stem-leaf separation of the prediction sample corn. The multi-dimensional eigenvalues ​​and angles of the stems and leaves of the remaining 30% of the corn in the experimental field that were actually calculated were tested and compared with the prediction results.

[0088] The field corn of the example of the present invention is located in the Xiaotangshan National Precision Agriculture Demonstration Base in Changping District, Beijing, with a geographical location of 115°50′17″~116°29′49″E, 40°20′18″~40°23′13″N. The agricultural demonstration base includes 15 breeding plots, including 5 varieties, with 3 replicates in each variety. The size of each plot is 3.6m×2.5m, and the plot interval is 1.5m. The planting density in the area is 40,000 plants / hectare, and the 5 varieties are Zhengdan 958, Xianyu 335, Jingnongke 728, Chengdan 30, and Jingpin 6, and are marked as A1, A2, A3, A4, and A5 respectively.

[0089] The ground-based LiDAR data was acquired on September 12, 2019. Before acquiring the data, 6 control points were evenly distributed in the test area. In this example, a 1m square box was used for the registration of data from different scanning stations. A total of 15 stations were deployed in 4 small areas, with 5 stations in each of the four corners and one in the middle. The ground-based LiDAR is FARO FocusS 350, and the scanning parameters are as follows: resolution / quality is 43.7Mpts / 3x, horizontal scanning angle is 0°~360°, vertical scanning angle is -60°~90°, and scanning time per station is 2′45″.

[0090] The preprocessing of ground-based lidar data mainly includes four parts: multi-site scanning data solution and stitching, denoising, cell segmentation and ground point filtering.

[0091] Multi-site scan data analysis and stitching: FARO SCENE software is used to scan and stitch the acquired ground-based LiDAR data. FARO SCENE can efficiently and easily process and manage scan data through automatic object recognition, scan stitching and positioning.

[0092] Denoising: The SOR algorithm integrated in the CloudCompare software is used to complete the denoising by determining two parameters: the standard deviation multiplier threshold and the number of points used to estimate the average distance.

[0093] Cell segmentation: First, draw the vector boundary lines of each cell based on the point cloud data, and adjust the boundary lines and point cloud data to be at the same height. Then, complete the cell segmentation of each route data in batches in the CloudCompare software.

[0094] Ground point filtering: Use the "cloth" filtering algorithm ClothSimulation Filter, CSF, to separate vegetation point cloud and ground point cloud.

[0095] like Figure 2-3As shown in the figure, according to the different structures of corn stalks and leaves, a cylinder is used to traverse the corn. The cylinder is 0.11m high and 0.7m in diameter. A plane is fitted to the points in the cylinder and its multidimensional eigenvalues ​​and angles are extracted. Then, random forest is used for training and classification to complete the separation of corn stems and leaves. Figure 4 shown.

[0096] from Figure 5 It can be seen that the method provided by the present invention can well separate the stems and leaves of corn of the density and variety. And the stems and leaves of corn with different planting densities, different growth periods and different platforms are separated. This further proves the robustness of the method provided by the present invention. Among them, Figure 5 (b) The corn lodging in the black frame area affected the stem-leaf separation results.

[0097] The invention realizes the high-precision separation of stems and leaves of field corn.

[0098] Finally, it should be noted that what is disclosed above is only a specific embodiment of the present invention, but the embodiment of the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A multi-platform laser radar high-throughput stem-leaf separation method for field corn, characterized in that: The steps include: Use LiDAR scanning to obtain point cloud data of corn in the field; Preprocessing the acquired point cloud data; The corn stem and leaf separation model is used to perform preliminary separation of corn stems and leaves. The specific separation steps are as follows: Establish multiple reference cylinders, use multiple reference cylinders to traverse all points of the preprocessed point cloud data, calculate the multidimensional eigenvalues ​​and angles of the traversed points in the reference cylinders, and realize the preliminary separation of corn stems and leaves according to the differences in multidimensional eigenvalues ​​and angles of corn stalks and leaves; Optimize the separation results. The specific optimization steps are: Statistical outlier removal (SOR) filtering was used to remove noise points in the stems and add noise points to the leaves; An optimized cylinder is used to traverse the separated stems, and the difference between the highest point and the lowest point in each optimized cylinder is counted. A threshold is set. If the difference is less than the threshold, the points in the optimized cylinder are separated from the stems and added to the leaf points. The diameter of the optimized cylinder is smaller than the diameter of the reference cylinder, and its height is greater than the height of the reference cylinder.

2. A multi-platform laser radar high-throughput stem-leaf separation method for field corn according to claim 1, characterized in that: The specific steps of calculating the multi-dimensional eigenvalues ​​and angles of the points traversed in the reference cylinder are as follows: A plane is fitted for all points in each of the reference cylinders, the sum of distances from the fitting plane to all points in the reference cylinder is minimized, and multi-dimensional eigenvalues ​​and angles of the fitting plane are obtained.

3. A multi-platform laser radar field corn high-throughput stem and leaf separation method according to claim 2, characterized in that: The constraints of the fitting plane are: a 2 +b 2 +c 2 =1 The fitting plane equation is: ax+by+cz=d The three-dimensional coordinates of all points in the reference cylinder are (x i ,y i ,z i ), i=1,2,3...n; Substitute the three-dimensional coordinates of all points in the reference cylinder into the fitting plane equation to obtain ax i +by i +cz i =d,i=1,2,3...n (1) Let the average coordinates of all points in the reference cylinder be And put it into the fitting plane equation, we get Subtracting equation (1) from equation (2), we get Let matrix Column Matrix Then formula (3) is equivalent to AX = 0, Performing singular value decomposition on the matrix A can obtain the eigenvalues ​​and eigenvectors of the fitting plane, that is, the multi-dimensional eigenvalues ​​and angles of the fitting plane.

4. The multi-platform laser radar high-throughput stem-leaf separation method for field corn according to claim 1, characterized in that: The method for obtaining point cloud data of field corn by laser radar includes obtaining point cloud data of field corn by ground-based laser radar and obtaining point cloud data of field corn by backpack laser radar; The ground-based laser radar obtains the point cloud data of field corn, including setting different laser radar scanning parameters for corn in different growth stages, and obtaining the point cloud data of field corn through a multi-site scanning method; The backpack laser radar acquires the point cloud data of field corn by planning a scanning route according to the actual field environment, and adopts a figure-eight method to eliminate the inertial navigation error before and after scanning.

5. The multi-platform laser radar high-throughput stem-leaf separation method for field corn according to claim 1, characterized in that: The preprocessing of the acquired point cloud data includes: Solving and stitching point cloud data obtained by ground-based LiDAR, and solving point cloud data obtained by backpack LiDAR; De-noising of point cloud data, filtering of ground point cloud data, and segmentation of point cloud data of field corn.

6. The multi-platform laser radar high-throughput stem-leaf separation method for field corn according to claim 4, characterized in that: The scanning parameters of the laser radar include scanning resolution / quality, horizontal scanning angle, and vertical scanning angle.