A soybean branch angle extraction method based on point cloud

Abstract

The application discloses a soybean branch angle extraction method based on point cloud, comprising: layered clustering detection is carried out on pretreated soybean single plant point cloud data, and a branch position is obtained; based on the branch position, a layered cluster containing multiple clusters is selected, a median point of the cluster is calculated, and a bifurcation point is obtained; based on the bifurcation point, the branch position is optimized, and an optimized branch point is obtained; based on the bifurcation point and the optimized branch point, a branch angle is calculated, and the soybean branch angle extraction based on the point cloud is realized. On the basis of bifurcation point detection and branch point optimization, the branch angle can be calculated based on a space vector, and the high-throughput extraction efficiency of plant phenotypes is promoted; the method provided by the application can obtain branch angles at different heights of plants, and is used for crop plant type parameter evaluation, serves for excellent germplasm resource identification and breeding utilization, and through the provision of a new technical means of nondestructive, high-throughput and high-precision, it is beneficial to accelerate the breeding process and optimize the cultivation management measures.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of research on crop phenotype monitoring methods, and particularly relates to a soybean branch angle extraction method based on point cloud. BACKGROUND

[0002] The branch angle is an important plant type trait of soybean, and is closely related to the canopy structure and soybean yield. The branch angle generally refers to the angle between the branch and the main stem. The traditional measurement method relies on manual operation, mainly using an inclinometer, a protractor, and a 3D digital measuring instrument. The measurement is not only time-consuming and labor-intensive, but also easily affected by the subjective factors of the measurer, and is difficult to be used for large-scale measurement and accurate quantification of differences between different materials.

[0003] The branch angle extraction based on two-dimensional images improves the work efficiency to some extent. For example, the RGB image of oilseed rape is taken by a mobile phone, the paths of the main stem and the branch are manually drawn in the photo using AutoCAD, and then the branch angle is calculated according to the paths. The processing efficiency is improved compared with manual operation. With the development of image processing technology, automatic image processing tools developed based on MATLAB (such as Leaf Angle Extractor) are used to quantify the leaf angle of corn and sorghum from image data, greatly improving the automation level. However, these methods mainly focus on crops with simple canopy structure (such as symmetrical leaf distribution) such as oilseed rape, corn, and sorghum, and require specific angles for shooting. Moreover, for soybean crops with complex three-dimensional canopy structure, the image acquisition method is seriously disturbed by the plant organs or branch and leaf shielding.

[0004] In addition, a method for measuring the branch angle of soybean is to use a depth camera to perform three-dimensional reconstruction on the soybean canopy. Although the three-dimensional reconstruction method has depth data as support, which is conducive to more accurate measurement of the branch angle of soybean, the data quality based on three-dimensional reconstruction is greatly affected by light, and the reconstruction accuracy of plant objects with discontinuous surfaces is poor. Moreover, the data registration and camera calibration processes have complex operation requirements.

[0005] At present, the branch angle calculation method based on point cloud data mainly focuses on crops with simple plant type structure (such as symmetrical leaf growth) such as corn and sorghum, and the mainstream branch angle extraction method needs to extract the plant skeleton. The method based on skeleton extraction is very effective for plants with large volume and simple canopy structure, but it is difficult to extract the skeleton of small plants with interlaced leaves. The main reason is that the Laplace shrinkage method incorrectly compresses the point cloud of the base of the new leaves of small plants into the same category of skeleton as the main stem.

[0006] The existing branch angle extraction method based on point cloud data is not suitable for soybeans, mainly because: (1) the soybean branch is more rigid than the corn, sorghum and other crops, and there is no obvious curvature increase; (2) the soybean canopy is more complex than the corn, sorghum and other crops, the stems and leaves are thinner, and the three-dimensional distribution of the pods is more serious, and the scanning of the soybean plant will cause the loss of the skeleton structure due to the shielding of the leaves; (3) the average height of a single soybean plant is about 50cm, while the average height of a corn plant is about 200cm, which leads to smaller signal-to-noise ratio of the soybean point cloud data under the same scanning conditions, and the analysis is more difficult.

[0007] Therefore, there is an urgent need for a point cloud-based soybean branch angle extraction method to solve the problems in the prior art. SUMMARY

[0008] The purpose of the present application is to provide a point cloud-based soybean branch angle extraction method, to develop a three-dimensional point cloud-based soybean branch detection method, to calculate the branch angle characteristics representing the differences in soybean plant type, and to serve the identification and genetic analysis of excellent germplasm resources.

[0009] To achieve the above purpose, the present application provides a point cloud-based soybean branch angle extraction method, which specifically comprises the following steps:

[0010] Layered clustering detection is performed on the pretreated soybean single plant point cloud data to obtain the branch position;

[0011] Based on the branch position, a layer containing multiple clustering clusters is selected, the median point of the clustering cluster is calculated, and the bifurcation point is obtained;

[0012] Based on the bifurcation point, the branch position is optimized to obtain the optimized branch point;

[0013] Based on the bifurcation point and the optimized branch point, the branch angle is calculated to realize point cloud-based soybean branch angle extraction.

[0014] Optionally, the layered clustering detection of the pretreated soybean single plant point cloud data to obtain the branch position comprises:

[0015] Based on the pretreated soybean single plant point cloud data, layering is performed according to the preset layering height to obtain the point cloud data of each layer;

[0016] The point cloud data of each layer is subjected to DBSCAN clustering, and the number of clustering clusters is counted;

[0017] The number of clustering clusters of the current layer and the number of clustering clusters of the next layer are analyzed to obtain the clustering point of the branch layer;

[0018] According to the clustering point of the branch layer, the branch position is obtained.

[0019] Optionally, the clustering point of the branch layer is obtained by analyzing the number of clustering clusters of the current layer and the number of clustering clusters of the next layer, comprising:

[0020] If the number of clustering clusters of the current layer is greater than the number of clustering clusters of the next layer, the current layer contains a new branch, the clustering of the current layer is saved, and the branch positioning is performed to obtain the clustering point of the branch layer;

[0021] If the number of clustering clusters of the current layer is equal to the number of clustering clusters of the next layer, it is judged whether the current layer contains the new branch to obtain the clustering point of the branch layer;

[0022] If the number of clustering clusters of the current layer is less than the number of clustering clusters of the next layer, it is judged whether the current layer contains the new branch to obtain the clustering point of the branch layer.

[0023] Optionally, if the number of clustering clusters of the current layer is equal to the number of clustering clusters of the next layer, it is judged whether the current layer contains the new branch to obtain the clustering point of the branch layer, comprising:

[0024] The clustering center point of each layer is obtained;

[0025] The first nearest distance from the clustering center point of the next layer to the clustering center point of the current layer is calculated and compared with a threshold distance;

[0026] If the first nearest distance is less than or equal to the threshold distance, the current layer does not contain the new branch, the branch of the current layer is a continuation of the branch of the next layer, and the branch of the current layer is discarded;

[0027] If the first nearest distance is greater than the threshold distance, the current layer contains the new branch, and the branch positioning is performed to obtain the clustering point of the branch layer.

[0028] Optionally, if the number of clustering clusters of the current layer is less than the number of clustering clusters of the next layer, it is judged whether the current layer contains the new branch to obtain the clustering point of the branch layer, comprising:

[0029] The second nearest distance from the clustering of the next layer to the clustering of the current layer is calculated and compared with a threshold distance, and the number of second nearest distances less than the threshold distance is counted;

[0030] If the number of second nearest distances less than the threshold distance is greater than or equal to the number of clustering clusters of the current layer, the current layer does not contain the new branch, the branch of the current layer is a continuation of the branch of the next layer, and the branch of the current layer is discarded;

[0031] If the second nearest distance is less than a threshold distance and the number of the second nearest distance is less than the number of the cluster in the current layer, the current layer contains the new branch, and branch positioning is performed to obtain the cluster point of the branch layer.

[0032] Optionally, based on the branch position, a hierarchical layer containing a plurality of cluster groups is selected, a median point of the cluster groups is calculated, and the bifurcation point is obtained.

[0033] S1, based on the branch position, a hierarchical layer containing a plurality of cluster groups is selected, and then a plurality of cluster groups are selected;

[0034] S2, the median points of the plurality of cluster groups are calculated as seed points, a median point height value is obtained, and points with a height value greater than the median point height value are deleted;

[0035] S3, the seed points are subjected to a nearest neighbor search, a new cluster is obtained, and the median point of the new cluster is calculated.

[0036] S4, S2-S3 are repeated until the number of overlapping points exceeds a growth point number threshold, the growth is ended, a target cluster median point is calculated, and the target cluster median point is the bifurcation point.

[0037] Optionally, based on the bifurcation point, the branch position is optimized, and an optimized branch point is obtained.

[0038] Based on the bifurcation point, a branch point of a target slice thickness is selected.

[0039] The upper end of the branch point of the target slice thickness is taken as a lower limit, and points of a target height are selected upward to obtain a target branch.

[0040] The target branch is subjected to DBSCANM clustering, the median points of a plurality of clusters after division are calculated, and the optimized branch point is obtained.

[0041] Optionally, based on the bifurcation point and the optimized branch point, a branch angle is calculated.

[0042] Based on the bifurcation point and the optimized branch point, a space vector is constructed.

[0043] The space vector is used to calculate the branch angle through a space vector included angle.

[0044] The present application has the following beneficial effects:

[0045] Compared with the prior art, the application has advantages in measuring plants with small volume and complex crown layer structure, and the method is also suitable for measuring the branch angle of large crops, and creates a simple and efficient algorithm for automatically obtaining the branch angle parameters of plants; based on the detection of the bifurcation point and the optimization of the branch point, the branch angle can be calculated based on the spatial vector, which promotes the high-throughput extraction efficiency of plant phenotypes; the method provided by the application can obtain the branch angles at different heights of plants, which is used for evaluating crop plant type parameters, identifying excellent germplasm resources for density tolerance, and breeding utilization, and through providing a new technical means of nondestructive, high-throughput and high-precision, the breeding process is accelerated and the cultivation management measures are optimized. BRIEF DESCRIPTION OF DRAWINGS

[0046] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein for a purpose of explanations of the application and are not intended as an undue limitation of the application. In the drawings:

[0047] Figure 1 A flowchart of a soybean branch angle extraction method based on point cloud according to an embodiment of the application;

[0048] Figure 2 A flowchart of slice clustering and branch detection according to an embodiment of the application, wherein (a) is a slice clustering flowchart, (b) is a branch detection flowchart, (b.1) is a branch detection flowchart when CNU>CNL, (b.2) is a branch detection flowchart when CNU=CNL, and (b.3) is a branch detection flowchart when CNU<CNL;

[0049] Figure 3 A flowchart of bifurcation position and downward region growth according to an embodiment of the application;

[0050] Figure 4 A flowchart of branch point optimization and branch angle calculation according to an embodiment of the application;

[0051] Figure 5 A branch detection and bifurcation position result chart of 12 representative soybean varieties according to an embodiment of the application, (a) is a maximum and minimum point distance sample, (b) is a maximum and minimum plant height sample, (c) is a maximum and minimum crown height sample, (d) is a maximum and minimum crown width height ratio sample, (e) is a maximum and minimum angle number sample, and (f) is a maximum and minimum average branch angle sample;

[0052] Figure 6 A comparison chart of branch angles of different data sets according to an embodiment of the application, (a) is a simple sample group, (b) is a medium sample group, and (c) is a complex sample group;

[0053] Figure 7The visual result figures of different corn and tomato plants provided by the embodiments of the present application are shown in (a1)-(a6) and (b1)-(b6). (a1)-(a6) are the branch point and bifurcation point detection effects of the algorithm in 6 different corn plants; (b1)-(b6) are the branch point and bifurcation point detection effects of the algorithm in 6 different tomato plants.

[0054] In the figure, the triangle represents the bifurcation point, and the square represents the branch point.

[0055] Figure 8 The quantitative result figures of branch angle extraction of different plants provided by the embodiments of the present application are shown in (a) and (b). (a) is 6 corn plants with representative plant types, and (b) is 6 tomato plants with representative plant types. DETAILED DESCRIPTION

[0056] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0057] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0058] The development of three-dimensional reconstruction technology greatly promotes the extraction of three-dimensional phenotypic traits of crops. In the field of computer vision research, the devices and technologies currently used for plant three-dimensional reconstruction mainly include stereo vision, laser scanner and depth camera. Laser radar, as an active three-dimensional measurement technology, can obtain high-precision three-dimensional point cloud data of the target, bringing new opportunities for soybean branch angle extraction. Therefore, it is particularly important to develop a soybean branch angle extraction method based on three-dimensional point cloud data, which is not only conducive to the timely monitoring of crop growth (photosynthesis and stress physiology) and the precise adjustment of management measures, but also conducive to the genetic breeding of crop plant types and the realization of variety improvement.

[0059] Based on the laser radar point cloud data, the branch angle is calculated. First, the branch position in the plant is located, and the point cloud data is disordered. How to let the machine recognize and accurately locate the branch position in the soybean point cloud single plant is worth exploring. The present application is based on the growth process of the plant, that is, when the plant grows upward, a new branch is formed every time a new branch grows, which forms a branch angle. The branch angle detection is divided into three steps: (1) branch positioning; (2) bifurcation point detection; (3) branch point optimization.

[0060] As shown in Figure 1 The present embodiment provides a soybean branch angle extraction method based on point cloud, which specifically includes the following steps:

[0061] Step one, branch localization based on hierarchical slicing method;

[0062] As shown in Figure 2 (a), based on the pretreated single plant point cloud data of soybean, the data is divided into layers according to a certain layer height (H) from bottom to top. The DBSCAN clustering is performed on each layer of point cloud, and the number of clusters is counted. The number of clusters of each layer and the next layer is analyzed to determine whether each layer is a new branch. As shown in Figure 2 (b), there are mainly three cases for the number of clusters of the current layer and the next layer:

[0063] (1) The number of clusters in the current layer slice (CNU) is greater than the number of clusters in the next layer slice (CNL), that is, CNU> CNL;

[0064] As shown in Figure 2 (b.1), when CNU> CNL, the number of clusters of the current layer and the next layer is the simplest case, that is, the number of clusters increases from bottom to top, which means that a new branch is formed. Therefore, the clusters of CNU are retained for branch localization.

[0065] (2) The number of clusters in the current layer slice (CNU) is equal to the number of clusters in the next layer slice (CNL), that is, CNU= CNL;

[0066] As shown in Figure 2 (b.2), when CNU= CNL, the number of clusters of the current layer and the next layer includes two possible results: (I) the current layer contains a new branch and should be retained; (II) the current layer does not contain a new branch, but is a continuation of the upper layer branch and should be ignored.

[0067] In order to determine whether a layer contains a new branch, a distance-based rule is proposed to determine according to the nearest distance (i.e. D CNL = {d1, …, d CNL}) between each cluster of the next layer and the clusters of the current layer. It is worth noting that each cluster in each layer contains a number of points, and since the median point is robust in describing the center position of a cluster of points, each cluster of points is represented by the median point. If there is a cluster center point in the next layer, the distance between the center point and all cluster center points in the current layer is calculated. If the distance is greater than a given threshold, the center point cannot be found in the current layer, indicating that a new branch is generated between the two layers; otherwise, if each cluster in the next layer can find a neighboring cluster in the current layer within a given distance threshold, the current layer does not contain a new branch, but is a continuation. This judgment logic can be summarized as a distance-based rule, that is: if the maximum value (i.e. max(D CNL )) in the nearest distance is greater than the minimum value (i.e. min(D CNL)) big two times, the multiple is determined by trial and error, then this layer should be retained for positioning new branches; otherwise, this layer is just the continuation of the lower layer branch, should be discarded.

[0068] (3) the number of clusters in the current slice (CNU) is less than the number of clusters in the lower slice (CNL), i.e. CNU < CNL;

[0069] As shown in Figure 2 (b.3), when CNU < CNL, the number of cluster clusters of the current layer and the lower layer includes two possible results: (I) the current layer contains a new branch, which should be retained; (II) the current layer does not contain a new branch, but is the continuation of the previous layer branch, which should be ignored.

[0070] In order to determine whether a layer contains a new branch, the above-mentioned distance-based rule is still used. First, the nearest distance between each lower layer cluster and the current layer cluster is calculated (i.e. D CNL ). Then, count the number of elements in D CNL that meet the rule of less than twice Min(D CNL ), and name it N min ; if N min is less than CNU, then some clusters of the current layer cannot find the source branch in the lower layer, and the current layer contains a new branch; otherwise, if N min is equal to or greater than CNU, then all clusters of the current layer can find the source branch in the lower layer, which means that the current layer has no new branch. Similarly, the above logic can be summarized as a distance-based rule, that is, if there are CNU or more distances in D CNL that are less than twice Min(D CNL ), then the current layer is just the continuation of the lower layer branch, which should be discarded; otherwise, the current layer should be retained for positioning new branches.

[0071] In summary, by analyzing the hierarchical slices, clusters and distance relationships between clusters of upper and lower layers, all branches of a soybean individual can be detected from its point cloud. According to the cluster points of the detected branch layer, the position of the branch can be roughly determined.

[0072] Step two, bifurcation point detection based on downward region growing method;

[0073] As shown in Figure 3 , after determining the position of the branch, there will be more than two cluster clusters in some hierarchical layers. Considering that the cluster clusters of the same branch are usually the two closest clusters, by calculating the two clusters with the closest distance in each hierarchical layer, the two clusters belong to the main stem and the branch respectively.

[0074] The KD-Tree is established on the single-point cloud, two clustering clusters based on branches are calculated, the median points of the clusters are calculated, the points with a Z value greater than the two median points are removed (i.e. region growing downward), the median points of the clustering on the two sides of the branch are set as seed points, the nearest neighbor search is performed around the seed points respectively, N nearest points are extracted as new clusters, the median points of the two new clusters are calculated, the points with a Z value greater than the two median points are removed, the new median points are set as seed points to continue the nearest neighbor search around, N nearest points are extracted as new clusters, the above steps are repeated, the points searched on the two sides will overlap, until the number of the accumulated overlapping points on the two sides exceeds the threshold N of the growing points, the growth is ended, the last time of the growing overlapping points is extracted, the median point of the cluster composed of the overlapping points is calculated, and the median point is the bifurcation point to be found.

[0075] Step three, bifurcation point optimization based on upward segmentation method;

[0076] The bifurcation median point found based on the original hierarchical detection is close to the bifurcation point, which may cause the calculated angle to be generally large. Meanwhile, in order to keep consistent with the traditional manual measurement of the angle, the bifurcation position at a certain height above the bifurcation point is selected for measurement. As shown in the figure, Figure 4 The specific method is as follows: a slice thickness D is selected upward based on the bifurcation point (determined according to the experience of manual measurement), the upper end of the slice thickness D is taken as the lower limit, and a height of 1 cm is taken upward, the bifurcation in the height is extracted, and DBSCAN clustering is performed on the bifurcation, and the median points of the two clusters divided are calculated, and the two median points are the optimized bifurcation points.

[0077] Based on the bifurcation point and the two bifurcation points, a spatial vector is constructed The bifurcation angle is calculated through the spatial vector angle, as shown in formulas (1)-(3):

[0078]

[0079]

[0080]

[0081] Wherein, x1, y1, z1 represent the vector coordinates composed of the bifurcation point and one of the bifurcation points, x2, y2, z2 represent the vector coordinates composed of the bifurcation point and the other bifurcation point, |a| is the modulus of vector a, |b| is the modulus of vector b, and θ is the angle between the two vectors.

[0082] Accuracy verification:

[0083] The 152 soybean variety point clouds are divided into three different difficulty datasets, namely simple, medium and complex, according to the number of branches, the density and the missing degree of the point cloud, and the different datasets are tested by the algorithm. Part of the qualitative results of the experiment, such as Figure 5 shown, Figure 5 (a) The left and right subgraphs are the point clouds of plants with the maximum and minimum point spacing, which are 1.12 mm and 0.36 mm respectively; Figure 5 (b) The left and right subgraphs are the plants with the highest and lowest height, which are 1.08 m and 0.22 m respectively; Figure 5 (c) The left and right subgraphs are the plants with the widest and narrowest crown width, which are 0.48 m and 0.06 m respectively; Figure 5 (d) The left and right subgraphs are the plants with the largest and smallest crown height ratio, which are 0.6 and 0.1 respectively; Figure 5 (e) The number of angles, the left and right subgraphs are the plants with the most and least angles, which are 11 and 1 respectively; Figure 5 (f) The left and right subgraphs are the plants with the largest and smallest average branch angle, which are 93.71° and 26.42° respectively. The quantitative results of the experiment, such as Figure 6 shown, in the Soybean3D dataset, the branch angles of Figure 6 (a) the simple group, Figure 6 (b) the medium group and Figure 6 (c) the complex group are extracted by using SCAG and manual measurement, and the results are compared. The correlation coefficients of the algorithm in the simple, ordinary and complex groups are 0.88, 0.85 and 0.83. The qualitative results show that the algorithm has good applicability for angle calculation of soybean, but when the branch level or downward growth leads to too large angle, the algorithm cannot well identify the angle. The quantitative results show that the correlation coefficients (r) of the algorithm for simple, ordinary and complex calculation reach 0.88, 0.85 and 0.83.

[0084] In order to show the branch angle calculation of different crops by the embodiment, the method is used on 6 corns and 6 tomatoes. As Figure 7 shown, Figure 7 (a1)-(a6) are the detection effects of branch points and branch points of the algorithm in 6 different corn plants, Figure 7 (b1)-(b6) are the detection effects of branch points and branch points of the algorithm in 6 different tomato plants. The qualitative results show that the embodiment can be well used for detecting the branches of corn and tomato, and only when the branch curvature is too large or the branch leaf is too small, the missed detection will occur. As Figure 8 (a) shown, the quantitative results show that the correlation coefficient (r) of the embodiment for corn branch angle extraction reaches 0.95; as Figure 8(b) as shown, the quantitative results show that the correlation coefficient (r) of the tomato branch angle extraction of this embodiment reaches 0.94 quantification. The above results fully show that the SCAG method can be well adapted to other crops and accurately calculate the branch angle.

[0085] The above merely provides the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any changes or replacements within the technical scope disclosed by the present application, which can be easily thought by any person skilled in the art, should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for extracting the branch angle of soybeans based on point clouds, characterized in that, Specifically, the following steps are included: Hierarchical clustering detection was performed on the preprocessed soybean single-plant point cloud data to obtain the branching positions; Based on the branching position, a stratum containing multiple clusters is selected, the median of the clusters is calculated, and the bifurcation point is obtained. Based on the branching location, a stratum containing multiple clusters is selected, the median of the clusters is calculated, and the bifurcation point is obtained, including: S1. Based on the branch position, select a layer containing multiple clusters, and then select several clusters; S2. Calculate the median of several clusters as seed points, obtain the median height value, and delete points whose height value is greater than the median height value; S3. Perform a nearest neighbor search on the seed points to obtain new clusters, and calculate the median of the new clusters; S4. Repeat S2-S3 until the number of overlapping points exceeds the growth point threshold, then end the growth and calculate the site in the target cluster, where the site in the target cluster is the bifurcation point. The branching position is optimized based on the bifurcation point to obtain the optimized branching point; Based on the bifurcation point, the branch position is optimized to obtain the optimized branch point, including: Based on the bifurcation point, select the branch point for the target slice thickness; The upper end of the branch point of the target slice thickness is taken as the lower limit, and the point of the target height is selected upward to obtain the target branch; DBSCAN clustering is performed on the target branch to calculate the median of several clusters after division, and the optimized branch point is obtained. Based on the bifurcation point and the optimized branch point, the branch angle is calculated to realize the extraction of soybean branch angle based on point cloud.

2. The method for extracting soybean branch angles based on point clouds as described in claim 1, characterized in that, Hierarchical clustering detection is performed on the preprocessed soybean single-plant point cloud data to obtain the branching positions, including: Based on the preprocessed soybean single plant point cloud data, the data is divided into layers according to a preset layering height, and the point cloud data of each layer is obtained. DBSCAN clustering is performed on the point cloud data of each layer, and the number of clusters is counted. Analyze the number of clusters in the current layer and the number of clusters in the next layer to obtain the cluster points of the branch layer; The branching position is obtained based on the clustering points of the branching layer.

3. The method for extracting soybean branch angles based on point clouds as described in claim 2, characterized in that, Analyzing the number of clusters in the current layer and the number of clusters in the next layer, the clustering points of the branch layer are obtained, including: If the number of clusters in the current layer is greater than the number of clusters in the next layer, then the current layer contains a new branch. The current layer clusters are saved, and the branch is located to obtain the cluster points of the branch layer. If the number of clusters in the current layer is equal to the number of clusters in the next layer, then determine whether the current layer contains the new branch and obtain the cluster points of the branch layer. If the number of clusters in the current layer is less than the number of clusters in the next layer, then determine whether the current layer contains the new branch and obtain the cluster points of the branch layer.

4. The method for extracting soybean branch angles based on point clouds as described in claim 3, characterized in that, If the number of clusters in the current layer is equal to the number of clusters in the next layer, then determine whether the current layer contains the new branch, and obtain the cluster points of the branch layer including: Obtain the cluster center points of each layer; Calculate the first nearest distance from the cluster center point of the next layer to the cluster center point of the current layer, and compare it with a threshold distance; If the first nearest distance is less than or equal to the threshold distance, then the current layer does not contain the new branch, the branch of the current layer is a continuation of the branch of the next layer, and the branch of the current layer is discarded. If the first nearest distance is greater than the threshold distance, then the current layer contains the new branch, and branch positioning is performed to obtain the clustering points of the branch layer.

5. The method for extracting soybean branch angles based on point clouds as described in claim 3, characterized in that, If the number of clusters in the current layer is less than the number of clusters in the next layer, then determine whether the current layer contains the new branch, and obtain the cluster points of the branch layer including: Calculate the second nearest distance from the next cluster to the current cluster, compare it with the threshold distance, and count the number of clusters whose second nearest distance is less than the threshold distance; If the number of clusters in the current layer whose second nearest distance is less than the threshold distance is greater than or equal to the number of clusters in the current layer, then the current layer does not contain the new branch, the branch in the current layer is a continuation of the branch in the next layer, and the branch in the current layer is discarded. If the number of times the second nearest distance is less than the threshold distance is less than the number of clusters in the current layer, then the current layer contains the new branch, and branch positioning is performed to obtain the cluster points of the branch layer.

6. The method for extracting soybean branch angles based on point clouds as described in claim 1, characterized in that, Based on the bifurcation point and the optimized branch point, the branch angle is calculated as follows: Based on the bifurcation point and the optimized branch point, construct a spatial vector; Using the spatial vectors, the branch angle is calculated through the angle between the spatial vectors.

Images