Seedling information processing method and device based on biological prior constraint, and medium

Abstract

This application discloses a method, device, and medium for seedling information processing based on biological prior constraints, belonging to the field of information processing technology. The method includes the following steps: acquiring seedling data to be completed, and extracting the coordinates, node type, and local feature vector of each key point of the seedling from the data to be completed; wherein the local feature vector includes multiple local features of the key point in its neighborhood; generating an initial heterogeneous graph based on the coordinates and the node type; determining multiple hidden nodes based on a first determination rule, the coordinates, the node type, and the data to be completed; wherein the first determination rule is generated based on biological prior constraints; and completing the initial heterogeneous graph based on the hidden nodes, coordinates, node type, and local feature vector to obtain a target heterogeneous graph. This application can improve the accuracy of the seedling topology obtained after completion.

Description

Technical Field

[0001] This application belongs to the field of information processing technology, specifically relating to a method, device and medium for seedling information processing based on biological prior constraints. Background Technology

[0002] In modern agricultural breeding, cultivation monitoring, and plant phenomics analysis, high-throughput, automated measurement of phenotypic parameters such as seedling uprightness, crown width, main stem extension direction, and crown projection range is required. However, before these phenotypic parameters can be measured in a high-throughput, automated manner, seedlings often experience leaf shading, severe organ overlap, and limited shooting angles in actual field or seedling environments. Therefore, it is necessary to complete the seedling shading topology.

[0003] Currently, traditional techniques represent seedling structures as ordinary graphs or point sets, establishing connections based on geometric proximity or data-driven weights to complete the topology. However, this method is prone to generating erroneous topologies that deviate from the actual growth structure of seedlings, resulting in low accuracy of the completed seedling topology. Summary of the Invention

[0004] The purpose of this application is to provide a seedling information processing method, device, and medium based on biological prior constraints, which can improve the accuracy of the seedling topology obtained by completion.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide a seedling information processing method based on biological prior constraints, the method comprising: Acquire the seedling data to be completed, and extract the coordinates, node type, and local feature vector of each key point of the seedling from the data to be completed; wherein, the local feature vector includes multiple local features of the key point in its neighborhood; An initial heterogeneous graph is generated based on the coordinates and the node type; Based on the first determination rule, the coordinates, the node type, and the data to be completed, multiple hidden nodes are determined; wherein, the first determination rule is generated based on biological prior constraints; The initial heterogeneous graph is completed based on the hidden nodes, the coordinates, the node types, and the local feature vectors to obtain the target heterogeneous graph.

[0006] Secondly, embodiments of this application provide a seedling information processing device based on biological prior constraints, the seedling information processing device based on biological prior constraints comprising: An extraction module is used to acquire the data to be completed for the seedlings, and to extract the coordinates, node type, and local feature vector of each key point of the seedlings from the data to be completed; wherein, the local feature vector includes multiple local features of the key point in its neighborhood; The first generation module is used to generate an initial heterogeneous graph based on the coordinates and the node type; The first determining module is used to determine multiple hidden nodes based on a first determination rule, the coordinates, the node type, and the data to be completed; wherein the first determination rule is generated based on biological prior constraints. The completion module is used to complete the initial heterogeneous graph based on the hidden nodes, the coordinates, the node type, and the local feature vectors to obtain the target heterogeneous graph.

[0007] Thirdly, embodiments of this application provide a computer device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0008] Fourthly, embodiments of this application provide a computer-readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0009] Fifthly, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0010] In this embodiment, seedling data to be completed is acquired, and the coordinates, node type, and local feature vectors of each key point of the seedling are extracted from the data. The local feature vectors include multiple local features of the key point within its neighborhood. An initial heterogeneous graph is generated based on the coordinates and node type. Multiple hidden nodes are determined based on a first determination rule, the coordinates, the node type, and the data to be completed. The first determination rule is generated based on biological prior constraints. The initial heterogeneous graph is completed based on the hidden nodes, the coordinates, the node type, and the local feature vectors to obtain a target heterogeneous graph. That is, hidden nodes are generated by combining biological prior constraints, thereby ensuring that the completed target heterogeneous graph conforms to anatomical connection rules and physical laws, avoiding the generation of erroneous topologies that violate the true growth structure of the seedling, and thus improving the accuracy of the completed seedling topology. Attached Figure Description

[0011] Figure 1This is a flowchart illustrating a seedling information processing method based on biological prior constraints provided in some embodiments of this application; Figure 2 This is a structural block diagram of a seedling information processing device based on biological prior constraints provided in some embodiments of this application; Figure 3 These are internal structural diagrams of a computer device provided in some embodiments of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0014] In one exemplary embodiment, this application proposes a seedling information processing method based on biological prior constraints. The following, in conjunction with the accompanying drawings, provides a detailed description of the seedling information processing method based on biological prior constraints provided by this application through specific embodiments and application scenarios.

[0015] Reference Figure 1 The method includes steps 102-108. Wherein: Step 102: In response to the input of the seedling data to be completed, extract the coordinates, node type and local feature vector of each key point of the seedling from the data to be completed; wherein, the local feature vector includes multiple local features of the key point in the neighborhood.

[0016] In some embodiments, the data to be completed may include, but is not limited to: two-dimensional images, depth images, RGB-D images, and three-dimensional point cloud data.

[0017] In some embodiments, the data to be completed is used to extract feature maps and coordinates. The output is based on this feature map. When the data to be completed is a two-dimensional image, a depth image, or an RGB-D image, a convolutional neural network can be used to extract pixel-level feature maps; when the data to be completed is three-dimensional point cloud data, a point cloud feature extraction network or a voxel encoding network can be used to extract point-level features.

[0018] In some embodiments, node types may include, but are not limited to: root node, stem node, leaf node, growth vertex node, inflorescence node, fruit node, and branch node. It is understood that node types may differ for different seedlings, and the connection relationships are subject to anatomical semantic constraints.

[0019] It should be noted that for each keypoint, a Softmax classifier can be used to output the probability value of the keypoint belonging to each node type, thereby obtaining the type confidence vector corresponding to each keypoint. The node type of the keypoint is a type confidence vector. The maximum value in the keypoint corresponds to the node type, and this maximum value is the type confidence of the corresponding node type.

[0020] For example, if a key point is classified by Softmax and the output is a trunk probability of 0.85, a branch probability of 0.10, and a leaf probability of 0.05, then the predicted category of the node is the trunk, with a confidence level of 0.85.

[0021] In some embodiments, for each key point, local feature vectors within its neighborhood can be extracted separately. .

[0022] It should be noted that, It can be based on the current key point on the backbone path and its preset neighborhood. The 3D coordinates of key points are constructed to calculate local bending trend values, representing the geometric changes at the current key point. Specifically, taking the current key point as the center, the coordinates of the points before and after it along the main path are taken. 1 neighboring key points; calculate the vector difference between the central key point and its neighboring key points: ,in, Indicates the central key point and its neighboring key points. The coordinate difference along the axis, i.e., the three-dimensional vector difference. Components on the axis, The x-coordinate of the central key point The x-coordinate of the neighborhood key point; , The vector difference is respectively in shaft and The components on the axis are defined as follows: This will not be elaborated further here. The vector differences are concatenated in a preset order to form a multidimensional local feature vector. Among them, It is an odd number and is usually 3, 5 or 7, with the specific value determined based on the seedling point cloud density.

[0023] In summary, all key points can be represented as a set. .

[0024] Step 104: Generate an initial heterogeneous graph based on coordinates and node types.

[0025] In some embodiments, generating an initial heterogeneous graph based on coordinates and node type includes: If the first Euclidean distance, direction vector, and node type connection relationship of any node pair in all key points meet the second decision rule, an initial edge is established for any node pair; wherein, the first Euclidean distance and direction vector are calculated based on coordinates, the node type connection relationship is determined based on the two node types of any node pair, and the node type is determined based on type confidence; the second decision rule is generated based on biological prior constraints.

[0026] Generate an initial heterogeneous graph based on all key points and initial edges.

[0027] In some embodiments, the second determination rule includes a second distance rule, a second connection rule, and a second direction rule.

[0028] The second distance rule indicates that the first Euclidean distance is less than or equal to the first distance threshold.

[0029] The second connection rule indicates that the node type connection relationship is the target connection relationship.

[0030] The second direction rule indicates the direction vector. The angle of deviation from the preset direction is less than or equal to the first angle threshold.

[0031] In some embodiments, the preset direction may include, but is not limited to, the direction of gravity. Preset growth direction.

[0032] It should be noted that the connection rules are related to anatomy, while the distance rules, direction rules, and subsequent bending rules are related to physical laws.

[0033] In some embodiments, all keypoints can first be classified according to the node type to which they belong, resulting in a node set corresponding to each node type. For example, all keypoints can be classified as root node type. Stem node type Leaf node type and growth vertex node type wait.

[0034] An arbitrary node pair is a pair of nodes formed by drawing one node from each of two arbitrary node sets. For example, the two nodes in an arbitrary node pair come from... and .

[0035] In some embodiments, this can be done directly at all key points, i.e., the set of key points. Extract any two key points to form any node pair.

[0036] In some embodiments, biologically based prior constraints refer to biologically based compliance constraints, which are anatomical semantic and physical constraints. Compliance may include, but is not limited to, node type compliance, distance compliance, direction compliance, and bending compliance.

[0037] It should be noted that plant anatomy is the study of the internal structure and organ organization of plants. In the embodiments of this application, the classification of node types (such as root nodes, stem nodes, leaf nodes, growth vertex nodes, etc.) is itself based on the classification of plant organs in plant anatomy. Based on this, the constraint corresponding to the anatomical semantics refers to: restricting the allowed connection rules between corresponding types of nodes in the graph structure according to the actual spatial connection relationships between different organs in plant anatomy. It can be understood that the target connection relationship is the actual spatial connection relationship.

[0038] For example, in plant anatomy, roots and stems are directly connected (both belong to the main axis system), stems and leaves are directly connected (leaves grow at stem nodes), and stems and apical meristems are directly connected (apical meristems are located at the stem tip). Roots and leaves are generally not directly connected (they belong to the underground and aboveground systems respectively, and are usually connected through the stem). Therefore, this application defines node type combinations such as "root-stem," "stem-leaf," and "stem-apical" as the "allowed connection set," and excludes combinations that do not conform to anatomical facts (such as a direct "root-leaf" connection).

[0039] In other words, the specific criterion for judging the compliance of node types is whether the combination of types belongs to the set of allowed connections defined by anatomical semantics. The connection between node types and anatomy is that each node type corresponds to a category of anatomical organs in the plant, and the possibility of connections between organs is entirely determined by the anatomical facts of the plant.

[0040] In some embodiments, no initial edge is constructed for any pair of nodes that does not satisfy biological prior constraints, thereby limiting erroneous connections from the source.

[0041] In some embodiments, the first distance threshold It can be the average distance between the center points of adjacent true stem segments Double the standard deviation ,Right now .

[0042] In some embodiments, the first angle threshold These are empirical values, for example, given by the 95th percentile of the historical skewness distribution of the seedling trunk relative to the direction of gravity.

[0043] In some embodiments, the initial edge carries first metadata, which may include, but is not limited to, a first Euclidean distance, a direction vector, and an edge type. The edge type is the connection relationship between two nodes; for example, if the two nodes are a root node and a stem node, then the edge type is a connection between the root node and the stem node.

[0044] In some embodiments, the target connection relationship is constructed based on node type compliance, such as a connection between the root node and the stem node being legal, while a connection with a leaf node is illegal. The target connection relationship may include, but is not limited to: the connection between the root node and the stem node, the connection between the stem node and the leaf node, the connection between the leaf node and the growth vertex node, the connection between the stem node and the growth vertex node, the connection between the stem node and the inflorescence node, and the connection between the fruit node and the branch node.

[0045] In some embodiments, the initial heterogeneous graph ,in, A set of nodes or a set of key points ; Let it be the set of edges; The biological constraint matrix serves as the generation condition for the target heterogeneous graph.

[0046] Step 106: Based on the first decision rule, coordinates, node type and data to be completed, determine multiple hidden nodes; wherein, the first decision rule is generated based on biological prior constraints.

[0047] In some embodiments, the node type includes at least a stem node; based on a first determination rule, coordinates, node type, and data to be completed, multiple hidden nodes are determined, including: If the first Euclidean distance, node type matching relationship, spatial region, and direction vector of any node pair among all keypoints meet the first judgment rule, then any node pair is determined as a candidate completion pair. Among them, the first Euclidean distance and direction vector are calculated based on coordinates, the node type matching relationship is determined based on the two node types of any node pair and the stem node, the node type is determined based on type confidence, and the spatial region is the spatial region where any node pair is located in the data to be completed. The first judgment rule is generated based on biological prior constraints.

[0048] The number of hidden nodes is determined based on the second Euclidean distance corresponding to each candidate completion pair.

[0049] Generate the corresponding number of hidden nodes between the two nodes of each candidate completion pair.

[0050] In some embodiments, the first determination rule includes a first distance rule, a first connection rule, a first direction rule, and a first spatial region rule.

[0051] The first distance rule indicates that the first Euclidean distance is greater than the first distance threshold and less than or equal to the second distance threshold.

[0052] The first connection rule indicates that the node type matching relationship allows connections through at least one stem node.

[0053] The first direction rule indicates that the angle between the direction vector and the preset direction is less than or equal to the first angle threshold.

[0054] The first spatial region rule indicates that the spatial region to which it belongs is the target region.

[0055] In some embodiments, candidate completion pairs are used to generate hidden nodes, that is, the node between the two nodes of a candidate completion pair is a hidden node that is obscured by leaves, branches, etc.

[0056] Based on this, the first decision rule is used to determine whether the two nodes of any pair of nodes are occluded.

[0057] In some embodiments, the condition is: the first Euclidean distance is greater than the first distance threshold, used to exclude node pairs that have already had initial edges constructed; the condition is: the first Euclidean distance is less than or equal to the second distance threshold, used to exclude cases that cannot be occluded.

[0058] In some embodiments, the second distance threshold can be a linear combination of the mean and standard deviation of the distances between the center points of adjacent true stem segments, and satisfies the following conditions: ,in, This is the second distance threshold.

[0059] In some embodiments, allowing a connection through at least one stem node means that in any pair of nodes, at least one node, after being connected to the stem node, belongs to the target connection relationship.

[0060] In some embodiments, the target region is a low-response region or a region with no response.

[0061] In some embodiments, quantity It can be calculated using the following formula 1.

[0062] Formula 1 in, For nodes With nodes The Euclidean distance between them; This represents the floor function, used to ensure that the number of hidden nodes generated is sufficient to cover the nodes. With nodes The gap between them.

[0063] In one embodiment, the initial position of the hidden node can be generated by spline interpolation, local orientation field recursion, or Kalman prediction, in addition to linear interpolation.

[0064] Step 108: Complete the initial heterogeneous graph based on hidden nodes, coordinates, node types, and local feature vectors to obtain the target heterogeneous graph.

[0065] In some embodiments, the initial heterogeneous graph is completed based on hidden nodes to obtain the target heterogeneous graph, including: The biological constraint matrix is ​​generated based on the third decision rule, coordinates, node type, and local feature vectors; wherein the third decision rule is generated based on biological prior constraints.

[0066] Add all hidden nodes to the initial heterogeneous graph to obtain the graph to be updated.

[0067] Perform message passing operations on the graph to be updated in order to update the biological constraint matrix.

[0068] The target heterogeneity map is obtained based on the updated biological constraint matrix.

[0069] It should be noted that the updated biological constraint matrix is ​​used to remove invalid edges from the graph to be updated.

[0070] In some embodiments, the third determination rule includes a third distance rule, a third connection rule, a third direction rule, and a third bending rule.

[0071] The third distance rule indicates that the first Euclidean distance is less than or equal to the first distance threshold.

[0072] The third connection rule indicates that the node type connection relationship is a target connection relationship.

[0073] Third-direction rule indication The angle of deviation from the preset direction is less than or equal to the first angle threshold.

[0074] The third bending rule indicates that the bending trend value is less than or equal to the trend value threshold. The bending trend value is determined based on local feature vectors; the smaller the bending trend value, the smoother the local bending trend. The trend value threshold is an empirical value and is not limited in this embodiment. The bending trend value can be the gradient of the local feature vectors.

[0075] In addition, local eigenvectors can also indirectly reflect the bending angle by calculating the cosine of the angle between adjacent vectors in the local eigenvectors. Mapping the cosine of the included angle to the degree of bending tendency, the smaller the included angle ( The closer The value of θ indicates a more severe bend; the bend trend value is the negative of the cosine of the included angle (or other monotonic mapping). The smaller the trend value, the more obvious the bend.

[0076] In some embodiments, when generating the biological constraint matrix, if the node type does not conform to the third distance rule... If the node type conforms to the third distance rule, perform subsequent judgments.

[0077] Subsequent determinations include: Assign basic weights if the first Euclidean distance conforms to the third distance rule. ;exist If the third-party directional rules are met, directional weights are added to the base weights. If the bending trend value conforms to the third bending rule, bending weights can be added. .

[0078] In some embodiments, the biological constraint matrix .

[0079] In some embodiments, the graph to be updated includes an initial heterogeneous graph, and hidden nodes, edges between hidden nodes and their corresponding candidate completion pairs, and edges between each hidden node are added to the initial heterogeneous graph.

[0080] Understandable. It can also be generated when constructing the initial heterogeneous graph; therefore, in one embodiment, the initial heterogeneous graph... .

[0081] For message passing operations, the details are as follows: For each node in the graph to be updated Determine its set of neighboring nodes. .

[0082] For each adjacent edge The propagation weight (i.e., edge weight) of the edge is calculated using the following formula 2. .

[0083] Formula 2 in, The attention weights obtained through learning.

[0084] Node neighboring nodes Current layer features Through linear transformation matrix Mapping to obtain the message to be transmitted .

[0085] Use edge weight right Weighting is performed to form neighbor nodes. For nodes Effective contribution .

[0086] Node Sum the weighted messages from all the neighbors: .

[0087] Apply a nonlinear activation function to the summation result , obtain node The next layer of features: .

[0088] Dynamically updated during message passing. : If the node type of an edge is invalid, set it to zero; if the angle between the edge and the previous stem segment is too large, reduce its weight; if the edge direction is significantly against gravity or deviates from the main growth direction, reduce its weight; if the edge helps to shorten the main path cost between the root node and the growth vertex node, retain a higher weight.

[0089] Here, attenuation refers to the attenuation propagation mechanism of the cost value, and the attenuation operation is as follows: When the path starts from a node in the graph to be updated Pass to adjacent nodes At that time, node From node The value of the inheritance is calculated using the following formula 3.

[0090] Formula 3 in, For nodes Current path cost; The attenuation coefficient is... This is used to control the degree of attenuation of path cost as it is propagated along the search direction. For nodes To the node The edge cost (usually the Euclidean distance between the two points).

[0091] Wherein, the main path cost is the cumulative cost of the path traversed from the starting node (root node) to the current candidate node, denoted as . It can be calculated using the following formula 4.

[0092] Formula 4 in, For the first on the path The node to the first The edge cost of each node; This represents the depth of the current node in the search tree. This is the depth index of each node on the path.

[0093] It should be noted that during topological reasoning, the main path cost is used to prioritize multiple candidate paths; the path with the lowest cost is prioritized for expansion and retention; the main path cost comprehensively considers the geometric length of the path (through...). (reflected) and path attenuation characteristics (through) reflect).

[0094] In this context, retaining higher weights refers to retaining topological inference results with higher confidence levels and suppressing or discarding results with lower confidence levels during non-maximum suppression (or similar screening) operations. The weight refers to the type confidence level mentioned earlier, with a value range of [value missing]. It is determined by the maximum probability value output by the Softmax classifier.

[0095] Specifically, the procedure for retaining higher weights is as follows: Set weight threshold ( ),For example For all candidate topology inference results, calculate their corresponding confidence scores; if the confidence score of a candidate result is... If the confidence level is high, then retain it; if the confidence level is low, then retain it. If the result is not high enough, it will be eliminated; if multiple candidate results with high confidence overlap, the one with the highest confidence will be retained and the others will be suppressed.

[0096] In some embodiments, this embodiment for The specific value is not limited, and can be adjusted according to factors such as point cloud quality and seedling type in actual application.

[0097] When the updated edge weight satisfies If the edge is valid, keep it; otherwise, delete it. The threshold that maximizes the edge connection accuracy can be determined using the validation set.

[0098] The hidden nodes corresponding to the retained edges are formally incorporated into the initial heterogeneous graph to obtain the target heterogeneous graph. .in, This represents the set of hidden nodes corresponding to the edges that are preserved. express and The union of; This represents the union of the initial edges and the retained edges in the initial heterogeneous graph.

[0099] This embodiment, in response to the input of seedling data to be completed, extracts the coordinates, node type, and local feature vectors of each key point of the seedling from the data to be completed. The local feature vectors include multiple local features of the key point within its neighborhood. An initial heterogeneous graph is generated based on the coordinates and node type. Multiple hidden nodes are determined based on a first decision rule, the coordinates, node type, and the data to be completed. The first decision rule is generated based on biological prior constraints. The initial heterogeneous graph is completed based on the hidden nodes to obtain the target heterogeneous graph. That is, by combining biological prior constraints to generate hidden nodes, the completed target heterogeneous graph conforms to anatomical connection rules and physical laws, avoiding the generation of erroneous topologies that violate the true growth structure of the seedling, thereby improving the accuracy of the seedling topology obtained after completion.

[0100] Furthermore, seedling information processing methods based on biological prior constraints also include: Connectivity component analysis was performed on the target heterogeneous graph to obtain multiple candidate connected components.

[0101] Determine the target connected component from all candidate connected components.

[0102] A spatial response map is generated based on the response values ​​of nodes in the target heterogeneous graph.

[0103] The spatial mask is obtained based on the spatial response map.

[0104] Based on the target connected components, spatial mask, and target heterogeneous graph, multiple phenotypic parameters of the seedlings are obtained.

[0105] In some embodiments, the target connected component is the candidate connected component with the highest score, wherein the score can be calculated using the following formula 5.

[0106] Formula 5 in, This represents the number of nodes in the candidate connected components. This represents the total edge weight of the candidate connected components. Indicates whether a candidate connected component contains the root node. Indicates whether a candidate connected component contains growing vertex nodes. , , , The default weights are non-negative.

[0107] It is understandable that candidate connected components with scores lower than the target connected component belong to noise components that do not contain root nodes, have a total edge weight below a threshold, or consist only of isolated leaf nodes.

[0108] Treat the target connected component as the main connected subgraph That is, the connected subgraph that is most likely to correspond to the main trunk structure of the seedling obtained by screening from the target heterogeneous graph.

[0109] In some embodiments, when generating a spatial response map based on the response values ​​of nodes in the target heterogeneous graph, the spatial response map is generated by projecting the responses of each node in the target heterogeneous graph back into the input space according to their coordinates. The response value is a scalar value calculated by skeletonization algorithms (such as Zhang-Suen skeletonization or distance-transform-based skeletonization methods) for each spatial location when processing point cloud data, reflecting the degree to which that location belongs to a skeleton point.

[0110] In one embodiment, the spatial response graph can be generated not only by node activation projection, but also by edge response projection, node-edge joint projection, or backbone probability graph.

[0111] In some embodiments, a space mask By analyzing the spatial response map The threshold is obtained through smoothing and thresholding. This threshold can be the response mean plus standard deviation, a fixed-ratio threshold, or the optimal threshold determined based on the validation set.

[0112] In some embodiments, the area covered by the spatial mask is defined as the backbone candidate region, and the area covered by the inverted mask is defined as the canopy periphery candidate region.

[0113] In some embodiments, phenotypic parameters may include, but are not limited to: uprightness measurement branch. Crown width measurement branch Branching angle, internode length, trunk curvature, fruit distribution range, and leaf area index.

[0114] Among them, the phenotypic parameters are and Taking this as an example, we will explain how to obtain multiple phenotypic parameters of seedlings based on target connected components, spatial masks, and target heterogeneous graphs.

[0115] Among them, for , space mask Acting on the main connected subgraph To retain candidate nodes and edges for the main axis, a minimum-cost path is searched from the root node to the growing vertex node to obtain the main axis. Fit the principal direction vector based on the principal axis. ; with reference to the direction of gravity Based on the above, calculate using the following formula 6. .

[0116] Formula 6 in, It is the inverse cosine function; Let be the norm of the vector, i.e., its magnitude.

[0117] Among them, for Invert the mask Acting on enhancing heterogeneous graphs Retain leaf nodes and non-main stem nodes located in the candidate region on the periphery of the canopy; read the planar coordinates or projected coordinates of the retained nodes and calculate the circumscribed envelope; calculate the canopy width parameters based on the circumscribed envelope. The crown width parameter can be the maximum lateral span, the envelope area, or a combination of both.

[0118] In some embodiments, to improve the accuracy of keypoint detection, node classification, and edge connection determination, keypoint detection, occlusion completion, and decoupled phenotypic parameter measurement can be jointly trained. The total loss function during training can be set as shown in Formula 7 below: Formula 7 in, For detecting losses at key points; The loss is used for node classification. This represents the connectivity loss, used to constrain the continuity of the main path before and after completion.

[0119] This embodiment avoids directly regressing multiple phenotypic parameters within the same shared feature space, which would cause the trunk information and canopy boundary information to be coupled together. This operation does not consider that uprightness measurement relies more on the trunk skeleton and orientation information, and canopy width measurement relies more on the distribution and boundary information of the outer leaves. Instead, it improves measurement stability by decoupling the trunk and outer leaves of the target heterogeneous map before performing phenotypic measurements.

[0120] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0121] Based on the same inventive concept, this application also provides a seedling information processing device based on biological prior constraints for implementing the seedling information processing method based on biological prior constraints described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more seedling information processing device embodiments based on biological prior constraints provided below can be found in the limitations of the seedling information processing method based on biological prior constraints described above, and will not be repeated here.

[0122] In one exemplary embodiment, such as Figure 2 As shown, a seedling information processing device based on biological prior constraints is provided, comprising: an extraction module 100, a first generation module 200, a first determination module 300, and a completion module 400, wherein: The extraction module 100 is used to acquire the seedling data to be completed, and extract the coordinates, node type and local feature vector of each key point of the seedling from the data to be completed; wherein, the local feature vector includes multiple local features of the key point in the neighborhood.

[0123] The first generation module 200 is used to generate an initial heterogeneous graph based on the coordinates and the node type.

[0124] The first determining module 300 is used to determine multiple hidden nodes based on a first determination rule, the coordinates, the node type, and the data to be completed; wherein the first determination rule is generated based on biological prior constraints.

[0125] The completion module 400 is used to complete the initial heterogeneous graph based on the hidden nodes, the coordinates, the node type, and the local feature vectors to obtain the target heterogeneous graph.

[0126] In one embodiment, the first generation module 200 is specifically used for: If the first Euclidean distance, direction vector, and node type connection relationship of any node pair in all key points conform to the second determination rule, an initial edge is established for the arbitrary node pair; wherein, the first Euclidean distance and the direction vector are calculated based on the coordinates, the node type connection relationship is determined based on the two node types of the arbitrary node pair, and the node type is determined based on the type confidence; the second determination rule is generated based on the prior constraints of the biology.

[0127] An initial heterogeneous graph is generated based on all the key points and the initial edges.

[0128] In one embodiment, the second determination rule includes a second distance rule, a second connection rule, and a second direction rule.

[0129] The second distance rule indicates that the first Euclidean distance is less than or equal to the first distance threshold.

[0130] The second connection rule indicates that the node type connection relationship is a target connection relationship.

[0131] The second direction rule indicates that the angle between the direction vector and the preset direction is less than or equal to the first angle threshold.

[0132] In one embodiment, the node type includes at least a stem node; the first determining module 300 is specifically used for: If the first Euclidean distance, node type matching relationship, spatial region, and direction vector of any node pair among all key points meet the first determination rule, the arbitrary node pair is determined as a candidate completion pair; wherein, the first Euclidean distance and the direction vector are calculated based on the coordinates, the node type matching relationship is determined based on the two node types and the stem node of the arbitrary node pair, and the node type is determined based on the type confidence; the spatial region is the spatial region where the arbitrary node pair is located in the data to be completed; the first determination rule is generated based on the prior constraints of the biology.

[0133] The number of hidden nodes is determined based on the second Euclidean distance corresponding to each candidate completion pair.

[0134] A corresponding number of hidden nodes are generated between the two nodes of each candidate completion pair.

[0135] In one embodiment, the first determination rule includes a first distance rule, a first connection rule, a first direction rule, and a first spatial region rule; The first distance rule indicates that the first Euclidean distance is greater than a first distance threshold and less than or equal to a second distance threshold; The first connection rule indicates that the node type matching relationship allows connections through at least one stem node; The first direction rule indicates that the angle between the direction vector and the preset direction is less than or equal to a first angle threshold; The first spatial region rule indicates that the spatial region to which it belongs is the target region.

[0136] In one embodiment, the completion module 400 is specifically used for: A biological constraint matrix is ​​generated based on a third decision rule, the coordinates, the node type, and the local feature vector; wherein the third decision rule is generated based on the prior constraints of the biology.

[0137] Add all hidden nodes to the initial heterogeneous graph to obtain the graph to be updated.

[0138] A message passing operation is performed on the graph to be updated in order to update the biological constraint matrix.

[0139] The target heterogeneity map is obtained based on the updated biological constraint matrix.

[0140] In one embodiment, the apparatus further includes: The fractal module is used to perform connected component analysis on the target heterogeneous graph to obtain multiple candidate connected components.

[0141] The second determining module is used to determine the target connected component from all candidate connected components.

[0142] The second generation module is used to generate a spatial response map based on the response values ​​of the nodes in the target heterogeneous graph.

[0143] The third generation module is used to obtain a spatial mask based on the spatial response map.

[0144] The fourth generation module is used to obtain multiple phenotypic parameters of the seedling based on the target connected component, the spatial mask, and the target heterogeneous graph.

[0145] Each module in the aforementioned seedling information processing device based on biological prior constraints can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0146] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a seedling information processing method based on biological prior constraints.

[0147] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0148] In one embodiment, a computer-readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps in the above-described method embodiments.

[0149] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0150] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0151] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0152] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A seedling information processing method based on biological prior constraints, characterized in that, The seedling information processing method based on biological prior constraints includes: Acquire the seedling data to be completed, and extract the coordinates, node type, and local feature vector of each key point of the seedling from the data to be completed; wherein, the local feature vector includes multiple local features of the key point in its neighborhood; An initial heterogeneous graph is generated based on the coordinates and the node type; Based on the first determination rule, the coordinates, the node type, and the data to be completed, multiple hidden nodes are determined; wherein, the first determination rule is generated based on biological prior constraints; The initial heterogeneous graph is completed based on the hidden nodes, the coordinates, the node types, and the local feature vectors to obtain the target heterogeneous graph.

2. The seedling information processing method based on biological prior constraints according to claim 1, characterized in that, The process of generating an initial heterogeneous graph based on the coordinates and the node type includes: If the first Euclidean distance, direction vector, and node type connection relationship of any node pair among all key points conform to the second determination rule, an initial edge is established for the arbitrary node pair; wherein, the first Euclidean distance and the direction vector are calculated based on the coordinates, the node type connection relationship is determined based on the two node types of the arbitrary node pair, and the node type is determined based on type confidence; the second determination rule is generated based on the prior constraints of the biology. An initial heterogeneous graph is generated based on all the key points and the initial edges.

3. The seedling information processing method based on biological prior constraints according to claim 2, characterized in that, The second determination rule includes a second distance rule, a second connection rule, and a second direction rule; The second distance rule indicates that the first Euclidean distance is less than or equal to the first distance threshold; The second connection rule indicates that the node type connection relationship is a target connection relationship; The second direction rule indicates that the angle between the direction vector and the preset direction is less than or equal to the first angle threshold.

4. The seedling information processing method based on biological prior constraints according to claim 1, characterized in that, The node type includes at least stem nodes; the determination of multiple hidden nodes based on the first determination rule, the coordinates, the node type, and the data to be completed includes: If the first Euclidean distance, node type matching relationship, spatial region, and direction vector of any node pair among all keypoints meet the first determination rule, the arbitrary node pair is determined as a candidate completion pair; wherein, the first Euclidean distance and the direction vector are calculated based on the coordinates, the node type matching relationship is determined based on the two node types of the arbitrary node pair and the stem node, the node type is determined based on type confidence; the spatial region is the spatial region where the arbitrary node pair is located in the data to be completed; the first determination rule is generated based on the prior constraints of the biology. The number of hidden nodes is determined based on the second Euclidean distance corresponding to each candidate completion pair. A corresponding number of hidden nodes are generated between the two nodes of each candidate completion pair.

5. The seedling information processing method based on biological prior constraints according to claim 4, characterized in that, The first determination rule includes a first distance rule, a first connection rule, a first direction rule, and a first spatial region rule; The first distance rule indicates that the first Euclidean distance is greater than a first distance threshold and less than or equal to a second distance threshold; The first connection rule indicates that the node type matching relationship allows connections through at least one stem node; The first direction rule indicates that the angle between the direction vector and the preset direction is less than or equal to a first angle threshold; The first spatial region rule indicates that the spatial region to which it belongs is the target region.

6. The seedling information processing method based on biological prior constraints according to claim 1, characterized in that, The process of completing the initial heterogeneous graph based on the hidden nodes, the coordinates, the node type, and the local feature vectors to obtain the target heterogeneous graph includes: A biological constraint matrix is ​​generated based on the third decision rule, the coordinates, the node type, and the local feature vector; wherein the third decision rule is generated based on the prior constraints of the biology. Add all hidden nodes to the initial heterogeneous graph to obtain the graph to be updated; A message passing operation is performed on the graph to be updated in order to update the biological constraint matrix; The target heterogeneity map is obtained based on the updated biological constraint matrix.

7. The seedling information processing method based on biological prior constraints according to claim 1, characterized in that, The seedling information processing method based on biological prior constraints also includes: Connectivity component analysis is performed on the target heterogeneous graph to obtain multiple candidate connected components; Determine the target connected component from all candidate connected components; A spatial response map is generated based on the response values ​​of the nodes in the target heterogeneous graph; The spatial mask is obtained based on the spatial response map; Based on the target connected component, the spatial mask, and the target heterogeneity graph, multiple phenotypic parameters of the seedling are obtained.

8. A seedling information processing device based on biological prior constraints, characterized in that, The device includes: An extraction module is used to acquire the data to be completed for the seedlings, and to extract the coordinates, node type, and local feature vector of each key point of the seedlings from the data to be completed; wherein, the local feature vector includes multiple local features of the key point in its neighborhood; The first generation module is used to generate an initial heterogeneous graph based on the coordinates and the node type; The first determining module is used to determine multiple hidden nodes based on a first determination rule, the coordinates, the node type, and the data to be completed; wherein the first determination rule is generated based on biological prior constraints. The completion module is used to complete the initial heterogeneous graph based on the hidden nodes, the coordinates, the node type, and the local feature vectors to obtain the target heterogeneous graph.

9. A computer device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the seedling information processing method based on biological prior constraints as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the seedling information processing method based on biological prior constraints as described in any one of claims 1-7.

Images