A plant morphological phenotype information extraction method

Abstract

The application discloses a kind of plant morphology phenotype information extraction method, belong to computer vision technical field.The method includes: propose local neighborhood structure plant node search positioning algorithm, the geometric distance and direction vector angle between adjacent nodes are calculated, the local neighborhood topological structure between nodes is analyzed, the node class and its belonging branch are judged;Plant region growth connection algorithm is proposed, the mask contour of the plant part of the same adjacent node is reduced dimension, the point that the maximum distance from direction axis is found on the target contour along the line parallel to direction axis, the boundary point of target plant part is determined, adjacent boundary point is sequentially connected, a continuous curve segment is formed, that is, the path close to adjacent node in real plant image;All the same adjacent nodes are sequentially repeated this operation, connect each part of plant, and construct complete plant morphology, realize fast and effective extraction plant morphology phenotype information, with the characteristics such as robust output result, convergence fast and stable.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, and more particularly to a plant morphology phenotype information extraction method. BACKGROUND

[0002] Plant phenotypes play an important role in genetics, botany and agronomy. Traditional plant phenotype measurement and analysis methods, such as manual observation and recording, are time-consuming and labor-intensive and cannot meet the requirements of large-scale crop phenotype analysis. There are certain limitations in terms of cost, performance and spatial and temporal coverage. With the rapid development of imaging technology, computing power and algorithms, computer vision has gradually changed the plant phenotype analysis and has become the main tool for phenotype analysis.

[0003] Traditional machine vision technology for plant phenotype trait recognition and prediction is mainly based on texture and color features, which requires a large amount of professional knowledge reserve. With the rapid development of computer vision, various deep learning-based models have been widely applied to plant phenotype recognition due to their excellent feature extraction, which can effectively improve the recognition efficiency and accuracy.

[0004] Guo et al. in Transactions of the Chinese Society of Agricultural Engineering proposed a detection region search algorithm based on improved YOLOv5 for obtaining soybean plant phenotype parameters. This method uses the detected soybean branch region as the search region, combines with the stem node coordinate information to reconstruct the soybean plant skeleton, and thus obtains phenotype parameters such as plant height, stem node number and branch number. However, this method is mainly suitable for plant types appearing in experiments, limiting its application range, and the reconstructed soybean plant skeleton is not fine enough in details, which will cause large errors in measuring plant height and other phenotypes.

[0005] Fu et al. in Computers and Electronics in Agriculture proposed a method of using RGB-D depth camera images to perform three-dimensional reconstruction of legume plants. This method extracts skeleton points through point cloud hierarchical clustering, connects the main stem skeleton points, and selects growth paths according to plant morphological characteristics, finally realizes digital modeling of plants. However, the universality and robustness of this method still have room for improvement, especially when dealing with plants with dense branches and leaves and almost no space between branches and leaves.

[0006] Chen et al. in Frontiers in Plant Science proposed a method based on improved single-shot multibox detector and ant colony optimization algorithm for extracting the phenotype information of whole soybean plants. The method uses an improved SSD network structure for pod and stem detection and positioning, and further proposes a stem extraction method based on an ant colony optimization algorithm, thereby constructing a complete soybean plant stem. However, the ant colony optimization algorithm may cause ants to fall into infeasible points or detours in planning paths when dealing with large-scale images or complex plant structures, affecting the efficiency of the algorithm and the accuracy of stem extraction. In the data set labeling process, the pod tip labeling range is determined by the occlusion condition, and the stem is labeled in segments without occlusion. This labeling method may introduce some subjectivity, and different labelers may produce differences, thereby affecting the stability of model training and the final detection results.

[0007] From the above technology, it can be seen that although the existing plant morphology extraction methods have made significant progress in precision and efficiency, they still face some challenges, such as the adaptability to specific plant types, the fineness of skeleton reconstruction, and measurement errors. Therefore, how to further improve the universality of plant morphology extraction methods, reduce errors, and enhance the processing capacity of complex plant structures is particularly important. SUMMARY

[0008] 1. Technical problems to be solved

[0009] In view of the problem in the prior art that the current plant phenotype information extraction method has certain limitations in identifying branches, stem nodes, and plant heights, the present application provides a plant morphology phenotype information extraction method. On the one hand, all plant nodes are detected by a local neighborhood structure plant node search positioning algorithm, and the plant node category and its belonging branch are determined. On the other hand, a plant region growing connection algorithm is proposed based on the plant node classification result to obtain the path close to the adjacent nodes in the actual plant image, and then the connection of each branch and the main stem is realized to construct the plant morphology, thereby realizing the fast and effective extraction of plant morphology phenotype information, and having the characteristics of robust output results, fast convergence, and stability.

[0010] 2. Technical solutions

[0011] The purpose of the present application is achieved by the following technical solutions.

[0012] A plant morphology phenotype information extraction method, comprising the following steps:

[0013] Collect and input plant morphology image data set;

[0014] The plant node detection module is configured based on a target detection model, a block structure of a model feature extraction layer in a backbone network of the target detection model is replaced by an EFViT structure, a multi-scale linear attention mechanism is combined, and an MST-EFViT structure is formed; all plant nodes in a plant morphology image are detected by the configured plant node detection module;

[0015] The plant node classification module is configured, all plant nodes are traversed, and the class of the plant node and the branch to which the plant node belongs are determined by using a local neighborhood structure plant node search positioning algorithm;

[0016] The plant growth connection module is configured, the path of adjacent plant nodes close to a real plant morphology image is calculated by using a plant region growth connection algorithm according to the plant node classification result; the operation is repeated for all adjacent plant nodes of the same type, and all parts of the plant are connected to construct a plant morphology;

[0017] The plant morphology phenotype information is extracted and output according to the plant morphology.

[0018] As a further improvement of the application, the MST-EFViT structure is represented as:

[0019] F MST-EFViT (x)=A MST (F EFViT (x))

[0020] Wherein, x represents a plant morphology image, F EFViT (x) represents the result of feature extraction of the plant morphology image by the EFViT structure, A MST represents a multi-scale linear attention mechanism, and F MST-EFViT (x) represents a plant node obtained by the MST-EFViT structure by performing feature extraction on the plant morphology image.

[0021] As a further improvement of the application, the class of the plant node and the branch to which the plant node belongs are determined by using the local neighborhood structure plant node search positioning algorithm, and the steps include:

[0022] A rectangular coordinate system is established, and the coordinates of all plant nodes are stored in a node array;

[0023] The node array is traversed, the coordinates of each plant node and its three adjacent plant nodes are extracted, the included angle between the plant node and the two adjacent plant nodes in front of the plant node and the included angle between the plant node and the two adjacent plant nodes behind the plant node are calculated, and whether the two included angle difference values satisfy a preset condition is determined to determine whether the plant has a branch;

[0024] If the plant has branches, traverse the node array, calculate the Euclidean distance and the included angle between each plant node and the previous plant node, and determine whether the plant node is the plant main stem node according to the preset condition; obtain the plant main stem node array through screening, and divide the branch node array on the left and right sides of the plant main stem according to the coordinate information in the plant main stem node array through the region judgment function; for the non-plant main stem node, analyze the characteristics of the K nearest neighbor plant nodes to determine the branch to which it belongs; if the plant has no branches, the original node array is the plant main stem node array.

[0025] As a further improvement of the application, the step of determining whether the plant has branches comprises:

[0026] The position coordinates of the plant nodes are stored in the node array Node=[X,Y], wherein Node represents an n×2 node array containing all the center points of the prediction boxes, X represents the horizontal coordinates of the center points, Y represents the vertical coordinates of the center points, and n represents a natural number;

[0027] For the plant node Node[i], the included angle α between the plant node Node[i] and the plant node Node[i+1] and the plant node Node[i+2] is calculated i , and the included angle α between the plant node Node[i] and the plant node Node[i+1] and the plant node Node[i+3] is calculated i+1 , and the calculation formula is:

[0028] α i =arctan 2(y i+2 -y i ,x i+2 -+x i )-+arctan 2(y i+1 -y i ,x i+1 -x i )

[0029] α i+1 =arctan 2(y i+3 -y i+1 ,x i+3 -x i+1 )-arctan 2(y i+2 -y i+1 ,x i+2 -x i+1 )

[0030] wherein i represents a natural number, (x i ,y i ) represents the coordinates of the plant node Node[i], (x i+1 , y i+1) represents the coordinates of the plant node Node[i+1], (x i+2 , y i+2 ) represents the coordinates of the plant node Node[i+2], and (x i+3 , y i+3 ) represents the coordinates of the plant node Node[i+3];

[0031] An error angle preset value is set, if the absolute value difference between the included angle α i and the included angle α i +1 is less than the error angle preset value, it is judged that the plant does not have branches, and the original node array is the plant main stem node array; if the absolute value difference between the included angle α i and the included angle α i +1 is greater than or equal to the error angle preset value, it is judged that the plant has branches.

[0032] As a further improvement of the present application, if the plant has branches, a region judgment function is defined to delimit a rectangular region excluding the plant nodes not on the plant main stem, and the formula is represented as:

[0033] left = min(Node[0, 1], Node[-1, 1]) - pixel

[0034] right = max(Node[0, 1], Node[-1, 1]) + pixel

[0035] bottom = image height ; top = 0

[0036] Area = (left, right, bottom, top)

[0037] Wherein, Area represents a rectangular region delimited by the region judgment function, left, right, bottom and top represent the left, right, bottom and top coordinate values of the rectangular region respectively, pixel represents the pixel value required to increase the width of the rectangular region from left to right, and image height represents the height of the plant morphology image.

[0038] As a further improvement of the present application, the plant node position coordinates of the rectangular region are stored in the node array, the first plant node in the node array is selected as the initial plant main stem node, and the included angle between the first plant node and the second plant node after the first plant node in the plant node array and the previous plant main stem node is calculated, and the calculation formula is:

[0039] Angle1 = (Node1[j, :], Node1[j+1, :], Node1[j+2, :]), j = 0, 1,..., n-3

[0040] Angle2 = (Node1[j, :], Node1[j+1, :], Node1[j+3, :]), j = 0, 1,..., n-3

[0041] Angle3 = (Node1[j, :], Node1[j+1, :], Node1[j+4, :]), j = 0, 1,..., n-3

[0042] max a1 , max p1 = Max(Angle1, Angle2, Angle3)

[0043] max a2 , max p2 = SeMax(Angle1, Angle2, Angle3)

[0044] wherein n represents a coordinate array length of a node array, Angle1, Angle2 and Angle3 represent three-point angles obtained by the cosine law, j represents an index of a current plant node, j+1 represents an index of a next plant node, wherein j = 0, 1,..., n-3, Max represents finding the largest angle in three angle sums corresponding points, SeMax represents finding the second largest angle in three angle sums corresponding points, max a1 represents a maximum angle value in Angle1, Angle2 and Angle3 calculated by the Max function, max a2 represents a second largest angle value in Angle1, Angle2 and Angle3 calculated by the SeMax function, max p1 corresponds to a plant node position of max a1 , max p2 corresponds to a plant node position of max a2 ;

[0045] An error angle is set, if the difference between max a1 and max a2 is less than the error angle, then the Euclidean distance of max p1 to the node array Node1[j+1, :] and max p2 to the node array Node1[j+1, :] is calculated, the point with the smallest Euclidean distance is the plant main stem node, according to this step, the plant node array is traversed in turn until all plant nodes are evaluated, and the plant main stem node array is screened out;

[0046] After all the main stem nodes of the plant are determined, the area judgment function is executed to divide the branch node arrays on the left and right sides of the main stem of the plant, and each plant node in the areas is judged according to the above steps, and the plant branch nodes are stored in the plant branch node array until all the plant nodes are judged, and the categories of all the plant nodes and the branches to which the plant nodes belong are determined.

[0047] As a further improvement of the present application, the path between adjacent plant nodes in the real plant morphology image is calculated by the plant area growth connection algorithm, and the steps include:

[0048] The plant node information in the plant main stem node array and the plant branch node array is mapped back to the plant morphology image, each plant node is represented as a rectangular frame, the center point of the rectangular frame corresponds to the plant node coordinates, and the width and height of the rectangular frame are determined according to the size of the plant node;

[0049] The plant morphology image is subjected to mask processing, the plant main stem node array and the plant branch node array are traversed, for each group of adjacent plant nodes of the same type, the plant part boundary between the two adjacent plant nodes is determined according to the corresponding rectangular frame, the minimum boundary rectangle of the plant part boundary between the two adjacent plant nodes is calculated, and the path between the two adjacent plant nodes in the real plant morphology image is obtained;

[0050] After the path of each group of adjacent plant nodes of the same type is determined, each main stem node is connected in turn, the lowest point of each branch node is extracted, and the lowest points of the branch nodes are connected in turn to construct the plant morphology.

[0051] As a further improvement of the present application, each plant node is represented as a rectangular frame, which includes: the coordinates of each plant node are extracted from the plant main stem node array and the plant branch node array in turn; for each plant node, the coordinates (x, y) are set, wherein x and y represent the horizontal position and the vertical position of the plant node in the plant morphology image respectively, the width w and the height h of the plant node are set, the left upper corner coordinates (x lu , y lu ) and the right lower corner coordinates (x ru , y ru ) of the plant node in the real plant morphology image are obtained, and the position and size of the plant node rectangular frame are determined, and the calculation formula is:

[0052]

[0053]

[0054] xl u represents the horizontal coordinate of the left upper corner of the rectangular frame of the plant node in the real plant morphology image, y lu represents the vertical coordinate of the upper left corner of the rectangular frame of the plant node in the real plant morphology image, x ru represents the horizontal coordinate of the lower right corner of the rectangular frame of the plant node in the real plant morphology image, y ru represents the vertical coordinate of the lower right corner of the rectangular frame of the plant node in the real plant morphology image, i represents a natural number, and n represents the total number of plant nodes.

[0055] As a further improvement of the present application, the steps for obtaining the path close to the adjacent plant nodes in the real plant morphology image include:

[0056] The plant mask morphology image is obtained by processing the plant morphology image, and the coordinates (x lu , y lu ) of the upper left corner and the coordinates (x ru , y ru ) of the lower right corner of the rectangular frame of the adjacent plant nodes in the real plant morphology image are extracted in sequence, the maximum and minimum coordinates (x lmin , y lmin ) and (x rmax , y rmax ) between the adjacent two plant nodes are calculated, and the plant part boundary between the adjacent two plant nodes is determined.

[0057] The plant part boundary mask contour is processed by dimension reduction to obtain the center point, the main direction, and the two orthogonal direction axes of the target contour. In the orthogonal direction of the direction axis, the cutting point between the direction axis and the target contour after moving up and down is the boundary point of the minimum boundary rectangle of the target contour. The contour tangent line is the direction axis after moving, that is, y=k1x+b * , y=k2x+b * , and the specific formula is:

[0058]

[0059] contour represents the boundary point set, k1 and k2 represent the slopes of the two direction axes, b * represents the intercept of the straight line, and the maximum and minimum values of b * correspond to the points farthest from the center axis along the two orthogonal directions, respectively.

[0060] The point farthest from the direction axis is found along the contour line, the arithmetic mean of the x coordinates of the maximum off-axis distance points is taken to obtain the x coordinate C x of the center point, and the arithmetic mean of the y coordinates of the maximum off-axis distance points is taken to obtain the y coordinate C y of the center point, thereby obtaining the center point (C x , C y), the maximum off-axis distance point corresponding to the k value is calculated, the width and height of the target minimum boundary rectangle are determined, and the path of the adjacent plant nodes in the real plant morphology image is obtained.

[0061] As a further improvement of the present application, the plant morphology is constructed by the steps comprising:

[0062] The lowest point of each plant branch node is determined, and the Euclidean distance between the node on the plant main stem and the lowest point of the plant branch node is calculated.

[0063] Two direction vectors are determined, including the branch direction vector of the plant and the main stem direction vector of the plant, and the angle of the direction vector between the branch lowest point of the plant and the main stem node is calculated, and the calculation formula is:

[0064]

[0065] wherein, represents the branch direction vector of the plant, represents the main stem direction vector of the plant, and θ represents the angle of the direction vector between the branch lowest point of the plant and the main stem node.

[0066] From the calculated Euclidean distance, the plant node with the shortest distance is selected as the candidate plant node, and the angle difference between the main stem direction vector and the branch direction vector of the candidate plant node is checked to see if it is within θ, if the condition is met, the plant branch node is connected with the plant main stem node, and each plant branch node is connected in turn to construct the plant morphology.

[0067] 3. Advantages

[0068] Compared with the prior art, the present application has the following advantages:

[0069] (1) The plant morphology phenotype information extraction method of the present application adds an adaptive training sample selection strategy to the plant node detection module. The adaptive training sample selection strategy reduces the influence of invalid or redundant samples on the plant node detection module by screening effective samples, avoids the over-reliance of the plant node detection module on certain samples, balances the distribution of data, and reduces the risk of overfitting of the plant node detection module on training data. In addition, a dynamic class distribution optimization algorithm is integrated into the loss function of the plant node detection module. By setting a weight factor to adjust the weight of each plant node class, the number difference of samples of each class in the training data is balanced, and the deep features of the plant nodes are processed more accurately and balanced without increasing the calculation cost.

[0070] (2) The plant morphological phenotype information extraction method of the present application replaces the bounding box regression loss of the plant node detection module with SDTIoU Loss. SDTIoU Loss considers all relevant factors that affect the similarity of two bounding boxes, such as overlapping or non-overlapping areas, center point distance, and height-width deviation, while simplifying the calculation process. Using SDTIoU Loss as a measure of the similarity of two bounding boxes can effectively improve the accuracy and efficiency of bounding box regression.

[0071] (3) The plant morphological phenotype information extraction method of the present application proposes a local neighborhood structure plant node search positioning algorithm. The initial stem node is selected as the starting point of the trunk from the preprocessed plant morphological image, and the subsequent stem nodes are evaluated step by step. By calculating the angle formed by three consecutive nodes and combining the error angle judgment, the nodes on the trunk are identified. Further, using the Euclidean distance and angle information, the nodes of each branch are evaluated and determined, thereby significantly improving the efficiency and accuracy of plant stem-related phenotype feature extraction.

[0072] (4) The plant morphological phenotype information extraction method of the present application proposes a plant region growing connection algorithm. For each group of adjacent nodes of the same type, the plant part boundary between the two adjacent nodes is determined according to the corresponding rectangular frame. The minimum boundary rectangle of the plant part between the two adjacent nodes is calculated using the contour dimension reduction algorithm, thereby minimizing the distance error between the adjacent nodes in the real plant morphological image, and obtaining a path close to the adjacent nodes in the real plant image. For branch nodes, the Euclidean distance of each point in the branch node array to all points in the main stem skeleton is calculated, and the point with the shortest distance is selected as the connection candidate point. Then, the branch direction vector and the main stem direction vector are calculated. By calculating the direction vector angle between the branch lowest point and each node on the main stem, it is determined whether it meets the specified conditions, thereby determining the connection point of each branch, and further completing the connection of the branch to the main stem. Finally, a complete plant morphology is constructed, and the plant morphological phenotype information is extracted quickly and effectively. This has important application value in the fields of plant phenotype research, plant growth monitoring, and precise management of crops, and has strong practicality and wide applicability. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 The plant morphological phenotype information extraction method flowchart of the present application embodiment;

[0074] Figure 2 The MST-EFViT structure schematic diagram of the present application embodiment;

[0075] Figure 3 The dynamic class distribution optimization algorithm schematic diagram of the present application embodiment;

[0076] Figure 4This is a schematic diagram of the local neighborhood structure plant node search and localization algorithm according to an embodiment of the present invention;

[0077] Figure 5 This is a schematic diagram of the plant region growth connectivity algorithm according to an embodiment of the present invention. Detailed Implementation

[0078] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0079] Example

[0080] like Figure 1 The image shown illustrates a method for extracting plant morphological phenotypic information provided in this embodiment. The method includes: collecting and inputting a plant morphological image dataset; constructing a plant node detection module based on a target detection model, replacing the block structure of the model feature extraction layer in the backbone network of the target detection model with an EFViT structure, and then combining it with a multi-scale linear attention mechanism to form an MST-EFViT structure; detecting all plant nodes in the plant morphological image through the plant node detection module; constructing a plant node classification module, traversing all plant nodes, and determining the plant node category and its branch through a local neighborhood structure plant node search and localization algorithm; constructing a plant growth connectivity module, calculating the path between adjacent plant nodes in the real plant morphological image based on the plant node classification results using a plant region growth connectivity algorithm; repeating this operation for all adjacent plant nodes of the same type to connect different parts of the plant and construct the plant morphology; and extracting and outputting plant morphological phenotypic information based on the plant morphology.

[0081] In this specific embodiment, plant morphology images are collected, and a plant morphology image dataset is constructed and input.

[0082] Furthermore, a plant node detection module is constructed based on the object detection model. The block structure of the feature extraction layer in the backbone network of the object detection model is replaced with an EFViT structure, and then combined with a multi-scale linear attention mechanism to form an MST-EFViT structure. For example... Figure 2 The diagram shown is a schematic of the MST-EFViT structure. In this embodiment, the MST-EFViT structure is represented as follows:

[0083] F MST-EFViT (x)=A MST (F EFViT (x))

[0084] Where x represents the plant morphology image, F EFViT (x) represents the result of feature extraction from the plant morphology image x using the EFViT structure. It should be noted that F EFViT The (x) function contains all feature extraction and transformation steps in the EFViT structure, AMST represents a multi-scale linear attention mechanism, F MST-EFViT (x) represents the plant node obtained by the MST-EFViT structure performing feature extraction on the plant morphological image x.

[0085] In the target detection model training process of the plant node detection module, an adaptive training sample selection strategy is introduced to filter effective samples from the original data to reduce the influence of invalid or redundant samples on the target detection model and reduce the risk of overfitting of the target detection model on the training data.

[0086] In this embodiment, the adaptive training sample selection strategy is represented as:

[0087]

[0088]

[0089] σIoU threshold = IoU avg + x x σIoU

[0090] wherein i represents a natural number, N represents the total number of anchor points, IoU avg represents the average value of the intersection over union of the N anchor points and the real bounding box, IoU(i) represents the intersection over union of the ith anchor point and the real bounding box, σ represents the standard deviation of the intersection over union value, x represents an experience parameter that determines which anchor points will be selected as positive samples, and IoU threshold represents a threshold value for determining whether a candidate anchor point can be selected as a positive sample. When the IoU of an anchor point and a real bounding box is greater than IoU threshold , the anchor point is selected as a positive sample, otherwise as a negative sample.

[0091] As shown in Figure 3 , in this embodiment, a dynamic class distribution optimization algorithm is integrated into the loss function of the target detection model, a weight factor is set to adjust the class weight, and the number difference of samples of each class in the training data is balanced. Without increasing the calculation cost, the deep features of the plant nodes are processed more accurately and balanced. In this embodiment, the weight factor is represented as:

[0092]

[0093] wherein W c represents the weight factor, pc represents the proportion of the cth class, i.e. the proportion of the sample number to the total sample, and β represents a hyperparameter in the range of [0, 1], which is used to adjust the strength of the weight. In this embodiment, when pc is very small, i.e. the class is rare, the weight factor W c will increase, thereby increasing the loss value of the class; on the contrary, when the class is common, the weight factor W cwill be reduced.

[0094] It is worth noting that in this embodiment, SDTIoU Loss is used as the bounding box regression loss of the plant node detection module. Thus, the bounding box regression loss of the plant node detection module is replaced by SDTIoU Loss, which takes into account all relevant factors that affect the similarity of two bounding boxes, such as overlapping or non-overlapping areas, center point distance, and width-height deviation, while simplifying the calculation process. Using SDTIoU Loss as a measure of the similarity of two bounding boxes can effectively improve the accuracy and efficiency of bounding box regression. In this embodiment, the bounding box regression loss is represented as:

[0095]

[0096]

[0097]

[0098] where x c represents the horizontal coordinate of the center point of the predicted box, y c represents the vertical coordinate of the center point of the predicted box, represents the horizontal coordinate of the center point of the target box, represents the vertical coordinate of the center point of the target box, t represents the width and height of the target box, which is used to calculate the shape loss, w represents the width of the predicted box, h represents the height of the predicted box, γ represents a dynamic parameter used to adjust the distance loss, ρt represents the predicted value in the regression process, Λ represents the angle loss used to calculate the minimum angle between the line connecting the center points of the predicted box and the real box and the coordinate axis, ρ x represents the relative deviation of the center point of the predicted box from the center point of the real box in the x direction, ρ y represents the relative deviation of the center point of the predicted box from the center point of the real box in the y direction, w c represents the width of the minimum bounding box covering the target box and the predicted box, h c represents the height of the minimum bounding box covering the target box and the predicted box, SDTIoU represents the bounding box regression loss, IoU represents the intersection over union, Δ and Ω both represent additional terms used to consider the angular difference of the center points and the width-height ratio of the bounding boxes.

[0099] In this embodiment, after the plant node detection module is constructed based on the target detection model, the plant morphology image is input into the plant node detection module, and all plant nodes in the plant morphology image are detected by the plant node detection module.

[0100] As Figure 4As shown, a plant node classification module is constructed. This module traverses all plant nodes and uses a local neighborhood structure plant node search and localization algorithm to determine the plant node category and its branch. Specific steps include:

[0101] S11. Establish a rectangular coordinate system and store the coordinates of all plant node positions in the node array Node = [X,Y], where Node represents an n×2 node array containing the center points of all prediction boxes, X represents the x-coordinate of the center point, and Y represents the y-coordinate of the center point.

[0102] S12. Traverse the node array, extract the rectangular coordinates of each plant node and its three adjacent plant nodes, calculate the angle between the plant node and the two preceding adjacent plant nodes and the angle between the plant node and the two following adjacent plant nodes, calculate whether the difference between the two angles meets the preset conditions, and determine whether the plant has branches.

[0103] Specifically, extract the coordinates of each group of three consecutive plant nodes in the node array. For each plant node Node[i], calculate the angle α formed by it and its two adjacent plant nodes Node[i+1] and Node[i+2]. i And the angle α formed by Node[i] with Node[i+1] and Node[i+3]. i+1 The calculation formula is as follows:

[0104] α i =arctan 2(y i+2 -y i x i+2 -x i )-arctan 2(y i+1 -y i x i+1 -x i )

[0105] α i+1 =arctan 2(y i+3 -y i+1 x i+3 -x i+1 )-arctan 2(y i+2 -y i+1 x i+2 -x i+1 )

[0106] Among them, (x i ,y i (x) represents the coordinates of the current plant node Node[i], (x) i+1 ,y i+1 (x) represents the coordinates of the plant node Node[i+1], (x) i+2 yi+2 ) represents the coordinates of the plant node Node[i+2], (x i+3 ,y i+3 ) represents the coordinates of the plant node Node[i+3].

[0107] S13. If the plant has branches, traverse the node array, calculate the Euclidean distance and the included angle between each plant node and the previous plant main stem node, and determine whether the plant node is the plant main stem node according to the preset condition.

[0108] Specifically, the error angle preset value Angle k is set, and if the absolute value difference between the included angle a i and the included angle a i +1 is less than the error angle preset value Angle k , all elements in the node array are traversed in turn. If all elements meet the preset condition, it is determined that the plant has no branches, and the node array Node is the main stem node array. If there is a case where the absolute value difference between the included angle a i and the included angle a i +1 is greater than or equal to the error angle preset value Angle k , it is determined that the plant has branches.

[0109] S14. The plant main stem node array is screened, and according to the determined coordinate information in the plant main stem node array, the branch node array on the left and right sides of the plant main stem is divided by the area judgment function; for non-plant main stem nodes, analyze the characteristics of the K nearest neighbor plant nodes to determine the branch to which it belongs.

[0110] Specifically, if the plant has branches, define the area judgment function to exclude the plant nodes that are not on the main stem in a rectangular area, and its formula is represented as:

[0111] left = min(Node[0, 1], Node[-1, 1]) - pixel

[0112] right = max(Node[0, 1], Node[-1, 1]) + pixel

[0113] bottom = image height ; top = 0

[0114] Area = (left, right, bottom, top)

[0115] Wherein, Area represents the rectangular area determined by the area judgment function, left, right, bottom, and top represent the left, right, bottom, and top coordinate values of the rectangular area, respectively, and pixel represents the pixel value required to increase the width of the rectangular area from left to right.

[0116] S15. The node coordinates of the rectangular region are stored in the node array Node1, the first node in the node array is selected as the initial main stem node, and the initial operation is started. The angle Angle between the first node and the second node after the first node in the node array and the previous main stem node is calculated, and the calculation formula is:

[0117] Angle1 = (Node1[j, :], Node1[j+1, :], Node1[j+2, :]), j = 0, 1,..., n-3

[0118] Angle2 = (Node1[j, :], Node1[j+1, :], Node1[j+3, :]), j = 0, 1,..., n-3

[0119] Angle3 = (Node1[j, :], Node1[j+1, :], Node1[j+4, :]), j = 0, 1,..., n-3

[0120] max a1 , max p1 = Max(Angle1, Angle2, Angle3)

[0121] max a2 , max p2 = SeMax(Angle1, Angle2, Angle3)

[0122] wherein n represents the length of the coordinate array of Node1, Angle1, Angle2, and Angle3 represent the angles between three points obtained by the cosine law, j represents the index of the current node, j+1 represents the index of the next node, wherein j = 0, 1,..., n-3 (n represents the length of the coordinate array of Node1), Max represents the largest angle in the three angles and corresponding points, SeMax represents the second largest angle in the three angles and corresponding points, max al represents the maximum angle value in Angle1, Angle2, and Angle3 calculated by the Max function, max a2 represents the second largest angle value in Angle1, Angle2, and Angle3 calculated by the SeMax function, max p1 corresponds to the node position of max a1 , max p2 corresponds to the node position of max a2 .

[0123] S16. If max a1 and max a2If the difference is less than the error angle a, then calculate max p1 to Node1[j+1, :] and max p2 to Node1[j+1, :] is the main stem node, and the process is repeated for each node in the array Node1 until all nodes are evaluated, and the main stem node array MNode is filtered out.

[0124] S17. After determining all main stem nodes, continue to execute the region judgment function to divide the branch node arrays on the left and right sides of the main stem, and judge the nodes in each region according to the above steps, and store the branch nodes in the corresponding branch array BNode j , where j represents the corresponding plant branch number, until all nodes are judged and all plant node categories and their corresponding branches are determined.

[0125] Thus, in this embodiment, by using the local neighborhood structure plant node search and positioning algorithm, the initial stem node is selected as the starting point of the main stem from the preprocessed plant morphology image, and the subsequent stem nodes are gradually evaluated, the nodes on the main stem are identified by calculating the angle formed by three consecutive nodes and combining the error angle judgment, and further, the nodes of each branch are evaluated and determined using the Euclidean distance and angle information, thereby significantly improving the efficiency and accuracy of plant stem-related phenotype feature extraction.

[0126] As shown in Figure 5 , a plant growth connection module is constructed, and according to the plant node classification result, the path of adjacent plant nodes close to the real plant morphology image is calculated by the plant region growth connection algorithm, and this operation is repeated for all similar adjacent plant nodes to connect each part of the plant to obtain the plant morphology, and the specific steps include:

[0127] S21. Map each plant node information in the plant main stem node array and the plant branch array back to the plant morphology image, and each plant node is represented as a rectangular box, the center point of the rectangular box corresponds to the coordinates of the plant node, and the width and height of the rectangular box are determined according to the actual size of the plant node.

[0128] Specifically, the coordinates of each plant node are extracted from the plant main stem node array and the plant branch node array. For each plant node, set the coordinates (x, y), where x and y represent the horizontal and vertical positions of the plant node in the plant morphology image, respectively. Assuming that the width of each plant node is w and the height is h, the left upper corner coordinates (x lu , y lu ) and the right lower corner coordinates (x ru , y ru), and then the position and size of the rectangular frame of each plant node are determined, and the calculation formula is:

[0129]

[0130]

[0131] wherein x lu represents the horizontal coordinate of the upper left corner of the rectangular frame of the plant node in the real plant morphology image, y lu represents the vertical coordinate of the upper left corner of the rectangular frame of the plant node in the real plant morphology image, x ru represents the horizontal coordinate of the lower right corner of the rectangular frame of the plant node in the real plant morphology image, and y ru represents the vertical coordinate of the lower right corner of the rectangular frame of the plant node in the real plant morphology image, i represents a natural number, and n represents the total number of plant nodes.

[0132] S22. Mask processing is performed on the plant morphology image, and the plant main stem node array and the plant branch array are traversed. For each group of adjacent plant nodes of the same type, the plant part boundary between the two adjacent plant nodes is determined according to the corresponding rectangular frame, and the contour dimension reduction algorithm is used to calculate the minimum boundary rectangle of the plant part between the two adjacent plant nodes, to obtain the path close to the adjacent plant nodes in the real plant morphology image. The specific steps include:

[0133] S221. The plant morphology image is converted from the RGB color space to the HSV color space, the corresponding HSV range is defined, the plant mask morphology image is obtained, and then the rectangular frame left upper corner coordinates (x lu , y lu ) and the right lower corner coordinates (x ru , y ru ) of each adjacent plant node of the same type in the real plant morphology image are extracted in turn, the maximum coordinates (x lmin , y lmin ) and (x rmax , y rmax ) between the two adjacent plant nodes are calculated, and the plant part boundary between the two adjacent plant nodes is determined.

[0134] S222. The existing principal component analysis technology is used to perform dimension reduction processing on the plant part boundary mask contour, to obtain the center point and the main direction and the two orthogonal direction axes of the target contour. It should be noted that the direction axis is a linear function, and moves up and down in the orthogonal direction of the direction axis. After moving, the tangent point between the direction axis and the contour is the point farthest from the center axis, which is also the boundary point of the minimum boundary rectangle of the contour. At this time, the tangent line of the contour is the direction axis after moving, i.e. y=k1x+b * , y=k2x+b * , and the specific formula is:

[0135]

[0136] where contour represents the set of contour points, k1 and k2 represent the slopes of the two direction axes respectively, b * represents the intercept of the straight line, b * represents the maximum and minimum values of the contour, respectively corresponding to the points farthest from the center axis along the two orthogonal directions.

[0137] S223. Find the points farthest from the direction axis along the contour points, a total of four points. According to the four points, take the arithmetic mean of the x coordinates of the four points farthest from the axis to obtain the x coordinate C x of the center point, take the arithmetic mean of the y coordinates of the four points to obtain the y coordinate C y of the center point, and obtain the center point (C x , C y ). Then calculate the maximum distance point corresponding to each k value, and further determine the width and height of the target minimum bounding rectangle, and finally obtain the path of adjacent plant nodes in the real plant morphology image.

[0138] S23. After determining the path of each group of similar plant nodes of the same type, sequentially connect the plant main stem nodes, extract the lowest points of each branch, find the plant stem nodes in the plant main stem array that are similar and have appropriate angles, and sequentially connect the plant nodes of each branch, so as to construct a complete plant morphology. The specific steps include:

[0139] S231. Determine the lowest point of each branch, i.e., the first element of each branch array. For each node on the main stem, calculate the Euclidean distance between the node and the lowest point of the branch. It should be noted that in plant morphology image processing, the node coordinates are usually stored in a bottom-up scanning manner, and thus the first element is the lowest point of each branch.

[0140] S232. Determine two direction vectors, a branch direction vector, i.e., a vector from the lowest point of the branch to other nodes of the branch, and a main stem direction vector, i.e., a vector from the lowest point of the branch to a node on the main stem. Calculate the angle between the direction vectors between the lowest point of the branch and each node on the main stem, and the calculation formula is:

[0141]

[0142] where, represents the branch direction vector, represents the main stem direction vector, and θ represents the angle between the branch direction vector and the main stem direction vector.

[0143] S233. From the calculated distances, the node with the shortest distance is selected as the candidate node, and the angle difference between the direction vector of the candidate node and the branch direction vector is checked to see if it is within an angle L = 45°, if both conditions are met, it is the main stem node to which the corresponding branch node should be connected. The connection is performed for each branch node array in turn, and the complete plant skeleton morphology is constructed.

[0144] Thus, in this embodiment, by using the plant region growing connection algorithm, the plant part boundary between the two adjacent nodes is determined according to the corresponding rectangular frame for each group of the same type of adjacent nodes, and the contour dimension reduction algorithm is used to calculate the minimum boundary rectangle of the plant part between the two adjacent nodes, thereby minimizing the distance error between the adjacent nodes in the real plant morphology image, and obtaining a path close to the adjacent nodes in the real plant image. For branch nodes, the Euclidean distance of each point in the branch node array to all points in the main stem skeleton is calculated, and the point with the shortest distance is selected as the connection candidate point. Then the branch direction vector and the main stem direction vector are calculated. By calculating the angle between the branch lowest point and each node on the main stem, it is determined whether it meets the predetermined condition, thereby determining the connection point of each branch, and then completing the connection of the branch to the main stem. Finally, the complete plant morphology is constructed, and the plant morphology phenotype information is extracted and output. This embodiment provides a plant morphology phenotype information extraction method, which can realize accurate measurement of plant morphology phenotype information, and has important application value in the fields of plant phenotype research, plant growth monitoring, etc.

[0145] Finally, the plant morphology phenotype information is extracted and output according to the plant morphology. In this embodiment, the plant phenotype information includes the number of internodes, the length of internodes, the number of branches, the branch angle, the plant height, and the main stem length, etc. Thus, the plant morphology phenotype information extraction method provided in this embodiment can realize accurate measurement of plant morphology phenotype information, and has important application value in the fields of plant phenotype research, plant growth monitoring, etc.

[0146] The above description of the application and its embodiments is illustrative and not restrictive, and the application can be practiced in other specific forms without departing from the spirit or essential character thereof. The drawings described herein are only one of the many embodiments of the application and are not limiting, and any reference signs in the claims should not be construed as limiting the claims to the figure in which the reference signs are used. Therefore, if a person skilled in the art is inspired by the disclosure, and without departing from the spirit of the application, designs a similar structure and embodiment to the technical solution without creativity, it should be within the protection scope of the application. In addition, the word "comprising" does not exclude other elements or steps, and the word "one" before an element does not exclude the inclusion of "multiple" elements. The multiple elements stated in the product claim can also be implemented by one element through software or hardware. The words "first", "second" and the like are used to indicate names, and not to indicate any particular order.

Claims

1. A method for extracting plant morphological phenotypic information, comprising the following steps: Collect and input a dataset of plant morphology images; Based on the target detection model, a plant node detection module is configured. The block structure of the model feature extraction layer in the backbone network of the target detection model is replaced with an EFViT structure, and then combined with a multi-scale linear attention mechanism to form an MST-EFViT structure. The configured plant node detection module detects all plant nodes in the plant morphology image. Configure a plant node classification module, traverse all plant nodes, and determine the category of plant nodes and their branches through a local neighborhood structure plant node search and positioning algorithm. Configure the plant growth connectivity module. Based on the plant node classification results, calculate the path of adjacent plant nodes in the image that closely resembles the real plant morphology using the plant region growth connectivity algorithm. Repeat this operation for all adjacent plant nodes of the same type to connect the various parts of the plant and construct the plant morphology. Extract and output plant morphological phenotypic information based on plant morphology; The steps for determining the category and branch of a plant node using the local neighborhood structure plant node search and localization algorithm include: Establish a rectangular coordinate system and store the coordinates of all plant node positions in a node array; Traverse the node array, extract the coordinates of each plant node and its three adjacent plant nodes, calculate the angle between the plant node and the two preceding adjacent plant nodes and the angle between the plant node and the two following adjacent plant nodes, and determine whether the plant has branches by calculating whether the difference between the two angles meets the preset conditions. If the plant has branches, the node array is traversed, and the Euclidean distance and angle between each plant node and the previous plant node are calculated. Based on preset conditions, it is determined whether the plant node is a main stem node. The main stem node array is obtained by filtering. Based on the coordinate information in the main stem node array, the branch node arrays on the left and right sides of the main stem are divided by a region judgment function. For non-main stem nodes, the characteristics of their K nearest neighbor plant nodes are analyzed to determine their branch. If the plant has no branches, the original node array is the main stem node array. The steps of calculating the paths of adjacent plant nodes in the image that closely resemble the actual plant morphology using the plant region growth connectivity algorithm include: The plant node information in the main stem node array and the branch node array is mapped back to the plant morphology image. Each plant node is represented as a rectangle, the center point of the rectangle corresponds to the coordinates of the plant node, and the width and height of the rectangle are determined according to the size of the plant node. The plant morphology image is masked, and the main stem node array and the branch node array are traversed. For each group of adjacent plant nodes of the same type, the plant part boundary between two adjacent plant nodes is determined according to its corresponding rectangle. The minimum boundary rectangle between two adjacent plant nodes is calculated to obtain the path of adjacent plant nodes in the real plant morphology image. After determining the paths of adjacent plant nodes of the same type in each group, connect the main stem nodes of each plant in sequence, extract the lowest point of each branch node, and connect the lowest points of each branch node in sequence to construct the plant morphology.

2. The method for extracting plant morphological phenotypic information according to claim 1, characterized in that, The MST-EFViT structure is represented as follows: F MST-EFViT (x)=A MST (F EFViT (x)) Where x represents the plant morphology image, F EFViT (x) represents the feature extraction result of the EFViT structure on the plant morphology image, A MST F represents a multi-scale linear attention mechanism. MST-EFViT (x) represents the plant node obtained by feature extraction of the plant morphology image using the MST-EFViT structure.

3. The method for extracting plant morphological phenotypic information according to claim 1, characterized in that, The steps for determining whether a plant has branches include: The coordinates of the plant node positions are stored in the node array Node = [X,Y], where Node represents an n×2 node array containing all the center points of the prediction boxes, X represents the x-coordinate of the center point, Y represents the y-coordinate of the center point, and n represents a natural number. For plant node Node[i], calculate the angle α formed by plant node Node[i], plant nodes Node[i+1], and plant nodes Node[i+2]. i And the angle α formed by plant node Node[i], plant node Node[i+1], and plant node Node[i+3]. i+1 The calculation formula is: α i =arctan2(y i+2 -and i ,x i+2 -x i )-arctan2(y i+1 -and i ,x i+1 -x i ) α i+1 =arctan2(y i+3 -and i+1， x i+3 -x i+1 )-arctan2(y i+2 -and i+1 ,x i+2 -x i+1 ) Where i represents a natural number, (x i ,y i ) represents the coordinates of the plant node Node[i], (x i+1 ,y i+1 (x) represents the coordinates of the plant node Node[i+1], (x) i+2 ,y i+2 (x) represents the coordinates of the plant node Node[i+2], (x) i+3 ,y i+3 () represents the coordinates of the plant node Node[i+3]; Set the error angle preset value, if the included angle α i and included angle α i If the absolute value difference of +1 is less than the preset error angle value, it is determined that the plant has no branches, and the original node array is the main stem node array of the plant; if the included angle α i and included angle α i If the absolute difference of +1 is greater than or equal to the preset error angle value, then the plant is determined to have branches.

4. The method for extracting plant morphological phenotypic information according to claim 3, characterized in that, If the plant has branches, the defined region judgment function delineates a rectangular region to exclude plant nodes that are not on the main stem of the plant. The formula is expressed as: left=min(Node[0,1],Node[-1,1])-pixel right=max(Node[0,1],Node[-1,1])+pixel bottom=image height ;top=0 Area=(left,right.bottom,top) Where Area represents a rectangular region defined by the region determination function, left, right, bottom, and top represent the left, right, bottom, and top coordinates of the rectangular region, respectively, and pixel represents the number of pixels required to increase the width of the rectangular region from left to right. height This indicates the height of the plant morphology image.

5. The method for extracting plant morphological phenotypic information according to claim 4, characterized in that, The coordinates of the plant nodes within the rectangular area are stored in a node array. The first plant node in the array is selected as the initial main stem node. The angle between the two consecutive plant nodes following the first plant node and the preceding main stem node is calculated using the following formula: Angle1=(Node1[j,:], Node1[j+1,:], Node1[j+2,:]), j=0, 1,..., n-3 Angle2=(Node1[j,:], Node1[j+1,:], Node1[j+3,:]), j=0, 1, ..., n-3 Angle3=(Node1[j,:], Node1[j+1,:], Node1[j+4,:]), j=0, 1,..., n-3 max a1 ,max p1 =Max(Angle1,Angle2,Angle3) max a2 ,max p2 =SeMax(Angle1,Angle2,Angle3) Where n represents the length of the coordinate array of the node array, Angle1, Angle2, and Angle3 represent the included angles of the three points obtained by the law of cosines, j represents the index of the current plant node, j+1 represents the index of the next plant node, and j = 0, 1, ..., n-3, Max represents finding the largest angle among the three angles and their corresponding points, SeMax represents finding the second largest angle among the three angles and their corresponding points, and max represents finding the second largest angle among the three angles and their corresponding points. a1 This represents the maximum angle value among Angle1, Angle2, and Angle3 calculated using the Max function. a2 This represents the second largest angle value among Angle1, Angle2, and Angle3 calculated using the SeMax function, where max represents the maximum angle. p1 Corresponding max a1 The position of the plant node, max p2 Corresponding max a2 The location of the plant node; Set the error angle; if max a1 and max a2 If the difference is less than the error angle, then calculate the maximum value. p1 To the node array Node1[j+1,:] and max p2 The Euclidean distance to the node array Node1[j+1,:] is calculated. The node with the smallest Euclidean distance is the main stem node of the plant. This process is repeated to traverse the plant node array until all plant nodes have been evaluated, and then the main stem node array is selected. After determining all the main stem nodes of the plant, the region judgment function is executed to divide the branch node arrays on the left and right sides of the main stem of the plant. The plant nodes in each region are judged according to the above steps, and the plant branch nodes are stored in the plant branch node array until all plant nodes are judged, and the category of all plant nodes and their respective branches are determined.

6. The method for extracting plant morphological phenotypic information according to claim 1, characterized in that, Each plant node is represented as a rectangle, including: extracting the coordinates of each plant node sequentially from the main stem node array and the branch node array; for each plant node, setting a coordinate pair (x, y), where x and y represent the horizontal and vertical positions of the plant node in the plant morphology image, respectively; setting the width of the plant node to w and the height to h; and obtaining the coordinates of the upper left corner (x, y) of the rectangle of the plant node in the actual plant morphology image. lu ,y lu ) and the coordinates of the lower right corner (x ru ,y ru To determine the position and size of the rectangular frame of the plant node, the calculation formula is as follows: Where, x lu The x-coordinate of the top-left corner of the rectangle containing the plant node in the actual plant morphology image is y. lu The x-coordinate represents the top-left corner of the rectangle containing the plant node in the image of the actual plant morphology. ru The x-coordinate of the lower right corner of the rectangle containing the plant node in the actual plant morphology image is y. ru The vertical coordinate of the lower right corner of the rectangle containing the plant node in the actual plant morphology image is represented by i, where i is a natural number and n is the total number of plant nodes.

7. The method for extracting plant morphological phenotypic information according to claim 6, characterized in that, The steps for obtaining the paths of adjacent plant nodes in an image that closely resembles the actual plant morphology include: The plant morphology image is processed to obtain a plant mask morphology image. The coordinates (x, y, y) of the top left corner of the rectangle containing adjacent plant nodes of the same type in the real plant morphology image are extracted sequentially. lu ,y lu ) and the coordinates of the lower right corner (x ru ,y ru The extreme coordinates (x, y) between two adjacent plant nodes are calculated. lmin ,y lmin ), (x rmax ,y rmax ), to determine the plant location boundary between two adjacent plant nodes; The boundary mask contour of the plant part is dimensionally reduced to obtain the center point, principal direction, and two orthogonal direction axes of the target contour. The contour is then moved up and down along the orthogonal directions of these direction axes. After the movement, the point of tangency between the direction axis and the target contour is the boundary point of the minimum boundary rectangle of the target contour. The contour tangent is the direction axis after the movement, i.e., y = k1x + b. * y = k2x + b * The specific formula is expressed as follows: Where, contour represents the set of boundary points, k1 and k2 represent the slopes of the two direction axes, respectively, and b * b represents the intercept of the line. * The maximum and minimum values ​​correspond to the points farthest from the central axis along two orthogonal directions, respectively. Find the point with the greatest distance from the direction axis along the contour lines, and take the arithmetic mean of the x-coordinates of the points with the greatest distance from the axis to obtain the x-coordinate C of the center point. x The y-coordinate C of the center point is obtained by taking the arithmetic mean of the y-coordinates of the points with the maximum off-axis distance. y The center point (C) is obtained. x C y ), calculate the maximum off-axis distance point corresponding to the k value, determine the width and height of the target minimum boundary rectangle, and obtain the path of adjacent plant nodes in the real plant morphology image.

8. The method for extracting plant morphological phenotypic information according to claim 7, characterized in that, The steps for constructing the plant morphology include: Determine the lowest point of each plant branch node. For a node on the main stem of the plant, calculate the Euclidean distance between it and the lowest point of the plant branch node. Determine two direction vectors, including the branch direction vector and the main stem direction vector. Calculate the angle between the direction vectors of the lowest point of the branch and the node of the main stem. The formula for calculation is: in, This represents the branching direction vector of the plant. θ represents the direction vector of the main stem of the plant, and θ represents the angle between the direction vector of the lowest point of the branch and the node of the main stem. From the calculated Euclidean distances, the plant node with the shortest distance is selected as the candidate plant node. The angle difference between the main stem direction vector and the branch direction vector of the candidate plant node is checked to see if it is within θ. If the condition is met, the plant branch node is connected to the plant main stem node. Each plant branch node is connected in turn to construct the plant morphology.

Images