Fruit and vegetable new variety cultivation management supervision method and system

Abstract

The invention relates to the technical field of agricultural management, in particular to a fruit and vegetable new variety cultivation management supervision method and system.The method comprises the following steps that fruit and vegetable full-cycle data is collected to construct an environment and phenotype time sequence monitoring set, sequence transfer entropy is calculated to generate a dynamic character causal map, hidden variable nodes are set to construct a metabolism hidden variable supervision network structure; the probability is corrected through an expectation maximization algorithm, a growth state transition probability parameter table is determined, seedling stage data is input, the mature stage posterior probability is calculated, germplasm classification is determined, environment compensation parameters are inverted, and a cultivation management supervision instruction is generated. According to the method, the dynamic causal map is constructed by calculating the environment phenotype time sequence transfer entropy, the potential metabolic node reconstruction character dependency path is set, the growth state transition probability is corrected by using the expectation maximization algorithm, the mature period joint posterior probability is predicted, and the environment compensation parameter is inversed, so that the early germplasm grading and accurate intervention of crops are realized; the breeding decision lag problem is solved; and the breeding management efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural management technology, and in particular to methods and systems for supervising the cultivation, management and supervision of new fruit and vegetable varieties. Background Technology

[0002] The field of agricultural management technology encompasses a comprehensive management system covering the entire process from crop planting planning, resource allocation, growth monitoring to final harvest. Its core lies in the scientific and coordinated arrangement of factors such as land, water, seeds, fertilizers, and labor in agricultural production, aiming to establish an efficient and sustainable agricultural production model. This field not only focuses on the regulation of crop growth cycles but also involves the systematic coordination and control of multiple aspects such as pest and disease control, yield prediction, and agricultural product quality traceability.

[0003] The traditional method for supervising and managing the breeding of new fruit and vegetable varieties refers to the monitoring and standardization of matters such as hybridization combinations, offspring screening, trait recording, and experimental field management during the breeding process of new varieties. It relies on breeders manually entering data on parent sources, hybridization time, and offspring traits through paper files or spreadsheet software. In the field, they use basic measuring tools such as tape measures, vernier calipers, and saccharimeters to obtain plant height, fruit size, and sugar-acid ratio indicators. They also observe and record crop growth status and the occurrence of pests and diseases through regular manual inspections. Subsequently, based on expert experience, they compare the results with the established breeding objectives to make phenotypic screening and elimination decisions. Finally, they archive and save the pedigree information of the remaining germplasm resources in the form of text files.

[0004] Existing technologies rely on manual methods combined with basic tools for discrete data recording. Limited by low-frequency measurements and subjective experience, they cannot capture the nonlinear dynamic relationship between the environment and phenotype throughout the entire growth cycle. Screening based solely on external traits is insufficient to quantify and reveal the potential regulatory mechanisms of internal metabolic allocation on fruit yield and quality. Furthermore, the lack of in-depth exploration of the causal evolutionary relationship between traits at the seedling and maturity stages results in breeding decisions lagging significantly behind crop growth status. This not only prolongs the screening cycle for superior germplasm and wastes resources, but also makes it difficult for environmental control measures to maintain optimal effectiveness due to the lack of precise metabolic data. Summary of the Invention

[0005] To address the limitations of existing technologies that rely on manual methods combined with basic tools for discrete data recording, which are restricted by low-frequency measurements and subjective experience, resulting in the inability to capture the nonlinear dynamic relationship between environment and phenotype throughout the entire growth cycle, and the difficulty in quantifying and revealing the potential regulatory mechanisms of internal metabolic allocation on fruit yield and quality based solely on external trait screening, and the lack of in-depth exploration of the causal evolutionary relationship between traits at the seedling and maturity stages, resulting in breeding decisions lagging significantly behind crop growth status, this invention not only prolongs the selection cycle for superior germplasm and wastes resources, but also makes it difficult for environmental control measures to maintain optimal effectiveness due to the lack of precise metabolic data support, this invention provides a method for supervising the breeding, management, and supervision of new fruit and vegetable varieties.

[0006] To achieve the above objectives, this invention employs a method for supervising and managing the cultivation of new fruit and vegetable varieties, comprising the following steps: S1: Obtain the vegetative growth phenotype, single fruit weight, and fruit sugar content of fruits and vegetables from seedling to maturity using crop phenotyping equipment, record environmental control data, arrange them according to the collection time, and construct a time series monitoring set of environment and phenotype. S2: Calculate the transfer entropy value of adjacent sequences in the environment and phenotype time series monitoring set, configure unidirectional driving connection edges for node pairs that are greater than the causal determination threshold, and generate a dynamic trait causal transfer graph. S3: Establish latent variable nodes in the dynamic trait causal transmission map, redirect the parent nodes of single fruit weight and fruit sugar content nodes to the latent variable nodes and connect them to both, and construct a metabolic latent variable supervision network structure. S4: Based on the environmental and phenotypic time-series monitoring set, perform expectation-maximization calculation on the metabolic latent variable supervision network structure, estimate the state distribution of latent variable nodes and correct the conditional probability, and determine the growth state transition probability parameter table. S5: Call the growth state transition probability parameter table, collect the seedling stage phenotypic input metabolic latent variable supervision network structure, calculate the joint posterior probability of mature stage traits, determine the germplasm label based on the distribution status, invert and calculate the environmental compensation parameters, and generate cultivation management supervision instructions.

[0007] As a further aspect of the present invention, the environmental and phenotypic time-series monitoring set includes plant height and leaf area index sequences, soluble solids content time-series data, and greenhouse light, temperature, and humidity records; the dynamic trait causal transmission map includes a set of network nodes, directed edges, and a transmission entropy matrix; the metabolic latent variable supervision network structure includes latent variable nodes, parent-child dependencies, and network topology; the growth state transition probability parameter table includes state distribution parameters, a conditional probability table, and a state transition matrix; and the cultivation management supervision instructions include variety grade classification, environmental control parameters, and water and fertilizer regulation schemes.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: The crop phenotypic acquisition device performs continuous periodic scanning of the target fruits and vegetables from the seedling stage to the maturity stage, collects plant height, stem diameter and leaf area index, detects single fruit weight and fruit sugar content, and aggregates data according to the batch to generate a crop full-cycle growth phenotypic feature vector. S102: Obtain environmental control record data throughout the entire growth cycle of the target fruits and vegetables, read the air temperature, relative humidity, light intensity and carbon dioxide concentration values ​​in the greenhouse, extract the effective values, map them to the collection time, and obtain the set of environmental control status parameters of the facility. S103: Based on the facility environment management status parameter set and the crop full-cycle growth phenotypic feature vector, match data items at the same time according to the consistency association of the collection timestamps and serialize and reassemble them to establish an environment and phenotypic time series monitoring set.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the environmental and phenotypic time-series monitoring set, perform sliding window sampling on the multidimensional sequence, construct the state transition probability distribution, calculate the conditional entropy difference after introducing source variables and quantify the information transmission intensity, and generate the environmental-phenotypic interaction transmission entropy matrix. S202: Call the environmental phenotype interactive transfer entropy matrix, identify non-zero elements whose intensity values ​​exceed the causal determination threshold, lock the row and column coordinates corresponding to the non-zero elements to determine the causal direction, and obtain the key causal driving node index set. S203: Initialize nodes and establish connection edges based on the key causal driving node index set, extract the transfer entropy values ​​from the environmental phenotype interaction transfer entropy matrix and map them as the weights of the connection edges to establish a dynamic trait causal transfer graph.

[0010] As a further aspect of the present invention, the non-zero elements whose identification intensity values ​​exceed the causal determination threshold refer to the statistical distribution mean and standard deviation of the off-diagonal elements in the environmental phenotypic interaction transfer entropy matrix. The sum of the statistical distribution mean and three times the standard deviation is set as the causal determination threshold, and elements in the environmental phenotypic interaction transfer entropy matrix whose values ​​are strictly greater than the causal determination threshold are screened.

[0011] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the dynamic trait causal transmission map, insert latent variable nodes into the network topology, assign photosynthetic product allocation attributes to the latent variable nodes, identify the abstract primitive objects as mediating latent variables connecting environmental factors and quality traits in the network, and generate latent variable nodes. S302: Call the hidden variable node, retrieve the target trait node corresponding to single fruit weight and fruit sugar content, backtrack to lock the upstream parent node, disconnect the original connection relationship, configure a one-way convergence connection edge from the upstream parent node to the hidden variable node, and obtain the driving factor redirection path set; S303: Based on the set of driving factor redirection paths, configure divergent connection edges from latent variable nodes to target trait nodes, integrate the recombined input convergence path and output divergent path, update the graph dependency relationship, and construct the metabolic latent variable supervision network structure.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the metabolic latent variable supervision network structure, map the environmental and phenotypic time series monitoring set indicators to the input and output layers of the network, perform expectation calculation on the latent variable nodes and derive the latent state posterior probability, and generate the metabolic latent node posterior distribution set. S402: Call the posterior distribution set of the metabolic hidden nodes, perform maximization on the log-likelihood function of the complete data, reweight the dependency parameters between network nodes using the posterior distribution values, iteratively update the conditional probability of the causal transmission strength between nodes until convergence, and obtain the modified conditional probability matrix. S403: Extract the converged node state transition values ​​based on the modified conditional probability matrix, index and arrange the node state transition values ​​according to the combination dimension of environmental input and phenotypic output, quantify the state transition probability and fill it into the data table, and establish a growth state transition probability parameter table.

[0013] As a further aspect of the present invention, the step of performing expectation calculation on the latent variable nodes and deriving the latent state posterior probability refers to constructing a forward recursive calculation path and a backward recursive calculation path for the distribution of latent variable nodes in the time series. The joint probability density of environmental and phenotypic time-series monitoring indicators prior to the current moment is accumulated along the forward recursive calculation path to generate forward probability variable values; The conditional probability density of environmental and phenotypic time-series monitoring indicators after the current time is accumulated and calculated along the backward recursive calculation path to generate backward probability variable values.

[0014] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the growth state transition probability parameter table, input the seedling phenotype into the metabolic latent variable supervision network structure, use weight-driven forward inference to calculate the conditional probability of evolution to mature traits, and obtain the joint posterior probability matrix of mature traits. S502: Extract the trait combination feature coordinates based on the joint posterior probability matrix of the maturity traits, project them onto the hierarchical geometric space, calculate the Euclidean distance between the feature coordinates and the cluster center, and determine the category according to the minimization principle to obtain the germplasm prediction hierarchical label. S503: Set target metabolic level constraints for the germplasm prediction and grading labels, perform reverse causal reasoning in the metabolic latent variable supervision network structure, invert the combination of environmental variables driving node transitions and calculate the compensation gain required by environmental factors, and construct cultivation management supervision instructions.

[0015] The fruit and vegetable new variety breeding management and supervision system includes: The data acquisition and construction module acquires the vegetative growth phenotypes, single fruit weight, and fruit sugar content of fruits and vegetables from seedling to maturity through crop phenotyping equipment, records environmental control data, arranges them according to the acquisition time, and constructs a time-series monitoring set of environment and phenotype. The causal graph generation module calculates the entropy value of adjacent sequences in the environmental and phenotypic time-series monitoring set, configures unidirectional driving connection edges for node pairs that are greater than the causal determination threshold, and generates a dynamic trait causal transmission graph. The latent variable network construction module establishes latent variable nodes in the dynamic trait causal transmission graph, redirects the parent nodes of the single fruit weight and fruit sugar content nodes to the latent variable nodes and connects them to the two, thus constructing a metabolic latent variable supervision network structure. The state parameter estimation module performs expectation maximization calculation on the metabolic latent variable supervision network structure based on the environmental and phenotypic time-series monitoring set, estimates the state distribution of latent variable nodes and corrects the conditional probability, and determines the growth state transition probability parameter table. The cultivation supervision and inversion module calls the growth state transition probability parameter table, collects seedling stage phenotypic input metabolic latent variable supervision network structure, calculates the joint posterior probability of maturity stage traits, determines germplasm labels based on distribution status, inverts and calculates environmental compensation parameters, and generates cultivation management supervision instructions.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by constructing a time-series monitoring set of environment and phenotype and calculating the sequence transfer entropy value, a dynamic trait causal transmission relationship is established to overcome the limitations of static observation. Latent variable nodes are set up to reconstruct the single fruit weight and sugar content dependence path. The expectation-maximization algorithm is used to estimate the distribution of metabolic latent variables and correct the growth state transition probability. A joint posterior probability prediction model for traits from seedling stage phenotype to maturity stage is established. The environmental compensation parameters required to improve metabolic level are calculated through inversion. Germplasm grading and environmental intervention are completed in the early stage of crop growth, solving the problem of lagging breeding decisions and significantly improving the accuracy and efficiency of the breeding management process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] Please see Figure 1 This invention provides a method for supervising the cultivation, management, and control of new fruit and vegetable varieties, comprising the following steps: S1: Obtain the vegetative growth phenotype, single fruit weight, and fruit sugar content of fruits and vegetables from seedling to maturity using crop phenotyping equipment, record environmental control data, arrange them according to the collection time, and construct a time series monitoring set of environment and phenotype. S2: Calculate the propagation entropy value of adjacent sequences in the computational environment and phenotypic time series monitoring set, configure unidirectional driving connection edges for node pairs that are greater than the causal determination threshold, and generate a dynamic trait causal propagation graph. S3: In the dynamic trait causal transmission map, a latent variable node is set up, and the parent node of the single fruit weight and fruit sugar content node is redirected to the latent variable node and connected to both, thus constructing a metabolic latent variable supervision network structure. S4: Based on the environmental and phenotypic time-series monitoring set, perform expectation-maximization calculation on the metabolic latent variable supervision network structure, estimate the state distribution of latent variable nodes and correct the conditional probability, and determine the growth state transition probability parameter table. S5: Call the growth state transition probability parameter table, collect seedling phenotypic input metabolic latent variable supervision network structure, calculate the joint posterior probability of maturity traits, determine germplasm labels based on distribution status, invert and calculate environmental compensation parameters, and generate cultivation management supervision instructions.

[0022] The environmental and phenotypic time-series monitoring set includes plant height and leaf area index sequences, soluble solids content time-series data, and greenhouse light, temperature, and humidity records. The dynamic trait causal transmission map includes a set of network nodes, directed edges, and a transmission entropy matrix. The metabolic latent variable supervision network structure includes latent variable nodes, parent-child dependencies, and network topology. The growth state transition probability parameter table includes state distribution parameters, conditional probability tables, and a state transition matrix. The cultivation management supervision instructions include variety grade classification, environmental control parameters, and water and fertilizer regulation schemes.

[0023] Please see Figure 2 The specific steps of S1 are as follows: S101: The crop phenotypic acquisition device performs continuous periodic scanning of the target fruits and vegetables from the seedling stage to the maturity stage, collects plant height, stem diameter and leaf area index, detects single fruit weight and fruit sugar content, and aggregates data according to the batch to generate a crop full-cycle growth phenotypic feature vector. First, the high-throughput phenotypic acquisition equipment, integrated with a high-resolution 3D LiDAR, a hyperspectral imager, and a visible light depth camera, is activated on a track-mounted mobile platform. Before the scanning operation begins, an initialization self-test procedure is executed to align the origin of the 3D LiDAR coordinate system with the global coordinate system of the greenhouse, and the scanning frequency is set to 50 Hz and the angular resolution to 0.1 degrees. The mobile platform moves smoothly along the crop planting rows at a constant speed of 0.5 meters per second. The laser beam emitted by the 3D LiDAR covers the vertical space of the crop from the roots to the canopy, acquiring a dense point cloud data stream. For extracting the plant height value, the processing unit first uses a pass-through filtering algorithm to remove background noise, retaining the point cloud within the region of interest. Then, a random sampling consistency algorithm is used to identify the ground plane and calculate the vertical Euclidean distance from the highest point of the crop canopy to the ground plane; this distance is the plant height value. For example, when the Z-axis coordinate of the highest point of the canopy is identified as 150.5 cm and the Z-axis coordinate of the ground is 5.2 cm, the plant height value is obtained by subtraction as 145.3 cm. For stem diameter, a machine vision algorithm was used to extract the skeleton from the depth image. A cross-section of the stem was taken 5 cm from the root, and the diameter of the fitted circle of the cross-section was calculated. For leaf area index, a hyperspectral imager was used to acquire spectral images in the wavelength range of 400 nm to 1000 nm. The normalized vegetation index was calculated, and the leaf area index was inverted using a preset extinction coefficient model. For single fruit weight and fruit sugar content, near-infrared spectroscopy was used to acquire the diffuse reflectance spectrum of the fruit in the 900 nm to 1700 nm band. A partial least squares regression model was used to map the spectral absorbance data to the Brix sugar content value. At the same time, the fruit volume data acquired by the binocular vision system was multiplied by a preset fruit density coefficient to estimate the single fruit weight. The collected data were indexed by the batch ID. For example, for batch ID 20260122A, the plant height, stem diameter, leaf area index, single fruit weight, and sugar content data of plants in that batch were standardized. Data standardization employs the Z-score standardization method, which involves subtracting the arithmetic mean of the batch data from the original value, and then dividing the difference by the standard deviation of the batch data. For example, if a plant height is 150 cm, the batch average height is 140 cm, and the standard deviation is 10 cm, then the standardized feature value is 1.0. Finally, the standardized feature values ​​are concatenated in chronological order to generate a crop full-cycle growth phenotypic feature vector. This process relies on physical and statistical regression models. For the near-infrared spectroscopy sugar content retrieval model, partial least squares is used. The input layer consists of absorbance values ​​across 256 bands, which are then dimensionality-reduced to extract 10 principal components. The output layer is a single predicted sugar content value.

[0024] S102: Obtain environmental control record data throughout the entire growth cycle of the target fruits and vegetables, read the air temperature, relative humidity, light intensity and carbon dioxide concentration values ​​in the greenhouse, extract the effective values, map them to the collection time, and obtain the set of environmental control status parameters of the facility. The acquisition of environmental control data relies on a distributed IoT sensor network deployed within the greenhouse. This network includes high-precision temperature and humidity sensors, photonic flux density sensors, and non-dispersive infrared carbon dioxide sensors. Instantaneous values ​​are read from the sensor nodes every 5 minutes via the Modbus communication protocol. During the data cleaning phase, extreme value filtering is first performed, setting reasonable physical range thresholds. For example, the effective range for air temperature is set to 0°C to 50°C, relative humidity to 0% to 100%, light intensity to 0 lux to 100,000 lux, and carbon dioxide concentration to 200 ppm to 2000 ppm. If the read air temperature value is 250°C, significantly exceeding the upper threshold of 50°C, the data is considered abnormal noise and discarded. After discarding, for missing time points, cubic spline interpolation is used to fit a smooth curve using three adjacent valid data points to estimate the value at the missing time and fill the data gaps. After extracting the valid values, the environmental data needs to be mapped to the crop phenotypic collection time. Since environmental data is collected every 5 minutes, while phenotypic scanning can occur at any time, a linear weighted matching method is used. For example, if the phenotypic data collection time is 10:03, and the environmental records have two data points at 10:00 and 10:05, the time difference weights are calculated. The weight for 10:00 is 2 divided by 5, which equals 0.4, and the weight for 10:05 is 3 divided by 5, which equals 0.6. The weight calculation logic here is inverse distance weighting, meaning the closer to the target time, the greater the weight. Specifically, 10:00 is 3 minutes from the target time, and 10:05 is 2 minutes from the target time, for a total interval of 5 minutes. Therefore, the weight for 10:00 is 2 divided by 5, and the weight for 10:05 is 3 divided by 5. The temperature value at 10:00 AM (25 degrees Celsius) was then multiplied by 0.4, and the temperature value at 10:05 AM (26 degrees Celsius) was multiplied by 0.6, resulting in a mapped temperature value of 25.6 degrees Celsius at 10:03 AM. Using this method, the corresponding air temperature, relative humidity, light intensity, and carbon dioxide concentration were calculated for each phenotypic data collection time. Finally, the mapped values ​​were encapsulated to obtain a set of facility environmental management status parameters. This parameter set is stored in a structured data table, with each row containing a timestamp and the corresponding values ​​for the four environmental factors.

[0025] S103: Based on the facility environment management status parameter set and crop full-cycle growth phenotypic feature vector, the data items at the same time are matched and serialized and recombined according to the consistency of the collection timestamps to establish an environmental and phenotypic time series monitoring set; This step performs data fusion and serialization / reassembly operations. First, the generated set of facility environmental management status parameters and the generated crop full-cycle growth phenotypic feature vector are loaded. Since both sets of data already contain a unified formatted timestamp index during storage, an inner join query is performed, using the timestamp as the key to strictly match the two sets of data. For each successfully matched time point, a data frame containing multi-dimensional attributes is constructed. The first half of this data frame is filled with environmental management status parameters, including air temperature, relative humidity, light intensity, and carbon dioxide concentration; the second half is filled with crop phenotypic features, including plant height, stem diameter, leaf area index, single fruit weight, and fruit sugar content. For example, at timestamp T1, the environmental data is [25.6 degrees Celsius, 65%, 45000 lux, 600 ppm], and the phenotypic data is [145.3 cm, 1.2 cm, 3.5, 150 g, 8.5 degrees Celsius]. These two sets of vectors are merged into [25.6, 65, 45000, 600, 145.3, 1.2, 3.5, 150, 8.5]. After single-point merging, the merged data frames are arranged into a time series according to the order of the timestamps. To eliminate the influence of dimensional differences on subsequent calculations, the merged sequence is normalized again. The min-max normalization method is used to find the maximum and minimum values ​​of each attribute in the entire period for each column of the sequence. The normalization calculation logic is: subtract the minimum value from the current value, and then divide the difference by the difference between the maximum and minimum values. For example, if the maximum light intensity over the entire period is 50,000 lux, the minimum is 0 lux, and the current value is 45,000 lux, then the normalized value is 45,000 minus 0, divided by 50,000 minus 0, resulting in 0.9. After normalization, the values ​​of the data items are all distributed within a closed interval between 0 and 1. Finally, the processed ordered multidimensional data sequence is stored as an environmental and phenotypic time-series monitoring set. This dataset has temporal continuity and dimensional consistency, providing standard input for subsequent causal entropy calculations. See Table 1, which shows part of the data structure of the environmental and phenotypic time-series monitoring set.

[0026] Table 1: Normalized Time-Series Monitoring Set Fragments of Environment and Phenotypic Data

[0027] As shown in Table 1, the data has been normalized and aligned by time.

[0028] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the environmental and phenotypic time series monitoring set, a sliding window sampling is performed on the multidimensional sequence to construct the state transition probability distribution, calculate the conditional entropy difference after introducing source variables and quantify the information transmission intensity, and generate the environmental-phenotypic interaction transmission entropy matrix. First, define the length of the sliding window, for example, set the window length to 50 time steps. The sliding window starts from the beginning of the environmental and phenotypic time-series monitoring set and moves forward by one time step each time. Within each window, select an environmental factor sequence as the source variable X and a phenotypic trait sequence as the target variable Y, and calculate the transfer entropy between them. First, construct the state transition probability distribution, which involves discretizing the continuous data. Use the histogram method to map the continuous values ​​into 10 discrete state intervals. The specific logic for calculating the conditional entropy difference after introducing the source variable is as follows: First, calculate the Shannon entropy of the target variable Y's future state. The calculation logic is to traverse the possible states and calculate the negative of the sum of the natural logarithm of the probability of each state and the product of that probability. Next, calculate the conditional entropy under the condition of the known past state of the target variable, and the conditional entropy under the condition of the known past state of the target variable and the past state of the source variable. The value of the transfer entropy is equal to the conditional entropy of the known past state of the target variable minus the conditional entropy under the condition of the known past state of the target variable and the past state of the source variable. For example, if only historical information about the target variable is known, the uncertainty in predicting the future (conditional entropy) is 1.5 bits. However, after introducing historical information about the source variable, the uncertainty in predicting the future (conditional entropy) decreases to 1.2 bits. The difference of 0.3 bits is the transfer entropy value, quantifying the intensity of information transfer. This calculation is performed on pairwise variable combinations in the monitoring set. Assuming there are 4 environmental factors and 5 phenotypic traits, a total of 4 x 5, or 20 pairs of relationships, need to be calculated, as well as the transfer entropy between each pair within a phenotypic trait. The calculated transfer entropy values ​​are filled into a square matrix to generate an environment-phenotypic interaction transfer entropy matrix. The rows of this matrix represent driving nodes, the columns represent receiving nodes, and the element values ​​in the matrix represent the intensity of information flow.

[0029] S202: Call the environment phenotype interactive transfer entropy matrix, identify non-zero elements whose intensity values ​​exceed the causal determination threshold, lock the row and column coordinates corresponding to the non-zero elements to determine the causal direction, and obtain the key causal driving node index set. After invoking the environment phenotype interactive entropy matrix, the statistical distribution characteristics of the off-diagonal elements in the matrix are first calculated. The values ​​excluding the main diagonal are traversed, and the arithmetic mean and standard deviation of the values ​​are calculated. Assuming there are 90 off-diagonal elements in the matrix, these 90 values ​​are added together and divided by 90 to obtain the mean, for example, a mean of 0.15. Next, the square of the difference between each value and the mean is calculated, summed, divided by 90, and the square root is taken to obtain the standard deviation, for example, a standard deviation of 0.05. The logic for setting the causal determination threshold is: the mean plus a predetermined multiple of the standard deviation. Here, the predetermined multiple is set to 3, based on the 3σ criterion to identify significant causal connections. Substituting the aforementioned values ​​into the calculation, the causal determination threshold equals 0.15 plus 3 multiplied by 0.05, resulting in 0.30. Subsequently, each element in the matrix is ​​traversed again, and elements with values ​​strictly greater than 0.30 are selected. For example, if the transfer entropy value of the element pointing from "light intensity" to "fruit sugar content" in the matrix is ​​0.45, which is greater than 0.30, then this element is retained; if the value of the element pointing from "air humidity" to "plant height" is 0.20, which is less than 0.30, then this element is set to zero or ignored. For the retained non-zero elements, their row and column indices are locked. The row index represents the causal source node, and the column index represents the causal result node. For example, locking the 3rd row and 5th column indicates that the 3rd variable is a key driver of the 5th variable. The row and column index pairs that meet the conditions are collected to form a key causal driver node index set. This process effectively eliminates background noise and weak correlations, retaining statistically significant strong causal driving paths.

[0030] S203: Initialize nodes and establish connection edges based on the key causal driving node index set, extract the transfer entropy values ​​from the environmental phenotype interaction transfer entropy matrix and map them as the weights of the connection edges to establish a dynamic trait causal transfer graph. Based on the key causal driver node index set, a directed weighted graph data structure is initialized in memory. First, a graph node is created for each involved variable, including environmental factor nodes and crop trait nodes. Next, each pair of indices in the index set is traversed, and a directed connection edge is established between the corresponding source and target nodes. The weight of the connection edge is directly assigned the extracted corresponding transit entropy value. For example, if the index set contains a connection from "carbon dioxide concentration" to "photosynthetic rate," and the corresponding transit entropy value is 0.38, then a directed edge is established from the carbon dioxide node to the photosynthetic rate node, and the weight attribute of this edge is set to 0.38. This process is repeated until all causal pairs in the index set are processed. The established graph is the dynamic trait causal transit graph. To enhance the expressive power of the graph, the weighted sum of the out-degree and in-degree of each node is also calculated. The weighted sum of the out-degree of a node is the sum of the weights of all edges emanating from that node, representing the node's regulatory capacity; the weighted sum of the in-degree of a node is the sum of the weights of all edges pointing to that node, representing the degree to which the node is influenced by the environment or upstream traits. Ultimately, the graph is stored in the form of an adjacency matrix or adjacency list, providing a foundation for subsequent topology analysis.

[0031] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the dynamic trait causal transmission graph, latent variable nodes are inserted into the network topology, and the latent variable nodes are given photosynthetic product allocation attributes. The abstract graph primitive objects are identified as mediating latent variables connecting environmental factors and quality traits in the network, and latent variable nodes are generated. At the topology layer, the creation of abstract primitive objects is performed. This operation does not correspond to a physical entity, but rather creates a virtual node representing the physiological process of "photosynthetic product allocation." A graph node object is instantiated and named the "latent variable node." This object is defined with a set of attributes, including "energy input stock," "allocation ratio vector," and "metabolic efficiency coefficient." In the logical definition of the graph, this node is identified as an intermediary latent variable, designed to simulate the unobservable process within the plant of converting energy assimilated from the environment into specific traits. To generate the latent variable node, it is inserted into the existing dynamic trait causal transmission graph and assigned a unique node ID. At this point, the node is still suspended in the network and not connected to any existing nodes. The initial latent state posterior probability of the node is set to a uniform distribution, assuming that at the initial moment, metabolic resources are equally probable across multiple possible allocation states. This step completes the expansion of the network structure, introducing a latent variable layer constrained by biological mechanisms, making the model not merely a purely data-driven model, but a hybrid model integrating prior knowledge of plant physiology.

[0032] S302: Call the latent variable node, retrieve the target trait node corresponding to single fruit weight and fruit sugar content, backtrack to lock the upstream parent node, disconnect the original connection relationship, configure a one-way convergence connection edge from the upstream parent node to the latent variable node, and obtain the driving factor redirection path set; The newly created latent variable node is invoked, and the target trait node, corresponding to "single fruit weight" and "fruit sugar content," is retrieved in the graph. Then, a backtracking operation is performed to find the upstream parent node of the target trait node in the original graph. For example, in the original graph, "light intensity" and "carbon dioxide concentration" directly point to "fruit sugar content." After identifying these two parent nodes, a disconnection operation is performed, deleting the edges pointing from "light intensity" to "fruit sugar content," and from "carbon dioxide concentration" to "fruit sugar content." Next, new connection relationships are configured: directed edges are established from "light intensity" to the "latent variable node," and directed edges are established from "carbon dioxide concentration" to the "latent variable node." The newly established connection edges are defined as unidirectional converging connection edges, meaning that environmental driving forces first converge on metabolic latent variables. The indexes of the redirected parent nodes and the new connection relationships are recorded to obtain the set of driving factor redirection paths. This operation logically forces that environmental influences must go through the metabolic process before acting on the final quality trait, which conforms to biological laws. For example, increased light exposure first increases photosynthetic metabolites, which are then converted into sugar accumulation, rather than light directly turning into sugar.

[0033] S303: Based on the set of driving factor redirection paths, configure divergent connection edges from latent variable nodes to target trait nodes, integrate the recombined input convergence path and output divergent path, update the graph dependencies, and construct the metabolic latent variable supervision network structure. Based on the set of driving factor redirection paths, the connections originating from the latent variable nodes are further configured. Divergent connections are established from the latent variable nodes to single fruit weight and to fruit sugar content. At this point, the metabolic node becomes a hub, receiving input from upstream environmental factors and disseminating their influence to downstream quality traits. The weights of the new edges are initialized, with the initial weights set to the mean of the original direct connection weights or a preset prior value, such as 0.5. After completing the connection configuration, a graph dependency update operation is performed, recalculating the network's adjacency matrix to ensure that newly added nodes and edges are correctly indexed. The structure completed at this point is called the metabolic latent variable supervision network structure. This structure essentially forms a hierarchical model: the input layer is environmental factors, the intermediate layer is metabolic latent variables, and the output layer is phenotypic traits. This structure can capture the nonlinear effects of the synergistic interactions between environmental factors on metabolic processes. Finally, the integrity of the network structure will be verified to ensure that there are no isolated nodes and that each quality node has and only has direct input from the metabolic node (according to the recombined logic), thus guaranteeing the mediating control status of the metabolic node.

[0034] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the metabolic latent variable supervision network structure, map the environmental and phenotypic time series monitoring set indicators to the input and output layers of the network, perform expectation calculation on the latent variable nodes and derive the latent state posterior probability, and generate the metabolic latent node posterior distribution set. The metabolic latent variable supervision network structure is invoked to map the generated environmental and phenotypic time-series monitoring sets to the network's input layer (environmental factors) and output layer (phenotypic traits). For latent variable nodes, expectation calculation is performed to derive the state distribution of the latent variable at each time step. This involves the execution of a forward-backward algorithm. First, a forward recursive computation path is constructed. For each time step t in the time series, the forward variable α is defined as the joint probability of the metabolic latent node being in state i at time step t, based on environmental and phenotypic data observed before (including time step t). The calculation logic is as follows: the value of the forward variable in state i at the current time step t is equal to the sum of the products of the values ​​of the forward variables in all possible states j at the previous time step t-1 and the transition probabilities of state j transitioning to state i, multiplied by the emission probability of the observed data at the current time step. For example, if the forward variables of the two states at the previous time step were 0.2 and 0.3, respectively, and their transition probabilities were 0.5 and 0.4, respectively, and the current observation probability was 0.8, then the forward variable value of state i at the current time step was 0.2 × 0.5 + 0.3 × 0.4 = 0.22, and 0.22 × 0.8 = 0.176. This process is repeated along the time axis from the starting point to the ending point until the forward probability variable values ​​for all time steps are calculated. Next, a backward recursive calculation path is constructed. The backward variable β is defined as the conditional probability of observing all data after time t, given that the metabolic hidden node is in state i at time t. The calculation logic is: the backward variable value of state i at the current time t is equal to the sum of the products of the transition probabilities of all possible states j at the next time step t+1, the observation emission probability in state j, and the backward variable value of state j. This process is repeated backward along the time axis from the ending point to the starting point to generate the backward probability variable values. Finally, the posterior distribution set of the metabolic hidden node is generated. For any time t, the posterior probability γ of a metabolic hidden node being in state i is obtained by calculating the product of the forward and backward variables and dividing it by the sum of the products of the forward and backward variables of the state at that time (the normalization factor). For example, if at a certain time the forward variable of state 1 is 0.176, the backward variable is 0.5, and the product is 0.088; the product of state 2 is 0.012; and the sum is 0.100, then the posterior probability of state 1 is 0.088 divided by 0.100, which is 0.88. This set quantifies the probability that a metabolic node is in different metabolic intensities (such as high metabolism, low metabolism) at multiple times given complete observation data.

[0035] S402: Call the posterior distribution set of the metabolic hidden nodes, perform maximization on the log-likelihood function of the complete data, reweight the dependency parameters between network nodes using the posterior distribution values, iteratively update the conditional probability of the causal transmission strength between nodes until convergence, and obtain the modified conditional probability matrix. In the maximization phase, the network parameters are updated using the obtained posterior distribution set of metabolic hidden nodes to maximize the log-likelihood function of the complete data. The parameters that need updating mainly include the dependency parameters between nodes (i.e., connection weights or conditional probability tables). The update logic is as follows: the new transition probability (from state i to state j) equals the sum of the joint posterior probabilities of transitioning from state i to state j at each time step, divided by the sum of the posterior probabilities of being in state i at each time step. For example, if the expected number of transitions from "low metabolism" to "high metabolism" over the entire cycle is 20, and the total expected number of times in "low metabolism" is 100, then the updated transition probability is 20 divided by 100, which is 0.2. For the connection weights, gradient ascent is used for correction. The partial derivative of the log-likelihood function with respect to the weights is calculated, using the posterior distribution values ​​as weighting coefficients. The new weights equal the old weights plus the learning rate multiplied by the gradient value. Assuming a learning rate of 0.01, a calculated gradient of 0.5, and an old weight of 0.4, the new weight update is 0.4 plus 0.01 multiplied by 0.5, equaling 0.405. An iterative process is then executed, alternating between expectation calculation and parameter updates. After each iteration, the increment of the log-likelihood function is calculated. The iteration is considered converged when the increment of the log-likelihood function for two consecutive iterations is less than a preset convergence threshold (e.g., 0.001). The resulting parameter matrix is ​​the modified conditional probability matrix, describing the strength of dynamic causal transmission between environmental, metabolic latent variables, and traits.

[0036] S403: Extract the converged node state transition values ​​based on the modified conditional probability matrix, index and arrange the node state transition values ​​according to the combination dimension of environmental input and phenotypic output, quantify the state transition probability and fill it into the data table, and establish the growth state transition probability parameter table. Based on the modified conditional probability matrix, converged node state transition values ​​are extracted. These values ​​describe the probability of metabolic latent variables transitioning to the next state under different environmental inputs and the current state. A multidimensional lookup table is constructed, with dimensions including the current environmental input combination (discretized), the current metabolic state, and the next metabolic state. The node state transition values ​​are indexed and arranged according to the combination dimension of environmental input and phenotypic output. For example, if the environmental combination is "high temperature-high light" and the current metabolism is "high," the probability of transitioning to "high" metabolism in the next time step is 0.8, and the probability of transitioning to "low" is 0.2. To quantify the state transition probability, the probability flow along the state transition path is calculated, i.e., the probability value multiplied by the frequency of state occurrence. The data is populated into a data table to establish a growth state transition probability parameter table. This table is actually a very detailed dynamic programming table, capable of querying the probability distribution of future state evolution based on the currently observed environmental and crop states. See Table 2, which shows some growth state transition probability parameters.

[0037] Table 2: Fragment of Growth State Transition Probability Parameter Table

[0038] As shown in Table 2, different environmental combinations significantly affect the maintenance or switching of metabolic states.

[0039] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the growth state transition probability parameter table, input the seedling phenotype into the metabolic latent variable supervision network structure, use weight-driven forward inference to calculate the conditional probability of evolution to maturity traits, and obtain the joint posterior probability matrix of maturity traits. The forward inference process is initiated by calling the growth state transition probability parameter table. Input data includes measured phenotypic data of the crop seedling stage (e.g., seedling height, leaf area) and a preset sequence of future environmental scenarios. The seedling phenotypic values ​​are mapped to the initial state distribution of metabolic latent variables. Subsequently, weighted forward inference is used. For each future time step, based on the current predicted state distribution and environmental input, the probability parameter table is consulted to calculate the state distribution for the next time step. The calculation logic is matrix-vector multiplication: the state probability vector for the next time step equals the current state probability vector multiplied by the state transition matrix under the corresponding environment. This inference process continues until the crop reaches maturity. At the end of the inference, the conditional probability distribution of the maturity trait is calculated by combining the emission probabilities of metabolic latent variables to phenotypic traits. For example, the final inference shows that the probability of a "high" metabolic state at maturity is 0.7, and the probability of a "low" metabolic state is 0.3. It is known that the probability of producing a single fruit weighing more than 200 grams under the "high" metabolic state is 0.9, and under the "low" metabolic state it is 0.2. The joint posterior probability of a single fruit weighing more than 200 grams at maturity is 0.7 multiplied by 0.9 plus 0.3 multiplied by 0.2, resulting in 0.63 plus 0.06 equaling 0.69. Performing this calculation on traits of interest (single fruit weight, sugar content) yields the joint posterior probability matrix for maturity traits, which provides a probabilistic prediction of the final quality of the crop under specific environmental management conditions.

[0040] S502: Extract the feature coordinates of trait combinations based on the joint posterior probability matrix of mature traits, project them onto the hierarchical geometric space, calculate the Euclidean distance between the feature coordinates and the cluster center, and determine the category according to the minimization principle to obtain the germplasm prediction hierarchical label. Based on the joint posterior probability matrix of maturity traits, the most probable combination of trait values ​​(e.g., the point with the highest probability density or expected value) is extracted as feature coordinates. Assume the extracted feature coordinates are (single fruit weight: 210 grams, sugar content: 9.5 degrees). These coordinates are projected onto a pre-constructed hierarchical geometric space. This space is formed by clustering a large amount of historical germplasm data, containing multiple cluster centers representing different grades (e.g., premium, first-grade, second-grade). The Euclidean distance between the feature coordinates and multiple cluster centers is calculated. The Euclidean distance calculation logic is: √((feature single fruit weight minus center single fruit weight) squared plus (feature sugar content minus center sugar content) squared). Assume the center coordinates of "premium grade" are (220, 10), and the center coordinates of "first-grade" are (180, 8). Calculate the distance from the feature point (210, 9.5) to the center of "Special Grade": (210-220)^2 + (9.5-10)^2 = 100 + 0.25 = 100.25, approximately 10.01. Calculate the distance to the center of "First Grade": (210-180)^2 + (9.5-8)^2 = 900 + 2.25 = 902.25, approximately 30.04. Based on the minimization principle, 10.01 is less than 30.04, therefore the germplasm category is determined to be "Special Grade". The final germplasm prediction grading label is obtained.

[0041] S503: Set target metabolic level constraints for germplasm prediction and grading labels, perform reverse causal reasoning in the metabolic latent variable supervision network structure, invert the combination of environmental variables driving node transitions and calculate the compensation gain required by environmental factors, and construct cultivation management supervision instructions. For the obtained germplasm prediction grading labels (such as "premium grade"), stricter constraints are set on the target metabolic level, such as requiring that the occupancy rate of the "high metabolism" state is not less than 80% throughout the entire cycle. Backward causal reasoning is performed in the metabolic latent variable supervision network structure. This is a type of constraint satisfaction problem or backpropagation optimization problem. The target state of the output layer is fixed, and the required input distribution for the intermediate metabolic nodes is derived in reverse, thereby deriving the required range of values ​​for the input layer environmental factors. The logic for calculating the compensation gain required for the environmental factors is: calculate the difference between the target environmental value and the current predicted environmental value. For example, the inversion shows that to maintain "high metabolism," the light intensity needs to reach 50,000 lux, while the current predicted environment only has 40,000 lux at a certain time. The compensation gain is calculated as 50,000 minus 40,000, which equals 10,000 lux. This inversion process, applied to factors such as temperature, humidity, and CO2, generates a series of specific control commands, such as "turn on the supplemental lighting to increase illumination by 10,000 lux during the T20-T30 period" and "increase the CO2 concentration to 800 ppm at T40." These commands are then integrated to form the final cultivation management and supervision instructions, which are directly sent to the greenhouse environmental control system for execution. This process achieves a closed-loop feedback from "predicted results" to "intervention process," ensuring that the cultivation of new varieties proceeds according to predetermined goals.

[0042] Please see Figure 7 The fruit and vegetable new variety breeding management and supervision system includes: The data acquisition and construction module acquires the vegetative growth phenotypes, single fruit weight, and fruit sugar content of fruits and vegetables from seedling to maturity through crop phenotyping equipment, records environmental control data, arranges them according to the acquisition time, and constructs a time-series monitoring set of environment and phenotype. The causal graph generation module calculates the entropy value of adjacent sequences in the phenotypic time series monitoring set, configures unidirectional driving connection edges for node pairs that exceed the causal determination threshold, and generates a dynamic trait causal transmission graph. The latent variable network construction module sets up latent variable nodes in the dynamic trait causal transmission graph, redirects the parent nodes of the single fruit weight and fruit sugar content nodes to the latent variable nodes and connects them to the two, thus constructing a metabolic latent variable supervision network structure. The state parameter estimation module performs expectation maximization calculation on the metabolic latent variable supervision network structure based on the environmental and phenotypic time-series monitoring set, estimates the state distribution of latent variable nodes and corrects the conditional probability, and determines the growth state transition probability parameter table. The cultivation supervision and inversion module calls the growth state transition probability parameter table, collects seedling phenotypic input metabolic latent variable supervision network structure, calculates the joint posterior probability of maturity traits, determines germplasm labels based on distribution status, inverts and calculates environmental compensation parameters, and generates cultivation management supervision instructions.

[0043] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A method for supervising the cultivation, management, and supervision of new fruit and vegetable varieties, characterized in that: Includes the following steps: S1: Obtain the vegetative growth phenotype, single fruit weight, and fruit sugar content of fruits and vegetables from seedling to maturity using crop phenotyping equipment, record environmental control data, arrange them according to the collection time, and construct a time series monitoring set of environment and phenotype. S2: Calculate the transfer entropy value of adjacent sequences in the environment and phenotype time series monitoring set, configure unidirectional driving connection edges for node pairs that are greater than the causal determination threshold, and generate a dynamic trait causal transfer graph. S3: Establish latent variable nodes in the dynamic trait causal transmission map, redirect the parent nodes of single fruit weight and fruit sugar content nodes to the latent variable nodes and connect them to both, and construct a metabolic latent variable supervision network structure. S4: Based on the environmental and phenotypic time-series monitoring set, perform expectation-maximization calculation on the metabolic latent variable supervision network structure, estimate the state distribution of latent variable nodes and correct the conditional probability, and determine the growth state transition probability parameter table. S5: Call the growth state transition probability parameter table, collect the seedling stage phenotypic input metabolic latent variable supervision network structure, calculate the joint posterior probability of mature stage traits, determine the germplasm label based on the distribution status, invert and calculate the environmental compensation parameters, and generate cultivation management supervision instructions.

2. The method for supervising and managing the cultivation of new fruit and vegetable varieties according to claim 1, characterized in that, The environmental and phenotypic time-series monitoring set includes plant height and leaf area index sequences, soluble solids content time-series data, and greenhouse light, temperature, and humidity records. The dynamic trait causal transmission map includes a set of network nodes, directed edges, and a transmission entropy matrix. The metabolic latent variable supervision network structure includes latent variable nodes, parent-child dependencies, and network topology. The growth state transition probability parameter table includes state distribution parameters, conditional probability tables, and a state transition matrix. The cultivation management supervision instructions include variety grade classification, environmental control parameters, and water and fertilizer regulation schemes.

3. The method for supervising and managing the cultivation of new fruit and vegetable varieties according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: The crop phenotypic acquisition device performs continuous periodic scanning of the target fruits and vegetables from the seedling stage to the maturity stage, collects plant height, stem diameter and leaf area index, detects single fruit weight and fruit sugar content, and aggregates data according to the batch to generate a crop full-cycle growth phenotypic feature vector. S102: Obtain environmental control record data throughout the entire growth cycle of the target fruits and vegetables, read the air temperature, relative humidity, light intensity and carbon dioxide concentration values ​​in the greenhouse, extract the effective values, map them to the collection time, and obtain the set of environmental control status parameters of the facility. S103: Based on the facility environment management status parameter set and the crop full-cycle growth phenotypic feature vector, match data items at the same time according to the consistency association of the collection timestamps and serialize and reassemble them to establish an environment and phenotypic time series monitoring set.

4. The method for supervising and managing the cultivation of new fruit and vegetable varieties according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the environmental and phenotypic time-series monitoring set, perform sliding window sampling on the multidimensional sequence, construct the state transition probability distribution, calculate the conditional entropy difference after introducing source variables and quantify the information transmission intensity, and generate the environmental-phenotypic interaction transmission entropy matrix. S202: Call the environmental phenotype interactive transfer entropy matrix, identify non-zero elements whose intensity values ​​exceed the causal determination threshold, lock the row and column coordinates corresponding to the non-zero elements to determine the causal direction, and obtain the key causal driving node index set. S203: Initialize nodes and establish connection edges based on the key causal driving node index set, extract the transfer entropy values ​​from the environmental phenotype interaction transfer entropy matrix and map them as the weights of the connection edges to establish a dynamic trait causal transfer graph.

5. The method for supervising and managing the cultivation of new fruit and vegetable varieties according to claim 4, characterized in that, The non-zero elements whose identification intensity values ​​exceed the causal determination threshold refer to the statistical distribution mean and standard deviation of the off-diagonal elements in the calculated environmental phenotypic interaction transfer entropy matrix. The sum of the statistical distribution mean and three times the standard deviation is set as the causal determination threshold, and elements in the environmental phenotypic interaction transfer entropy matrix whose values ​​are strictly greater than the causal determination threshold are selected.

6. The method for supervising the cultivation, management, and supervision of new fruit and vegetable varieties according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the dynamic trait causal transmission map, insert latent variable nodes into the network topology, assign photosynthetic product allocation attributes to the latent variable nodes, identify the abstract primitive objects as mediating latent variables connecting environmental factors and quality traits in the network, and generate latent variable nodes. S302: Call the hidden variable node, retrieve the target trait node corresponding to single fruit weight and fruit sugar content, backtrack to lock the upstream parent node, disconnect the original connection relationship, configure a one-way convergence connection edge from the upstream parent node to the hidden variable node, and obtain the driving factor redirection path set; S303: Based on the set of driving factor redirection paths, configure divergent connection edges from latent variable nodes to target trait nodes, integrate the recombined input convergence path and output divergent path, update the graph dependency relationship, and construct the metabolic latent variable supervision network structure.

7. The method for supervising the cultivation, management, and supervision of new fruit and vegetable varieties according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Call the metabolic latent variable supervision network structure, map the environmental and phenotypic time series monitoring set indicators to the input and output layers of the network, perform expectation calculation on the latent variable nodes and derive the latent state posterior probability, and generate the metabolic latent node posterior distribution set. S402: Call the posterior distribution set of the metabolic hidden nodes, perform maximization on the log-likelihood function of the complete data, reweight the dependency parameters between network nodes using the posterior distribution values, iteratively update the conditional probability of the causal transmission strength between nodes until convergence, and obtain the modified conditional probability matrix. S403: Extract the converged node state transition values ​​based on the modified conditional probability matrix, index and arrange the node state transition values ​​according to the combination dimension of environmental input and phenotypic output, quantify the state transition probability and fill it into the data table, and establish a growth state transition probability parameter table.

8. The method for supervising the cultivation, management, and supervision of new fruit and vegetable varieties according to claim 7, characterized in that, The process of performing expectation calculation on latent variable nodes and deriving the posterior probability of latent states refers to constructing forward recursive calculation paths and backward recursive calculation paths for the distribution of latent variable nodes in the time series. The joint probability density of environmental and phenotypic time-series monitoring indicators prior to the current moment is accumulated along the forward recursive calculation path to generate forward probability variable values; The conditional probability density of environmental and phenotypic time-series monitoring indicators after the current time is accumulated and calculated along the backward recursive calculation path to generate backward probability variable values.

9. The method for supervising the cultivation, management, and supervision of new fruit and vegetable varieties according to claim 7, characterized in that, The specific steps of S5 are as follows: S501: Call the growth state transition probability parameter table, input the seedling phenotype into the metabolic latent variable supervision network structure, use weight-driven forward inference to calculate the conditional probability of evolution to mature traits, and obtain the joint posterior probability matrix of mature traits. S502: Extract the trait combination feature coordinates based on the joint posterior probability matrix of the maturity traits, project them onto the hierarchical geometric space, calculate the Euclidean distance between the feature coordinates and the cluster center, and determine the category according to the minimization principle to obtain the germplasm prediction hierarchical label. S503: Set target metabolic level constraints for the germplasm prediction and grading labels, perform reverse causal reasoning in the metabolic latent variable supervision network structure, invert the combination of environmental variables driving node transitions and calculate the compensation gain required by environmental factors, and construct cultivation management supervision instructions.

10. A management and supervision system for the cultivation of new fruit and vegetable varieties, characterized in that: The system is used to implement the method for supervising and managing the cultivation of new fruit and vegetable varieties as described in any one of claims 1-9, and the system includes: The data acquisition and construction module acquires the vegetative growth phenotypes, single fruit weight, and fruit sugar content of fruits and vegetables from seedling to maturity through crop phenotyping equipment, records environmental control data, arranges them according to the acquisition time, and constructs a time-series monitoring set of environment and phenotype. The causal graph generation module calculates the transfer entropy value of adjacent sequences in the environment and phenotypic time-series monitoring set, configures unidirectional driving connection edges for node pairs that are greater than the causal determination threshold, and generates a dynamic trait causal transfer graph. The latent variable network construction module establishes latent variable nodes in the dynamic trait causal transmission graph, redirects the parent nodes of the single fruit weight and fruit sugar content nodes to the latent variable nodes and connects them to both, thus constructing a metabolic latent variable supervision network structure. The state parameter estimation module performs expectation maximization calculation on the metabolic latent variable supervision network structure based on the environmental and phenotypic time-series monitoring set, estimates the state distribution of latent variable nodes and corrects the conditional probability, and determines the growth state transition probability parameter table. The cultivation supervision and inversion module calls the growth state transition probability parameter table, collects seedling stage phenotypic input metabolic latent variable supervision network structure, calculates the joint posterior probability of maturity stage traits, determines germplasm labels based on distribution status, inverts and calculates environmental compensation parameters, and generates cultivation management supervision instructions.

Images