A method for extracting and detecting traits for melon breeding

Abstract

The application relates to the field of agricultural information technology and discloses a trait extraction detection method for melon breeding, which comprises the following steps: based on time sequence three-dimensional point cloud data of melon plants, independent organ structures are recognized and separated; through data correlation, cross-time-point tracking is realized, and independent life trajectories of various organs are formed; through analysis of the relationship between different organ instance point clouds, a three-dimensional topological connection diagram is reasoned and generated; source organs and sink organs existing in assimilation product supply relationships are determined; for each source organ and sink organ pair, dynamic indexes and dynamic trait parameters are respectively extracted; through analysis of the synergistic change relationship between the dynamic indexes and the dynamic trait parameters, the dynamic process synergistic efficiency is quantified as a breeding evaluation index. Through automatic construction of organ interaction networks from time sequence data and quantification of dynamic synergistic efficiency, the application realizes a leap from static trait measurement to system interaction performance analysis.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a method for extracting and detecting traits used in melon breeding. Background Technology

[0002] Currently, the extraction of phenotypic traits in melon breeding mainly relies on static observation and measurement of plants or organs. For example, analyzing leaf biochemical content through a single hyperspectral image acquisition, or measuring fruit volume using three-dimensional scanning at a single time point. These techniques can obtain a phenotypic snapshot at a specific moment.

[0003] However, these static, snapshot-like methods have fundamental limitations and cannot meet the needs of analyzing key biological processes in melon breeding. First, melon yield formation and quality accumulation are continuous dynamic processes, as evidenced by daily changes in fruit yield, sugar accumulation curves, and fluctuations in the photosynthetic capacity of the source organs over time. Static measurements completely lose this information about dynamic processes, and the differences between different genotypes in these dynamic patterns are crucial for breeding selection.

[0004] Secondly, current technologies cannot automatically establish and quantify the functional connections between organs at the plant level. In melon plants, there is a directional supply relationship of assimilates between specific leaves and specific fruits, and the efficiency of this "source-sink" interaction directly affects the final yield. Currently, there is a lack of technology that can automatically and objectively identify which leaves provide nutrients to which fruits from observational data, let alone dynamically quantify the effectiveness of this supply relationship. Therefore, breeding selection remains at the level of separately evaluating the isolated traits of source and sink organs, failing to perform synergistic efficiency selection at the system level.

[0005] Finally, the core technologies required to achieve the aforementioned dynamic relationship analysis remain unsolved. In particular, how to stably track the complete life cycle of each organ from time-series three-dimensional observation data, and on this basis, deduce the three-dimensional spatial connection topology between organs, is a key challenge that current technology has not yet solved. Without this foundation, any dynamic analysis of organ interactions is impossible.

[0006] Therefore, this paper proposes a method for extracting and detecting traits used in melon breeding. Summary of the Invention

[0007] The purpose of this invention is to solve the problem that the dynamic interaction efficiency between melon organs cannot be quantified in the prior art, and to propose a method for extracting and detecting traits for melon breeding.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a method for extracting and detecting traits for melon breeding, used to digitally analyze and evaluate the organ interaction efficiency of melon plants throughout their entire growth period, comprising the following steps: Step S1: Based on the time-series three-dimensional point cloud data continuously collected during the entire growth period of the melon plant, perform three-dimensional instance segmentation on the plant point cloud at each time point, and identify and separate the independent organ structures of the melon. Step S2: Through temporal point cloud registration and data association based on organ spatial continuity, cross-time point tracking of the same organ instance is achieved, forming independent life trajectories for each organ. Step S3: Based on temporal 3D point cloud data and life trajectory, by analyzing the spatial adjacency and geometric continuity relationships between point clouds of different organ instances, a 3D topological connection map representing the connection relationship between organs is inferred and generated. Step S4: Based on the three-dimensional topological connectivity graph, identify source organ and sink organ pairs that have assimilation product supply relationships; Step S5: For each source organ and depot organ pair, extract dynamic indicators representing the physiological activity of the source organ and dynamic trait parameters representing the morphological development of the depot organ from the time-series data corresponding to its life trajectory. Step S6: By analyzing the synergistic relationship between the dynamic indicators of the source organ and the dynamic trait parameters of the deposit organ over time, the synergistic efficiency of the dynamic process between organ pairs is quantified and used as a breeding evaluation indicator.

[0009] The beneficial effects of the technical solution provided by this invention include at least the following: This invention achieves complete digital recording of the independent life history of each organ by performing organ-level instance segmentation and cross-time point tracking on the temporal three-dimensional point cloud of the entire growth period of melon plants. This breaks through the limitations of traditional static phenotypic analysis and accurately captures the dynamic development process of key traits such as fruit enlargement and leaf senescence, providing a data foundation for genotype screening based on growth models.

[0010] This invention automatically infers and constructs a three-dimensional topological connection map representing the connection relationship between organs from time-series data, and determines the supply relationship pairs between source organs and sink organs accordingly. It can transform the abstract concept of "source-sink" interaction into a computable and verifiable topological network model, realizing the leap from isolated trait measurement to system interaction relationship analysis, and providing a core technical means for directly screening efficient and synergistic genotypes.

[0011] This invention extracts and analyzes the synergistic relationship between the dynamic indicators of physiological activity of source organs and the dynamic morphological development parameters of sink organs over time. It can quantify the synergistic efficiency of dynamic processes between organ pairs, going beyond simple measurements of final yield or quality. It directly reflects the spatiotemporal matching efficiency of photosynthetic product supply and utilization, providing breeders with a new and efficient digital trait for evaluating the effectiveness of the genotype "source-sink" system. Attached Figure Description

[0012] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention. Detailed Implementation

[0014] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for extracting and detecting traits for melon breeding according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0016] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0017] The specific scheme of the trait extraction and detection method for melon breeding provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Please see Figure 1 The diagram illustrates a flowchart of a method for extracting and detecting traits for melon breeding according to an embodiment of the present invention. This method is used to digitally analyze and evaluate the organ interaction efficiency of melon plants throughout their entire growth period, and includes the following steps: Step S1: Based on the time-series three-dimensional point cloud data continuously collected during the entire growth period of the melon plant, perform three-dimensional instance segmentation on the plant point cloud at each time point, and identify and separate the independent organ structures of the melon. Step S2: Through temporal point cloud registration and data association based on organ spatial continuity, cross-time point tracking of the same organ instance is achieved, forming independent life trajectories for each organ. Step S3: Based on temporal 3D point cloud data and life trajectory, by analyzing the spatial adjacency and geometric continuity relationships between point clouds of different organ instances, a 3D topological connection map representing the connection relationship between organs is inferred and generated. Step S4: Based on the three-dimensional topological connectivity graph, identify source organ and sink organ pairs that have assimilation product supply relationships; Step S5: For each source organ and depot organ pair, extract dynamic indicators representing the physiological activity of the source organ and dynamic trait parameters representing the morphological development of the depot organ from the time-series data corresponding to its life trajectory. Step S6: By analyzing the synergistic relationship between the dynamic indicators of the source organ and the dynamic trait parameters of the deposit organ over time, the synergistic efficiency of the dynamic process between organ pairs is quantified and used as a breeding evaluation indicator.

[0019] It should be noted that the entire growth period refers to the complete growth cycle of melons from seedling germination to fruit maturity, covering key growth stages such as budding, flowering, fruit setting, fruit expansion, and maturity. It is the time range that ensures the integrity of the dynamic analysis of organ interactions.

[0020] Temporal 3D point cloud data refers to a set of 3D spatial data of melon plants collected continuously at fixed time intervals throughout the entire growth period. Each data point contains 3D spatial coordinates and non-visible light band reflectance information, which can completely record the changes in the morphology and physiological characteristics of the plant and organs over time.

[0021] 3D instance segmentation refers to the use of a multimodal fusion intelligent segmentation algorithm to process the 3D point cloud of a plant at a single time point, enabling the differentiation of different organ categories such as leaves, stems, and fruits, as well as the independent separation of different individual organs within the same category, thus providing a foundation for subsequent organ tracking.

[0022] Independent organ structures refer to the functionally and morphologically independent components of a melon plant, mainly including leaves, stems, and fruits, and are the basic units for analyzing organ interaction effectiveness.

[0023] Temporal point cloud registration refers to using spatial alignment techniques such as the iterative nearest point algorithm to unify the 3D point clouds of plants collected at different time points into the same coordinate system, eliminating spatial offset caused by differences in collection location, and ensuring the accuracy of organ comparison across time points.

[0024] Organ spatial continuity refers to the correlation and consistency of the morphology, location, and connectivity of the same organ at different points in time, and is the core basis for realizing cross-time point tracking of organs.

[0025] Data association refers to establishing correspondences between organ instances by calculating the similarity of geometric and spectral features of organ instances at different time points, combined with topological constraints, thereby achieving identity association for the same organ.

[0026] An instance of the same organ refers to an individual of the same organ that continues to grow and develop throughout the entire reproductive period. Its identity is kept unique at different points in time through data association to avoid confusion with other organs.

[0027] Cross-time point tracking refers to continuously locking onto and tracking the dynamic changes of the same organ instance from time-series data of the entire reproductive period, and fully recording the data of its entire process from formation to maturity or shedding.

[0028] Life trajectory refers to the time-series data of the same organ instance throughout its reproductive life, including morphological parameters, physiological activity indicators, spatial location and other information at different time points, which can intuitively reflect the growth and development process of the organ.

[0029] Spatial adjacency refers to the proximity of different organ instances in three-dimensional space. It is determined by calculating parameters such as the minimum distance between organ point clouds and is the basis for judging whether there is a physical connection between organs.

[0030] Geometric continuity refers to the degree of matching of geometric features in the contact areas of different organ instances, including normal vector consistency and curvature continuity, which is used to help determine the tightness of the physical connection between organs.

[0031] A three-dimensional topological connection graph is a graphical model constructed with organ instances as nodes, persistent physical connections between organs as edges, and connection strength as edge weights. It can clearly present the connection network of organs within a plant.

[0032] Representing the connections between organs refers to visually reflecting the connected objects, connection status, and degree of connection of organs through the nodes and edges of a three-dimensional topological connection graph, providing a structural basis for subsequent source-library relationship determination.

[0033] Source organs are organs that can synthesize and export assimilates. In melon plants, they are mainly functional leaves, and their physiological activity directly determines the supply capacity of assimilates.

[0034] Sink organs are organs that consume or store assimilates. In melon plants, they are mainly fruits, and their morphological development reflects the efficiency of assimilate utilization.

[0035] Assimilate supply relationships refer to the physiological connections through which assimilates synthesized in source organs are transferred to sink organs via transport channels such as stems, and are the core manifestation of organ interaction efficacy.

[0036] Dynamic indicators refer to physiological activity parameters that change over time and are extracted from the life trajectory of source organs, including the cumulative value of normalized differential vegetation index and the mean value of red edge reflectance, which can quantify the assimilation product supply potential of source organs.

[0037] Dynamic trait parameters refer to morphological developmental parameters that change over time and are extracted from the life trajectory of deposit organs, including integral of volume change and peak growth acceleration, which can quantify the utilization efficiency of deposit organs for assimilates.

[0038] Synergistic change relationship refers to the degree of matching between the temporal variation patterns of source organ dynamic indicators and the temporal variation patterns of sink organ dynamic trait parameters, reflecting the temporal synchronicity of assimilation product supply and utilization.

[0039] Dynamic process synergy efficiency refers to a numerical index obtained by quantifying synergistic change relationships. The value ranges from 0 to 1. The higher the value, the better the synergy between the supply and utilization of assimilates between source and bank organs. It is the core evaluation basis for breeding screening.

[0040] Breeding evaluation indicators are quantitative parameters that can objectively reflect the superior characteristics of melon plants. The dynamic process synergy efficiency in this method is directly related to breeding objectives such as yield and quality, providing a standardized basis for screening superior genotype materials.

[0041] In one specific implementation, the melon variety M147 was selected as the experimental material and planted in a standardized experimental field. Conventional field management was adopted to ensure uniform plant growth. Throughout the entire growth period, a field phenotyping robot equipped with a 16-line lidar and a multispectral camera was used to collect data. The data collection frequency was set to once every 5 days, covering 12 time points from the seedling stage to maturity. RTK positioning technology was used to ensure consistent spatial location for each data collection, generating temporal 3D point cloud data containing three-dimensional spatial coordinates and near-infrared and red-edge band reflectivity information.

[0042] For step S1, a 3D instance segmentation network is constructed, comprising a dual-branch feature extraction module and a cross-modal attention fusion module. The first branch uses a PointNet++ network to extract geometric features of the point cloud, including curvature, normal vector, and local density. The second branch uses a 1D convolutional network to extract spectral features corresponding to multispectral reflectance. The cross-modal attention fusion module dynamically allocates weights by calculating the mutual information entropy of the two types of features. Finally, the mask generation network outputs the organ category label and instance ID for each point, achieving accurate differentiation and independent separation of leaves, stems, and fruits, with a segmentation accuracy of over 92%.

[0043] For step S2, the iterative nearest-neighbor algorithm is used to register the point cloud data of adjacent time points, with the registration error controlled within 2mm. Based on the registered point cloud data, a plant-level organ connection topology graph is constructed for each time point. Using organ instances as nodes, edges are established according to the rules that the minimum distance between the leaf base point cloud and the stem surface point cloud is less than 6mm, the minimum distance between the fruit stem region and the stem surface point cloud is less than 4mm, and the minimum distance between point clouds of fruits on the same branch is less than 12mm. A graph neural network algorithm is used to solve the graph matching optimization model, applying topological consistency constraints. That is, if an organ node at a previous time point matches an organ node at a subsequent time point, the matching results of its adjacent nodes must maintain an adjacency relationship. Ultimately, cross-time point tracking of the same organ instance is achieved, generating a life trajectory containing morphological and physiological data at each time point.

[0044] For step S3, the point cloud data of any two organ instances are traversed to find point pairs with a distance of less than 2.5 mm that constitute a contact region. Geometric features such as normal vector consistency, surface area, and curvature continuity of the contact region are extracted. These geometric features are input into a pre-trained random forest classifier to determine whether a physical connection exists between organs. The connection strength is calculated by multiplying the normalized contact surface area by the normal vector consistency. Connections that persist for more than 70% of the time points are selected. A three-dimensional topological connection graph is constructed with organ instances as nodes, persistent connections as edges, and connection strength as edge weights.

[0045] For step S4, candidate source organs and candidate pool organs are identified from the 3D topological connectivity graph. Candidate source organs are leaf instances with a normalized difference vegetation index greater than 0.6, and candidate pool organs are fruit instances with a volume growth rate greater than 0.5 cm³ / day. Using each candidate pool organ as the endpoint, a directed path traversal is performed from leaf to stem to fruit, collecting all candidate source organs reachable from that pool organ to form an initial source-pool pair set. The product of the edge connection strengths on each initial source-pool pair path is calculated (taking the maximum value in the case of multiple paths). Source-pool pairs with a path comprehensive connection strength higher than 0.3 are identified as valid organ pairs with an assimilation product supply relationship.

[0046] For step S5, for each valid source-sink pair, the life trajectories of the source organ and the sink organ are extracted. Based on the life trajectory of the sink organ, a rapid development period time window is determined, and the cumulative value of the normalized differential vegetation index of the source organ within this window is calculated as a dynamic indicator; the integral of the volume change of the sink organ within this window is calculated as a dynamic trait parameter, quantifying the supply potential of the source organ and the utilization efficiency of the sink organ respectively.

[0047] For step S6, the sliding window method is used to analyze the life trajectory of the source organs. The starting time when the volume growth rate is greater than 0 is defined as T_start, and the time when the growth rate drops to 50% of the peak value is defined as T_end, thus determining the rapid development period. Sequences of source organ dynamic indicators within the time window from T_start minus 3 days to T_end plus 3 days are extracted to construct an ideal supply curve template with a single peak, whose peak time is aligned with the time point of the maximum growth rate of the source organs. The similarity between the source organ sequence and the ideal template is calculated using a dynamic time warping algorithm. Path constraints with a slope range of 0.5 to 2 are applied, and the minimum cumulative distance is transformed into dynamic process synergy efficiency through a monotonically decreasing mapping function. Finally, the mean of the synergy efficiency of all effective source-sink pairs of the same plant is used as a comprehensive evaluation index for screening superior materials in melon breeding.

[0048] Step S1 further includes the following sub-steps: S1-1, Obtain color time-series three-dimensional point cloud data of melon plants containing multispectral or hyperspectral reflectance information. The three-dimensional points include spatial coordinates and reflectance values ​​in non-visible light bands. S1-2, input the color temporal 3D point cloud data into a multimodal 3D instance segmentation network, process the spatial geometric features of the points through the first branch of the segmentation network, and process the spectral features of the points through the second branch of the segmentation network; S1-3 uses a cross-modal attention fusion module in the network to dynamically fuse geometric and spectral features, outputting the organ category semantic label and instance ID for each point. The semantic labels include leaf, fruit, and stem.

[0049] It should be noted that multispectral reflectance information refers to the reflection ratio of multiple specific bands of electromagnetic waves by melon plants, covering key bands such as red, green, blue, near-infrared, and red edge, which can reflect the physiological activity state of the plant.

[0050] Hyperspectral reflectance information refers to the proportion of electromagnetic waves reflected by melon plants in a continuous and narrow band. The number of bands usually exceeds 100, which can capture the differences in biochemical components and physiological state inside the plant more precisely.

[0051] Color time-series 3D point cloud data refers to a set of discrete points collected at continuous time nodes that contain color information and 3D spatial coordinates. It reflects both the spatial morphology of the plant and the changes in its surface optical characteristics over time.

[0052] Spatial coordinates refer to the x, y, and z axis position data in the preset coordinate system of a three-dimensional point cloud, which are used to locate the spatial position and morphological outline of melon plants and organs.

[0053] The non-visible light band refers to the electromagnetic wave band that is beyond the range of human vision, mainly including the near-infrared and red-edge bands. The reflectance data of this band can reflect the internal physiological activity of plants and make up for the limitations of visible light information.

[0054] Multimodal 3D instance segmentation network refers to an intelligent segmentation model that can process multiple types of input data. It can simultaneously receive both geometric and spectral data, and achieve integrated processing of organ category differentiation and individual separation.

[0055] The first branch refers to the functional module in the segmentation network that specializes in processing spatial geometric information. It adopts a network structure adapted to 3D point clouds and focuses on extracting morphology-related features.

[0056] Spatial geometric features refer to parameters extracted from three-dimensional point cloud data that characterize morphology and spatial relationships, including the curvature, normal vector, local density, and distance distribution of the point cloud, which are used to distinguish the morphological differences of different organs.

[0057] The second branch refers to the functional module in the segmentation network that specializes in processing spectral information. It adopts network structures such as one-dimensional convolution and focuses on extracting spectral features related to physiological activity.

[0058] Spectral features refer to parameters extracted from multispectral or hyperspectral reflectance data that characterize the physiological state of plants, including vegetation index-related features, peak and trough values ​​of band reflectance, etc., which are used to help distinguish organs with different functions.

[0059] The cross-modal attention fusion module is a core unit in the network used to integrate geometric and spectral features. It dynamically allocates weights by calculating the mutual information entropy of the two types of features, thereby enhancing the feature responses that are useful for organ segmentation.

[0060] Organ category semantic labels are classification markers used to identify the organ type to which a three-dimensional point belongs, clarifying that each point belongs to a certain category such as leaf, fruit, or stem, thus distinguishing organ categories.

[0061] An instance ID is an identification code used to uniquely identify different individual organs within the same organ category, ensuring that each independent organ in the same category has a unique identifier, thus providing a foundation for subsequent cross-time point tracking.

[0062] In one specific implementation, a field phenotyping robot equipped with a 16-line lidar and a hyperspectral camera is used to collect data. The lidar acquires the three-dimensional spatial coordinates of the melon plants, and the hyperspectral camera collects reflectance data in the 400 to 1000 nm wavelength band. Both are collected simultaneously to generate color time-series three-dimensional point cloud data. The data is collected once every 5 days, covering key nodes throughout the entire growth period of the melon.

[0063] The collected color time-series 3D point cloud data is input into a multimodal 3D instance segmentation network. The first branch of this network adopts the PointNet++ structure. The spatial coordinates of each 3D point are processed by a multilayer perceptron to extract spatial geometric features such as curvature, normal vector, and local density. Among them, the curvature feature is used to distinguish the cylindrical outline of the stem from the flat shape of the leaf, and the normal vector feature is used to identify the difference in the smoothness of the organ surface.

[0064] The second branch of the network adopts a one-dimensional convolutional network structure to process the hyperspectral reflectance data of each three-dimensional point. It extracts spectral features such as reflectance differences and peak positions between bands through convolutional kernels. The reflectance ratio feature between the red edge band and the near-infrared band is used to enhance the distinction between leaves and fruits.

[0065] The cross-modal attention fusion module dynamically weights the two types of features by calculating the mutual information entropy of geometric and spectral features. For leaf regions, the weight of spectral features is increased to highlight their physiological activity differences; for stem regions, the weight of geometric features is increased to enhance their morphological contour features. The fused features are input into a mask generation network, which outputs an organ category semantic label and instance ID for each 3D point. The leaf label corresponds to the green functional organ region, the fruit label corresponds to the enlarged reproductive organ region, and the stem label corresponds to the supporting structure region connecting the organs. The instance ID encodes each independent organ in a bottom-up, primary-to-secondary order to ensure the uniqueness of organ identity during subsequent tracking.

[0066] Step S2 further includes the following sub-steps: S2-1: For the organ instance segmentation results at each time point, construct its plant-level organ connection topology graph, where nodes are organ instances and edges represent the spatial connection relationships between organ instances. S2-2, obtain the spatial transformation matrix of the entire plant point cloud between adjacent time points obtained through time-series point cloud registration, and use the spatial transformation matrix to uniformly transform the three-dimensional spatial coordinates of each node in the topology map of the previous time point to the coordinate system of the next time point. S2-3 constructs the cross-time point instance matching problem into a graph matching optimization model and imposes hard constraints, requiring that the matching results must maintain the consistency of node connection relationships. The optimization objective of the graph matching optimization model is to maximize the sum of geometric and phenotypic similarities between corresponding nodes in the topological graph at adjacent time points. S2-4 uses integer programming or graph neural network algorithms to solve the graph matching optimization model, obtains the optimal instance matching pair, and links the IDs of the same organ instance at different time points based on the instance matching pair to form its life trajectory.

[0067] Furthermore, in sub-step S2-1, the establishment of edges is determined according to the following corresponding rules: If the two organ instances are a stem and a leaf, then when the minimum distance between the base point cloud region of the leaf and the surface point cloud of the stem is less than the first threshold, an edge is established between them. If the two organ instances are a stem and a fruit, then when the minimum distance between the point cloud of the fruit's stem region and the point cloud of the stem surface is less than the second threshold, an edge is established between them. If two organ instances are both fruits and belong to the same fruit branch, then when the minimum distance between their point clouds is less than the third threshold, an edge is established between them.

[0068] Furthermore, in sub-steps S2-3, the hard constraints imposed by the graph matching optimization model include topology consistency constraints; wherein, the topology consistency constraints are specifically configured as follows: When performing node matching between organ connectivity topologies at adjacent time points t and t+1, if node u in topology graph G_t and node v in topology graph G_t+1 are determined to be a match, then the following must be satisfied: for any adjacent node U of node u in topology graph G_t, after it is matched to the corresponding node V in topology graph G_t+1, node v and node V must also have an adjacency relationship in topology graph G_t+1. This symmetry constraint also applies in reverse to the matching process from topology graph G_t+1 to topology graph G_t. This constraint is not applicable to scenarios representing organ detachment or new organ growth.

[0069] It should be noted that the plant-level organ connection topology diagram refers to a graphical model constructed with all organ instances of a melon plant at a single point in time as the core elements and the actual spatial connection relationships between organs as the linking links. It is used to intuitively present the organ connection network of a plant at a single point in time.

[0070] A node is a core unit in a topology graph that represents a single organ instance. Each node uniquely corresponds to an organ instance and carries key information such as the organ's shape and location.

[0071] Spatial connectivity refers to the physical relationship between different organ instances in three-dimensional space. It is determined by a distance threshold and is the core basis for constructing a topology graph.

[0072] A spatial transformation matrix is ​​a mathematical matrix that describes the spatial positional relationship of plant point clouds at adjacent time points. It includes transformation parameters such as translation, rotation, and scaling, and is used to achieve accurate coordinate transformation under different coordinate systems.

[0073] Unified transformation to the coordinate system of the next time point refers to using a spatial transformation matrix to convert the three-dimensional coordinates of organ instances at the previous time point into a coordinate system consistent with the next time point, ensuring the comparability of organ positions across time points.

[0074] Cross-time point instance matching is a technical challenge that establishes a correspondence between organ instances at different time points. The core issue is to identify the identity of the same organ instance at different time points.

[0075] Graph matching optimization model refers to a mathematical model based on graph theory, used to solve the optimal correspondence between nodes in the topological graph at different time points, so as to achieve cross-time point matching of organ instances.

[0076] Hard constraints refer to the rules and conditions that must be strictly followed during the model solving process. They are used to limit the rationality of the matching results and avoid logically contradictory matching relationships.

[0077] Consistency of node connectivity means that after matching across time points, the adjacency relationship of organ instances remains unchanged. That is, the original adjacent organ instances must remain adjacent after matching to ensure that the matching result conforms to the actual logic of plant growth.

[0078] The optimization objective refers to the solution direction of the graph matching optimization model. By maximizing the sum of geometric and phenotypic similarities of corresponding nodes in the topological graph at adjacent time points, the goal is to achieve organ instance matching that best matches the actual growth state.

[0079] The sum of geometric and phenotypic similarity refers to the comprehensive consideration of the morphological parameter similarity and physiological characteristic similarity of organ instances. The sum of the two serves as a quantitative evaluation basis for node matching, thereby improving the accuracy of matching.

[0080] Integer programming is a mathematical optimization method that uses integer variables to describe matching relationships and solves for the optimal matching result under constraints. It is suitable for precise matching of small-scale topological graphs.

[0081] Graph neural network algorithms are intelligent algorithms that process graph-structured data based on neural network architectures. They can automatically learn node features and connection relationships, and are suitable for efficient matching of complex topological graphs.

[0082] The optimal instance matching pair refers to the cross-time point organ instance correspondence obtained through model solution that best meets the requirements of geometric and phenotypic similarity, ensuring the accurate association of instances of the same organ.

[0083] Linking the IDs of the same organ instance at different points in time means associating and binding the corresponding organ instance IDs in the optimal instance matching pair, so as to maintain the uniqueness of the same organ instance throughout its reproductive life.

[0084] The basal point cloud region of a leaf refers to the set of point clouds corresponding to the part of the leaf that is close to the stem and directly connected to the stem. It is a key region for determining the connection relationship between the leaf and the stem.

[0085] The first threshold is a critical distance value used to determine whether there is a connection between the stem and the leaves. It is set according to the growth morphology characteristics of melon leaves and stems to ensure the accuracy of the connection determination.

[0086] The point cloud of the fruit stem region refers to the set of point clouds corresponding to the part of the fruit that connects to the stem or fruit branch. It is a key area for determining the connection relationship between the fruit and the stem.

[0087] The second threshold is a critical distance value used to determine whether there is a connection between the stem and the fruit, and it is set according to the growth connection characteristics of the melon fruit and the stem.

[0088] A fruiting branch refers to a branch that extends from the main stem or lateral branches and carries multiple fruits. Fruits on the same fruiting branch have similar growth positions and nutrient supply channels.

[0089] The third threshold is a critical distance value used to determine whether there is a connection between fruits on the same fruit branch, and it is set according to the growth density characteristics of melon fruits.

[0090] Topological consistency constraints are constraints that ensure that the adjacency relationships of organ instances remain unchanged after the topological graph is matched across time points. They are the core constraints for maintaining the logical rationality of the matching results.

[0091] An adjacent node is a node in the topology graph that has a direct connection with the target node, corresponding to an organ instance that is adjacent to the target organ in actual growth.

[0092] Symmetry constraints refer to the characteristic that topological consistency constraints are applicable in both directions between topological graphs at adjacent time points. They are applicable to matching from one time point to the next, as well as to matching in the opposite direction.

[0093] Organ detachment refers to the phenomenon where an organ separates from the plant in the later stages of its growth due to physiological aging or environmental factors. At this time, the corresponding node will disappear from the topology graph.

[0094] New organ growth refers to the addition of new organ instances during plant growth. At this time, new nodes will appear in the topology graph, which do not need to follow the connection constraints of the original nodes.

[0095] In one specific implementation, for step S2-1, for the organ instance segmentation results at each time point, using each organ instance as a node, edges are established according to preset rules to construct a plant-level organ connection topology graph. A first threshold is set to 6 mm; when the minimum distance between the base point cloud region of a leaf and the point cloud region of the stem surface is less than this value, an edge is established between them. A second threshold is set to 4 mm; when the minimum distance between the point cloud region of a fruit stem and the point cloud region of the stem surface is less than this value, an edge is established between them. A third threshold is set to 12 mm; when the minimum distance between the point clouds of two fruits on the same fruit branch is less than this value, an edge is established between them, forming a complete topology graph structure.

[0096] For step S2-2, the iterative nearest-point algorithm is used to register the plant point clouds at adjacent time points, and the spatial transformation matrix is ​​calculated. Taking time points t and t+1 as examples, the spatial transformation matrix is ​​used to convert the three-dimensional spatial coordinates of all nodes in the topology map at time point t into the corresponding coordinates in the coordinate system of time point t+1, thereby unifying the node coordinates of the topology maps at the two time points.

[0097] For steps S2-3, a graph matching optimization model is constructed, with the optimization objective being to maximize the sum of the geometric similarity and phenotypic similarity of corresponding nodes in the topological graph at adjacent time points. A topological consistency constraint is applied: if node u at time point t matches node v at time point t+1, then after node u's adjacent node U at time point t matches node V at time point t+1, nodes v and V must remain adjacent. This constraint also applies in reverse. When a node is detected to disappear or be added, the constraint is automatically released.

[0098] For steps S2-4, a graph neural network algorithm is used to solve the graph matching optimization model. The node features and connectivity relationships of the topological graphs at two time points are input into the graph neural network. Through multi-layer iterative learning, the matching probability matrix between nodes is output. Based on the probability matrix, the optimal instance matching pairs are selected. The corresponding organ instance IDs in the matching pairs are associated. For example, if a leaf instance with ID L1 at time point t matches a leaf instance with ID L12 at time point t+1, the two IDs are bound together. This is then sequentially associated with the corresponding IDs at all time points throughout the entire growth period, forming the life trajectory of that leaf instance.

[0099] Step S3 further includes the following sub-steps: S3-1, For any two organ instances, calculate the contact area between their point clouds. The contact area is composed of point pairs in the two point clouds whose distance is less than a preset threshold. S3-2, Extract the geometric features of the contact area, including the consistency of the normal vectors of the point cloud of the contact area, and the continuity of the surface area and curvature of the contact area; S3-3, based on geometric features, uses a pre-trained classifier or a preset rule set to determine whether there is a physical connection between two organ instances and to estimate the connection strength. S3-4: Based on the judgment results of all time points, filter out the connection relationships that are judged to exist in time points exceeding a preset proportion, and use the existing connection relationships as edges to construct a three-dimensional topological connection graph together with the corresponding organ instance nodes.

[0100] It should be noted that the contact area refers to the overlapping and associated region between the point clouds of two organ instances that meets the distance condition. It consists of point pairs that are close to each other and the distance is less than a preset threshold. It is the core basis for judging the physical connection between organs.

[0101] A point pair is a combination of two points that come from the point clouds of two different organ instances and satisfy a distance condition. Each point pair corresponds to an adjacent position on the surface of the two organs.

[0102] The preset threshold is a critical distance value used to determine whether two points are in contact. It is set based on the collection density and growth characteristics of melon organ point clouds to ensure the accuracy of contact area determination.

[0103] Normal vector consistency refers to the degree of matching of the normal vector directions of corresponding points of two organ instances within the contact area. It is quantified by calculating the mean of the angle between the normal vectors, reflecting the degree of fit of the contact surfaces.

[0104] The surface area of ​​the contact region refers to the surface area formed by the set of contact point pairs in three-dimensional space. It is calculated using point cloud reconstruction technology and represents the size of the contact range between two organs.

[0105] Curvature continuity refers to the smoothness of curvature changes at points within the contact area. It is quantified by calculating the variance of curvature changes, reflecting the smooth connection state of the contact surfaces.

[0106] A pre-trained classifier is an intelligent discrimination model trained on point cloud data of melon organs labeled with connectivity relationships. It can automatically determine physical connectivity relationships based on the input geometric features.

[0107] The preset rule set refers to a series of logical rules formulated based on domain knowledge. By comparing geometric features with rule thresholds, the physical connection relationship can be quickly determined.

[0108] Physical connection refers to the stable association between two organ instances in three-dimensional space through direct contact, and is the structural basis for the functional cooperation of organs within a plant.

[0109] Connection strength is a quantitative indicator of the physical connection between two organ instances. It is calculated based on geometric features and ranges from 0 to 1. A higher value indicates a tighter connection.

[0110] The determination results at all time points refer to the independent determination conclusions of the connection relationship between two organ instances at each collection time point throughout the entire reproductive period, constituting a temporal change record of the connection relationship.

[0111] The preset ratio refers to the critical value of the time percentage used to screen stable connections. Only connections that persist for more than this ratio are considered stable connections.

[0112] Stable connectivity refers to the physical connection between organs that persists for most of the reproductive lifespan. It reflects the long-term structural relationship between organs and is distinct from the accidental relationship of temporary contact.

[0113] In one specific implementation, for step S3-1, all organ instance combinations of the melon plant at the same time point are traversed, and the distance between the point cloud data of any two organ instances is calculated. A preset threshold of 2.5 mm is set. For each point of organ A, points in organ B with a distance less than the threshold are found to form contact point pairs. All contact point pairs together constitute the contact area between the two organ instances.

[0114] For step S3-2, geometric features are extracted from the obtained contact area. The normal vector of each contact point pair is calculated using a point cloud processing algorithm, and the mean of the angle between the normal vectors of all point pairs is obtained to determine the consistency of the normal vectors. The contact point pairs are fitted to a surface using Poisson surface reconstruction technology, and the area of ​​this surface is calculated as the surface area of ​​the contact area. The curvature values ​​of all points within the contact area are calculated, and the curvature continuity is obtained through variance analysis.

[0115] For step S3-3, a pre-trained random forest classifier is used to determine connectivity. This classifier takes normal vector consistency, surface area of ​​the contact region, and curvature continuity as input features, and outputs a determination result of "connection exists" or "connection does not exist". Simultaneously, the three geometric features are normalized and then weighted and summed to obtain a value between 0 and 1, which serves as the quantification result of the connectivity strength. The weight coefficients are determined through training based on the importance of the features to the connectivity determination.

[0116] For steps S3-4, the connection relationship determination results of each organ instance combination throughout the entire growth period are statistically analyzed, and a preset proportion of 70% is set. Combinations that are determined to "have connections" in more than 70% of the time points are selected. These stable connections are used as edges, with the edge weight being the average value of the connection strength. Using all organ instances as nodes, a three-dimensional topological connection graph of the melon plant is constructed. This graph can completely present the stable connection network between organs throughout the entire growth period.

[0117] Step S4 further includes the following sub-steps: S4-1, In the three-dimensional topological connection graph, identify candidate source organ nodes and candidate library organ nodes. Candidate library organ nodes are determined from fruit instances based on organ type and developmental stage represented in life trajectory, while candidate source organ nodes are determined from leaf instances based on organ type. S4-2, On the three-dimensional topological connection graph, for each candidate library organ node, perform a directed path traversal to find all candidate source organ nodes that can be reached through the connection edges, forming an initial source-liquid pair set. The direction of the directed path traversal is preset according to the material transport direction from the source organ to the library organ. S4-3, For each source-library pair in the initial source-library pair set, calculate the comprehensive path connection strength from the source organ node to the library organ node based on the connection strength of each edge on the connection path. S4-4 identifies source-liquid pairs with a path integration connection strength higher than a preset strength threshold as organ pairs with an assimilation product supply relationship.

[0118] It should be noted that candidate source organ nodes refer to nodes selected from the three-dimensional topological connectivity graph that have the potential for assimilation product synthesis and supply. They are selected only from leaf instances and conform to the functional attributes of source organs.

[0119] Candidate organ nodes refer to nodes selected from the three-dimensional topological connectivity graph that have the ability to consume and store assimilates. They are selected from fruit instances and must meet the developmental stage requirements represented in the life trajectory.

[0120] The developmental stage of representation refers to the characteristic information extracted from the life trajectory of an organ that reflects the growth status of the organ, and is used to determine whether a fruit is in a rapid developmental stage that requires assimilation products.

[0121] Directed path traversal refers to searching for connected paths from candidate source organ nodes to candidate library organ nodes in a 3D topological connection graph in a preset single direction, without allowing reverse traversal.

[0122] The direction of material transport refers to the physiological pathway of assimilated products in melon plants from source organs to sink organs, presumably a unidirectional direction from leaf nodes through stem nodes to fruit nodes.

[0123] The initial source-liquid pair set refers to the combination of organ nodes that may have assimilation product supply relationships, obtained by traversing a directed path. It is the basis for subsequent screening of effective organ pairs.

[0124] A connection path refers to a series of continuous edges connecting candidate source organ nodes and candidate library organ nodes in a three-dimensional topological connection graph, corresponding to the transport channels of assimilated products in actual plants.

[0125] The overall path connectivity strength is a quantitative indicator obtained by comprehensively calculating the connectivity strength of all edges on the source-database connection path, reflecting the smoothness and stability of the path.

[0126] The preset intensity threshold is a critical value used to determine whether there is an effective assimilation product supply relationship between the source and the library. It is set according to the physiological characteristics of melon plants and the needs of breeding practice.

[0127] Organ pairs with assimilation product supply relationships refer to source-sink node combinations that have been identified as having stable assimilation product transport channels after screening based on the comprehensive connectivity strength of the pathway. These pairs are the core objects of subsequent collaborative efficiency analysis.

[0128] In one specific implementation, for step S4-1, candidate source organ nodes and candidate library organ nodes are screened in the three-dimensional topology connection graph. Candidate source organ nodes are directly selected from the nodes corresponding to all leaf instances, assuming that all functional leaves have the potential to supply assimilates; candidate library organ nodes are selected from the nodes corresponding to fruit instances. By analyzing the life trajectory of the fruit, fruit nodes with a volume growth rate greater than 0.5 cubic centimeters per day are screened and determined to be in a rapid development stage and possess library organ functions.

[0129] For step S4-2, the preset directed path traversal direction is a unidirectional direction from leaf nodes to stem nodes, and then from stem nodes to fruit nodes. Taking each candidate library organ node as the endpoint, a depth-first traversal is performed in the 3D topological connection graph along the preset direction to search for all candidate source organ nodes that can be reached through continuous connecting edges. Each library organ node is combined with all corresponding source organ nodes to form an initial set of source-library pairs.

[0130] For step S4-3, for each pair of nodes in the initial source-library pair set, extract all edges on its connection path. Multiply the connection strength of each edge (if there are multiple parallel paths, take the maximum value of the product results of each path) to obtain the comprehensive path connection strength of the source-library pair. The closer the product result is to 1, the better the path smoothness and stability.

[0131] For step S4-4, a preset strength threshold is set to 0.3. The path integration connection strength of each source-liquid pair is compared with this threshold. If the strength value is higher than the threshold, the source-liquid pair is determined to have a stable assimilation product transport channel and is identified as an organ pair with an assimilation product supply relationship. If the strength value is lower than the threshold, it is determined to have no effective supply relationship and is removed from the set.

[0132] Step S5 further includes the following sub-steps: S5-1, for each source organ and depot organ pair, extract the life trajectory of the source organ and the depot organ respectively; S5-2, based on the life trajectory of the source organ, calculates the cumulative or average value of its physiological activity indicators within the time window that overlaps with the rapid development period of the deposit organ, as a dynamic indicator characterizing the potential contribution of the source organ to the supply of the deposit organ. S5-3, based on the life trajectory of the reservoir organ, calculates the peak value of its morphological change integral or growth acceleration during the rapid development period, as a dynamic trait parameter characterizing the assimilate utilization efficiency of the reservoir organ.

[0133] It should be noted that the source organ and sink organ pair refers to the corresponding combination of source organ nodes and sink organ nodes that have an assimilation product supply relationship, as determined by step S4. It is the core object for extracting dynamic indicators and dynamic trait parameters.

[0134] Extracting the life trajectory of source and bank organs refers to separating the independent life trajectory data of the corresponding organ instance in each source-bank pair from the time-series data of the entire reproductive period, ensuring the relevance of indicator calculation.

[0135] The rapid development period of sink organs refers to the stage in which the sink organ (fruit) grows at its fastest rate and has the strongest demand for assimilates. It is determined by the rate of change of morphological parameters in the life trajectory.

[0136] The overlapping time window refers to the interval in the life trajectory of the source organ that coincides with the rapid development period of the sink organ in time. The physiological activity of the source organ in this interval directly affects the development of the sink organ and is a key evaluation period for the potential contribution of the supply.

[0137] Physiological activity indicators are parameters extracted from the spectral characteristics of source organs that reflect the ability to synthesize assimilates, including normalized difference vegetation index and red edge reflectance, which can quantify the physiological functional status of source organs.

[0138] The cumulative value of physiological activity indicators refers to the sum of physiological activity indicators over time within an overlapping time window, reflecting the total supply capacity of the source organ during critical periods.

[0139] The mean value of physiological activity indicators refers to the average value of physiological activity indicators within an overlapping time window, reflecting the stable supply capacity of the source organ during critical periods.

[0140] Supply contribution potential refers to the potential ability of a source organ to deliver assimilates to a corresponding depot organ. It is quantified by dynamic indicators, and the higher the value, the stronger the supply capacity.

[0141] The morphological change integral refers to the integral value of morphological parameters (such as volume and surface area) over time during the rapid development period of the reservoir organ, reflecting the total growth of the reservoir organ during the critical period.

[0142] The peak growth acceleration refers to the maximum rate of change of growth rate of the sink organ during the rapid development period, reflecting the maximum utilization rate of assimilates by the sink organ.

[0143] Assimilation utilization efficiency refers to the ability of a sink organ to convert the assimilated products it receives into its own growth. It is quantified by dynamic trait parameters, and the higher the value, the better the utilization efficiency.

[0144] In one specific implementation, for step S5-1, from the full life cycle time series database, according to the organ instance ID of each source-sink pair, the life trajectory data of the corresponding source organ (leaf) and sink organ (fruit) are extracted respectively. Each trajectory contains spectral data, morphological parameters and spatial coordinate information at each time point.

[0145] For step S5-2, by analyzing the rate of volume change in the life trajectory of the source organ, its rapid development period is determined to be 10 to 25 days after fruit set. Using this interval as a benchmark, a time window overlapping with the source organ's life trajectory is determined. The normalized difference vegetation index (NDVI) within the overlapping time window is extracted from the source organ's life trajectory as a physiological activity indicator. The cumulative value and mean of this indicator are calculated, and the maximum value is taken as the dynamic indicator to quantify the source organ's supply contribution potential.

[0146] For step S5-3, based on the volume data of the rapid development period (10 to 25 days after fruit set) in the life trajectory of the sink organ, the trapezoidal integral method is used to calculate the morphological change integral, reflecting the total growth. Simultaneously, by calculating the difference in volume growth rate between adjacent time points, a growth acceleration sequence is obtained, and the maximum value in the sequence is extracted as the peak value of the growth acceleration. After normalizing the morphological change integral and the peak value of the growth acceleration, a weighted sum is taken as a dynamic trait parameter to quantify the assimilate utilization efficiency of the sink organ. The weights are set according to the degree of influence of both on the utilization efficiency.

[0147] Step S6 further includes the following sub-steps: S6-1, based on the life trajectory of the organ, determine the start time T_start and end time T_end of its rapid development period; S6-2, Based on the dynamic index time series of the source organ, calculate its activity intensity sequence within the time window [T_start-Δ, T_end+Δ], where Δ is the preset advance and lag tolerance; S6-3, based on prior knowledge of plant physiology, predefines an ideal supply curve template. This curve has a single peak shape in [T_start, T_end], and its peak time is aligned with the time point of the maximum growth rate of the sink organ. S6-4, the activity intensity sequence of the source organ is temporally aligned and normalized and compared with the ideal supply curve template to calculate the shape similarity between the two. S6-5 uses the calculated shape similarity as a quantification of the collaborative efficiency of the dynamic process.

[0148] Furthermore, in sub-step S6-4, the shape similarity between the two is calculated using a dynamic time warping algorithm, including the following steps: Discretize the ideal supply curve template within the time interval [T_start, T_end] to generate a discrete ideal supply sequence; The source organ activity intensity sequence and the discrete ideal supply sequence are time-axis normalized to make them have the same sequence length; Define a local distance function to calculate the Euclidean distance or absolute difference between two sequences at any two corresponding time points; During dynamic time warping, a path slope constraint is applied to limit the continuity and monotonicity of the warped path. Calculate the minimum cumulative distance that satisfies the constraints, and use it as the minimum regularized path distance; The shape similarity is obtained by inputting the minimum regular path distance into a monotonically decreasing mapping function.

[0149] It should be noted that the start time T_start refers to the beginning of the rapid development period of the organ, which is determined by analyzing the time point when the growth rate of morphological parameters in the life trajectory of the organ first exceeds the preset growth threshold.

[0150] The end time T_end refers to the point at which the rapid development period of the reservoir organ ends. It is determined by analyzing the time point when the growth rate of morphological parameters in the life trajectory of the reservoir organ drops to 50% of its peak value.

[0151] Dynamic index time series refers to the numerical sequence of dynamic indexes of source organs changing over time, which fully records the temporal change characteristics of the source organ's supply contribution potential.

[0152] The time window [T_start-Δ, T_end+Δ] refers to the interval that covers the rapid development period of the source organ and extends forward and backward for a certain period of time, and is used to fully capture the synergistic relationship between the activity of the source organ and the development of the source organ.

[0153] The advance and lag tolerance Δ refers to the time extension set to accommodate the time difference in assimilation product transport, ensuring that the lag or advance effect of source organ activity on sink organ development is not overlooked.

[0154] The activity intensity sequence refers to the numerical sequence within the corresponding time window extracted from the time series of dynamic indicators of source organs, which is specifically used for similarity comparison with the ideal supply curve template.

[0155] Prior knowledge in plant physiology refers to existing research conclusions based on the synergistic relationship between plant assimilate supply and sink organ development, which provides a theoretical basis for the design of ideal supply curve templates.

[0156] An ideal supply curve template refers to a curve that simulates the optimal supply mode of source organs and can reflect the ideal synergistic relationship between the supply of assimilates and the development of sink organs.

[0157] The unimodal morphology refers to the ideal supply curve template showing a trend of first rising and then falling during the rapid development of the sink organ, which is consistent with the physiological law that the source organ gradually increases its supply to meet the needs of the sink organ and then gradually weakens in the later stage.

[0158] The maximum growth rate time point refers to the moment when the growth rate of the sink organ reaches its peak during the rapid development period, and it is the alignment benchmark for the peak of the ideal supply curve template.

[0159] Time alignment refers to calibrating the time axis of the source organ activity intensity sequence with that of the ideal supply curve template to ensure that the correspondence between the two at key time points is accurate.

[0160] Normalization comparison refers to converting the values ​​of two sequences to the same numerical range to eliminate the influence of dimensional differences on similarity calculation and ensure the fairness of the comparison.

[0161] Shape similarity refers to the degree of agreement between the source organ activity intensity sequence and the ideal supply curve template in terms of change trend, peak position, and curve shape. It is the core quantitative basis for synergistic efficiency.

[0162] Discretization sampling refers to extracting numerical points from a continuous ideal supply curve template at fixed time intervals and converting them into a discrete sequence to meet the input requirements of dynamic time warping algorithms.

[0163] Discrete ideal supply sequence refers to the set of values ​​obtained after discretization sampling, arranged in chronological order, retaining the core morphological characteristics of the ideal supply curve template.

[0164] Time axis normalization refers to adjusting the length of two sequences through interpolation or sampling so that they contain the same number of numerical points, ensuring that distance calculations can be performed point by point during dynamic time warping.

[0165] Local distance functions are mathematical functions used to calculate the numerical differences between corresponding time points of two sequences. Euclidean distance or absolute difference is a commonly used quantification method that reflects the degree of matching at a single point.

[0166] Path slope constraint refers to the restriction condition on the slope of path change during dynamic time warping, which is used to avoid unreasonable time matching and ensure that the physical meaning of the warped path conforms to physiological laws.

[0167] The continuity of a regular path refers to the smooth transition of the regular path on the time axis, which does not allow for skipped matches and ensures the rationality of the time correspondence.

[0168] The monotonicity of regular paths means that regular paths extend only in the direction of increasing time and do not allow time backtracking matches, which is consistent with the irreversible nature of time series data.

[0169] The minimum cumulative distance refers to the minimum sum of local distances between all corresponding points in two sequences, under the premise of satisfying constraints, reflecting the degree of difference between the two sequences as a whole.

[0170] Minimum regular path distance refers to the quantification of the minimum cumulative distance, which is a core indicator for measuring the dissimilarity between two sequences. The smaller the value, the more similar the sequences are.

[0171] A monotonically decreasing mapping function is a mathematical function that converts the minimum regular path distance into shape similarity. This function satisfies the rule that the greater the distance, the smaller the similarity, and the smaller the distance, the greater the similarity, ensuring that the similarity is positively correlated with the degree of matching.

[0172] In one specific implementation, for step S6-1, the volume change data in the life trajectory of the organ (fruit) is analyzed, and the volume growth rate at adjacent time points is calculated. A growth rate threshold of 0.3 cubic centimeters per day is set. When the growth rate first exceeds this threshold, this moment is recorded as the start time T_start; when the growth rate drops to 50% of its peak value, this moment is recorded as the end time T_end, thus determining the rapid development period interval.

[0173] For step S6-2, the advance and lag tolerance Δ is set to 3 days, and a time window [T_start-3, T_end+3] is constructed. From the dynamic index time series of the source organ, all values ​​within this time window are extracted to form an activity intensity sequence, which fully covers the cycle of the influence of source organ activity on the development of deposit organ.

[0174] For step S6-3, based on the principle in plant physiology that "the supply of source organs must be synchronized with the peak demand of sink organs," an ideal supply curve template is predefined. This curve has a single-peak shape within [T_start, T_end], with the peak height being the maximum value of the activity intensity sequence. The peak time is aligned with the peak time of the sink organ volume growth rate. The curve smoothly rises from the baseline to the peak and then smoothly declines at both ends, simulating the optimal supply mode.

[0175] For step S6-4, the shape similarity is calculated using a dynamic time warping algorithm. First, the ideal supply curve template is discretized within [T_start, T_end] at 1-day intervals to generate a discrete ideal supply sequence. The source organ activity intensity sequence and the ideal supply sequence are time-axis normalized and adjusted to sequences of 20 data points each through linear interpolation. The local distance function is defined as Euclidean distance, and the numerical difference between corresponding points is calculated. A path slope constraint of 0.5 to 2 is applied to limit the variation range of the warped path. The minimum cumulative distance satisfying the constraint is solved using a dynamic programming algorithm to obtain the minimum warped path distance. This distance is input into the mapping function Sim=1-D / D_max (where D is the minimum warped path distance and D_max is the preset maximum possible distance) to obtain the shape similarity Sim, with a value ranging from 0 to 1.

[0176] For step S6-5, the calculated shape similarity Sim is directly used as the dynamic process synergistic efficiency of the source-library pair. For the same melon plant, the average synergistic efficiency of all effective source-library pairs is calculated as a comprehensive evaluation index of the plant's organ interaction effectiveness, which is used for screening superior materials during the breeding process.

[0177] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for extracting and detecting traits used in melon breeding, for digitally analyzing and evaluating the organ interaction efficiency of melon plants throughout their entire growth period, characterized in that... Includes the following steps: Step S1: Based on the time-series three-dimensional point cloud data continuously collected during the entire growth period of the melon plant, perform three-dimensional instance segmentation on the plant point cloud at each time point, and identify and separate the independent organ structures of the melon. Step S2: Through temporal point cloud registration and data association based on organ spatial continuity, cross-time point tracking of the same organ instance is achieved, forming independent life trajectories for each organ. Step S3: Based on temporal 3D point cloud data and life trajectory, by analyzing the spatial adjacency and geometric continuity relationships between point clouds of different organ instances, a 3D topological connection map representing the connection relationship between organs is inferred and generated. Step S4: Based on the three-dimensional topological connectivity graph, identify source organ and sink organ pairs that have assimilation product supply relationships; Step S5: For each source organ and depot organ pair, extract dynamic indicators representing the physiological activity of the source organ and dynamic trait parameters representing the morphological development of the depot organ from the time-series data corresponding to its life trajectory. Step S6: By analyzing the synergistic relationship between the dynamic indicators of the source organ and the dynamic trait parameters of the deposit organ over time, the synergistic efficiency of the dynamic process between organ pairs is quantified and used as a breeding evaluation indicator.

2. The method for extracting and detecting traits for melon breeding according to claim 1, characterized in that: Step S1 further includes the following sub-steps: S1-1, Obtain color time-series three-dimensional point cloud data of melon plants containing multispectral or hyperspectral reflectance information, wherein the three-dimensional points include spatial coordinates and reflectance values ​​in non-visible light bands; S1-2, input the color temporal 3D point cloud data into a multimodal 3D instance segmentation network, process the spatial geometric features of the points through the first branch of the segmentation network, and process the spectral features of the points through the second branch of the segmentation network; S1-3, through the cross-modal attention fusion module in the network, geometric features and spectral features are dynamically fused to output the organ category semantic label and instance ID of each point. The semantic label includes leaf, fruit and stem.

3. The method for extracting and detecting traits for melon breeding according to claim 1, characterized in that: Step S2 further includes the following sub-steps: S2-1: For the organ instance segmentation results at each time point, construct its plant-level organ connection topology graph, where nodes are organ instances and edges represent the spatial connection relationships between organ instances. S2-2, obtain the spatial transformation matrix of the entire plant point cloud between adjacent time points obtained through time-series point cloud registration, and use the spatial transformation matrix to uniformly transform the three-dimensional spatial coordinates of each node in the topology map of the previous time point to the coordinate system of the next time point. S2-3, the instance matching problem across time points is constructed as a graph matching optimization model, and a hard constraint is imposed, requiring that the matching results must maintain the consistency of the node connection relationship. The optimization objective of the graph matching optimization model is to maximize the sum of the geometric and phenotypic similarities of corresponding nodes between adjacent time point topological graphs. S2-4 uses integer programming or graph neural network algorithms to solve the graph matching optimization model, obtains the optimal instance matching pair, and links the IDs of the same organ instance at different time points based on the instance matching pair to form its life trajectory.

4. The method for extracting and detecting traits for melon breeding according to claim 3, characterized in that, In sub-step S2-1, the establishment of the edge is determined according to the following corresponding rules: If the two organ instances are a stem and a leaf, then when the minimum distance between the base point cloud region of the leaf and the surface point cloud of the stem is less than the first threshold, an edge is established between them. If the two organ instances are a stem and a fruit, then when the minimum distance between the point cloud of the fruit's stem region and the point cloud of the stem surface is less than the second threshold, an edge is established between them. If two organ instances are both fruits and belong to the same fruit branch, then when the minimum distance between their point clouds is less than the third threshold, an edge is established between them.

5. The method for extracting and detecting traits for melon breeding according to claim 3, characterized in that, In sub-steps S2-3, the hard constraints imposed by the graph matching optimization model include topology consistency constraints; wherein, the topology consistency constraints are specifically configured as follows: When performing node matching between organ connectivity topologies at adjacent time points t and t+1, if node u in topology graph G_t and node v in topology graph G_t+1 are determined to be a match, then the following must be satisfied: for any adjacent node U of node u in topology graph G_t, after it is matched to the corresponding node V in topology graph G_t+1, node v and node V must also have an adjacency relationship in topology graph G_t+1. This symmetry constraint also applies in reverse to the matching process from topology graph G_t+1 to topology graph G_t. This constraint is not applicable to scenarios representing organ detachment or new organ growth.

6. The method for extracting and detecting traits for melon breeding according to claim 1, characterized in that: Step S3 further includes the following sub-steps: S3-1, For any two organ instances, calculate the contact area between their point clouds, where the contact area is composed of point pairs in the two point clouds whose distance is less than a preset threshold. S3-2, Extract the geometric features of the contact area, including the consistency of the normal vectors of the point cloud of the contact area and the continuity of the surface area and curvature of the contact area; S3-3, based on geometric features, uses a pre-trained classifier or a preset rule set to determine whether there is a physical connection between two organ instances and to estimate the connection strength. S3-4: Based on the judgment results of all time points, filter out the connection relationships that are judged to exist in time points exceeding a preset proportion, and use the existing connection relationships as edges to construct a three-dimensional topological connection graph together with the corresponding organ instance nodes.

7. The method for extracting and detecting traits for melon breeding according to claim 1, characterized in that: Step S4 further includes the following sub-steps: S4-1, In the three-dimensional topological connection diagram, candidate source organ nodes and candidate library organ nodes are identified. The candidate library organ nodes are determined from fruit instances based on organ type and developmental stage represented in life trajectory. The candidate source organ nodes are determined from leaf instances based on organ type. S4-2, On the three-dimensional topological connection graph, for each candidate library organ node, a directed path traversal is performed to find all candidate source organ nodes that can be reached through the connection edge, forming an initial source-liquid pair set. The direction of the directed path traversal is preset according to the material transport direction from the source organ to the library organ. S4-3, For each source-library pair in the initial source-library pair set, calculate the comprehensive path connection strength from the source organ node to the library organ node based on the connection strength of each edge on the connection path. S4-4 identifies source-liquid pairs with a path integration connection strength higher than a preset strength threshold as organ pairs with an assimilation product supply relationship.

8. The method for extracting and detecting traits for melon breeding according to claim 1, characterized in that: Step S5 further includes the following sub-steps: S5-1, for each source organ and depot organ pair, extract the life trajectory of the source organ and the depot organ respectively; S5-2, based on the life trajectory of the source organ, calculates the cumulative or average value of its physiological activity indicators within the time window that overlaps with the rapid development period of the deposit organ, as a dynamic indicator characterizing the potential contribution of the source organ to the supply of the deposit organ. S5-3, based on the life trajectory of the reservoir organ, calculates the peak value of its morphological change integral or growth acceleration during the rapid development period, as a dynamic trait parameter characterizing the assimilate utilization efficiency of the reservoir organ.

9. The method for extracting and detecting traits for melon breeding according to claim 1, characterized in that: Step S6 further includes the following sub-steps: S6-1, based on the life trajectory of the organ, determine the start time T_start and end time T_end of its rapid development period; S6-2, Based on the dynamic index time series of the source organ, calculate its activity intensity sequence within the time window [T_start-Δ, T_end+Δ], where Δ is the preset advance and lag tolerance; S6-3, based on prior knowledge of plant physiology, predefines an ideal supply curve template. This curve has a single peak shape in [T_start, T_end], and its peak time is aligned with the time point of the maximum growth rate of the sink organ. S6-4, the activity intensity sequence of the source organ is temporally aligned and normalized and compared with the ideal supply curve template to calculate the shape similarity between the two. S6-5 uses the calculated shape similarity as a quantification of the collaborative efficiency of the dynamic process.

10. The method for extracting and detecting traits for melon breeding according to claim 9, characterized in that, In sub-step S6-4, the calculation of the shape similarity between the two is achieved through a dynamic time warping algorithm, including the following steps: Discretize the ideal supply curve template within the time interval [T_start, T_end] to generate a discrete ideal supply sequence; The source organ activity intensity sequence and the discrete ideal supply sequence are time-axis normalized to make them have the same sequence length; Define a local distance function to calculate the Euclidean distance or absolute difference between two sequences at any two corresponding time points; During dynamic time warping, a path slope constraint is applied to limit the continuity and monotonicity of the warped path. Calculate the minimum cumulative distance that satisfies the constraints, and use it as the minimum regularized path distance; The shape similarity is obtained by inputting the minimum regular path distance into a monotonically decreasing mapping function.

Images