Chela ratio prediction and breeding method of macrobrachium rosenbergii based on phenomics analysis

Abstract

The application discloses a method for predicting and breeding Macrobrachium rosenbergii chela ratio based on phenomics analysis, comprising the following steps: selecting a plurality of Macrobrachium rosenbergii for multi-generation breeding, recording pedigree data of each individual, and constructing a Macrobrachium rosenbergii breeding database; extracting the phenotypic characteristics of Macrobrachium rosenbergii at different breeding stages, analyzing the relationship between the growth index and the chela ratio, and evaluating the genetic force to construct a Macrobrachium rosenbergii phenomics network; analyzing the growth path of Macrobrachium rosenbergii, identifying abnormal events during breeding, and generating a Macrobrachium rosenbergii growth trend network; constructing a Macrobrachium rosenbergii phenotype knowledge graph, and constructing a Macrobrachium rosenbergii chela ratio prediction model; collecting early phenotype data of the Macrobrachium rosenbergii population to be bred, making breeding decisions, and recommending breeding paths for the breeding individuals by using the Macrobrachium rosenbergii phenotype knowledge graph. Whether the breeding path is deviated in the breeding process is analyzed at the same time, and a correction scheme is formulated and pushed, so that the breeding cycle of Macrobrachium rosenbergii is shortened, and the accuracy and efficiency of breeding are improved.

Description

A Method for Predicting and Breeding Giant Freshwater Prawn Cheliceroid Ratio Based on Phenomics Analysis Technical Field

[0001] This invention relates to the field of giant freshwater prawn breeding technology, and in particular to a method for predicting and breeding giant freshwater prawn claw-body ratio based on phenomics analysis. Background Technology

[0002] The giant freshwater prawn (Macrobrachium rosenbergii), a large freshwater shrimp of significant economic value, has seen widespread development in its aquaculture industry globally. However, a long-standing and difficult-to-eradicate technical bottleneck severely restricts further increases in yield and the sustainable development of the industry during large-scale farming: the intense cannibalistic behavior among individual giant freshwater prawns. This behavior directly leads to a significant decrease in survival rate, an increase in feed conversion ratio, and ultimately, substantial economic losses. Research shows that male giant freshwater prawns have abnormally developed chelipeds (second walking legs), with their length sometimes reaching twice their body length. This morphological feature is significantly positively correlated with their strong territoriality and aggressive behavior. In other words, the more developed the chelipeds of an individual, the stronger its aggression, and the greater the harm it causes to its own kind in high-density farming environments, which is the main cause of cannibalism.

[0003] Traditional breeding improvement methods typically rely on the experience of shrimp farmers, relying on manual selection based on a few readily observable phenotypic traits such as body size and weight in the later stages of cultivation. This method has fundamental flaws when it comes to reducing the complex trait of cheliped-to-body ratio (the ratio of cheliped length to body length). First, it prolongs the breeding cycle, requiring shrimp to reach sexual maturity and fully develop their chelipeds before accurate measurement and selection can be performed. This significantly extends the generation interval, severely slows down the breeding process, and increases breeding costs. Second, traditional methods suffer from low throughput and high subjectivity, making it difficult to conduct precise and efficient measurements on large populations. Furthermore, they cannot predict future cheliped growth trends in the early stages, resulting in low accuracy and efficiency in breeding. Therefore, the industry urgently needs an innovative technological solution that enables early, rapid, and precise selection of new Macrobrachium rosenbergii strains with low cheliped-to-body ratios and low aggression.

[0004] With the development of modern information technology, phenomics, big data analysis, and machine learning technologies have provided new solutions to this problem. Through automated, high-throughput image acquisition and processing technologies, massive amounts of morphological phenotypic data of shrimp throughout their entire growth cycle can be obtained non-destructively. Furthermore, by combining pedigree information and using quantitative genetics methods, the heritability of the chelicerae ratio trait can be accurately assessed, confirming its potential for genetic improvement. Therefore, this invention provides a phenomics-based method for predicting and breeding the chelicerae ratio of Macrobrachium rosenbergii, thereby shortening the shrimp breeding cycle, improving the accuracy and efficiency of breeding, and simultaneously assisting in the rearing of selected shrimp individuals. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method for predicting and breeding the claw-body ratio of Macrobrachium rosenbergii based on phenomics analysis.

[0006] To achieve the above objectives, the first aspect of this invention provides a method for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis, comprising:

[0007] Several giant freshwater prawns were selected for multi-generational culture. During the culture process, pedigree data of each individual was recorded, and phenotypic data and culture management data of the prawns at different culture stages were collected to construct a giant freshwater prawn culture database.

[0008] Phenotypic characteristics of prawns at different farming stages were extracted from the prawn farming database, the relationship between individual growth indicators and chelicerate ratio was analyzed, heritability was assessed, and a prawn phenomics network was constructed.

[0009] The growth path of giant freshwater prawns during the farming process is analyzed using the aforementioned prawn farming database. Abnormal events during the farming period are identified and corresponding treatment plans are associated with them, generating a prawn growth status network.

[0010] A phenotypic knowledge graph of the prawn was constructed by combining the prawn phenomics network and the prawn growth status network, and a prawn chelicerae ratio prediction model was constructed using the prawn phenotypic knowledge graph.

[0011] Early phenotypic data of the selected giant freshwater prawn population were collected. Breeding decisions were made using the giant freshwater prawn claw-to-body ratio prediction model. The giant freshwater prawn phenotypic knowledge graph was used to recommend farming paths for the selected individuals.

[0012] Acquire multimodal monitoring data of selected individuals during the breeding process, analyze whether the actual growth trajectory or environmental parameters deviate from the expected path, and if deviations are found, formulate and implement corrective measures.

[0013] In this scheme, several giant freshwater prawns are selected for multi-generation culture. During the culture process, pedigree data for each individual is recorded, and phenotypic data and culture management data of the prawns at different culture stages are collected to construct a giant freshwater prawn culture database. Specifically, this includes:

[0014] Several giant freshwater prawns with chelicerae ratios meeting the desired threshold were selected from existing culture ponds to form a first-generation culture population. Each giant freshwater prawn parent and offspring was assigned a unique electronic identifier, and a giant freshwater prawn culture pedigree was constructed through multiple generations of culture.

[0015] During the aquaculture process, an array of aquaculture monitoring sensors is deployed to periodically collect individual image data of prawns at different growth stages, environmental data of prawn farming ponds, and prawn management data to generate a raw multimodal dataset.

[0016] The obtained original multimodal dataset is preprocessed, and the collected individual images of prawns are imported into an image segmentation model built with a convolutional neural network to obtain individual prawn segmentation images. The morphological phenotypic data of the individuals are calculated according to the preset pixel-actual size conversion coefficient and bound to the corresponding individual ID and collection timestamp to generate a prawn phenotypic dataset.

[0017] For the environmental data and management data of prawn farming ponds, an outlier monitoring algorithm is used to clean the data, identify and remove abnormal measurement values, extract the environmental data fluctuation characteristics before and after the event point based on the timestamp of the management event recorded in the farming log, and generate a management tag for each event. The management tag includes management measures, changes in environmental parameters before and after management, and prawn status data, and finally generates a farming management dataset with event semantics.

[0018] A database of giant freshwater prawn (Gastropoda rosenbergii) farming was constructed by combining the pedigree of giant freshwater prawn farming, prawn phenotypic datasets, and farming management datasets.

[0019] In this scheme, the step of extracting phenotypic characteristics of prawns at different farming stages from the prawn farming database and analyzing the relationship between individual growth indicators and chelicerae ratio specifically includes:

[0020] Obtain the prawn farming database, extract complete time-series phenotypic data of all prawn individuals at various time points from the prawn farming database, generate prawn phenotypic measurement sequences, and use a sliding window of fixed time length to segment the prawn phenotypic measurement sequences to generate several prawn phenotypic measurement subsequences.

[0021] For each phenotypic measurement subsequence of freshwater prawns, the first-order difference between adjacent time points is calculated as the instantaneous growth rate characteristic, and the difference between the first-order difference sequences is calculated as the second-order difference to characterize the growth acceleration.

[0022] Meanwhile, cubic spline interpolation or polynomial fitting is performed on the complete growth curve of each phenotypic index, the coefficients of the fitting function are extracted as shape features describing the overall growth pattern, and the statistical features of each sequence are calculated, including mean, variance, skewness and kurtosis, and finally the phenotypic derived feature sequence of prawn is obtained.

[0023] A multimodal feature set characterizing individual growth dynamics was constructed by integrating the phenotypic measurement sequence and the phenotypic derived feature sequence of the prawn. The association analysis was performed by combining the final measured chelicerae ratio of each prawn individual, and the maximum mutual information coefficient and Pearson correlation coefficient between each feature and the chelicerae ratio were calculated.

[0024] Subsequently, a random forest regression model was invoked, with the final measured chelicerae ratio as the target variable and the phenotypic measurement features and extended phenotypic features of the prawn as decision variables. The importance score of each feature was calculated through 10-fold cross-validation. The key growth features and weight coefficients that are significantly related to the chelicerae ratio were analyzed. A Gaussian process regression model was established by selecting key growth features to fit the nonlinear mapping relationship between phenotypic features and target traits, and finally, a set of association rules between phenotypic features and chelicerae ratio was generated.

[0025] In this scheme, the genetic assessment and construction of the prawn phenomics network specifically include:

[0026] Obtain the phenotypic feature-chelicer ratio association rule set, extract key growth features and weight coefficients that are significantly related to the chelicer ratio from the phenotypic feature-chelicer ratio association rule set to construct a fixed-effects design matrix, extract the pedigree table from the prawn farming database, and use the pedigree table to construct a kinship matrix of all prawn individuals;

[0027] Based on the fixed effects design matrix and kinship matrix, the additive genetic effects of individual prawns are treated as random effects, and the kinship matrix is ​​defined as the variance-covariance matrix prior of the random effects to construct a mixed linear model based on an animal model.

[0028] The restricted maximum likelihood method was used to fit the mixed linear model. The variance components were solved iteratively by the EM algorithm to obtain the additive genetic variance component and residual variance component of the chelate ratio trait. The heritability estimate was obtained by calculating the ratio of additive genetic variance to phenotypic variance. The breeding value of each individual was solved by the BLUP method.

[0029] Based on the phenotypic feature-chelicere ratio association rule set, key growth features are extracted as network nodes. The dynamic time warping algorithm is used to calculate the phenotypic distance matrix between individuals to represent the asynchronous growth trajectory of individual prawns. The graphicallasso algorithm is used to estimate the conditional independence relationship between features and to construct a sparse Gaussian graph model.

[0030] The sparse Gaussian graph model is defined as the basic topology of the phenomics network, where nodes represent phenotypic features and edge weights represent the conditional correlation coefficients between features. The association strength between each phenotypic feature and the cheliceroid ratio is extracted as the initial attribute of the node through the phenotypic feature-cheliceroid ratio association rule set. The heritability estimate is used as the global attribute of the network, and the individual breeding value is used as the additional attribute of the corresponding individual phenotypic node to generate the initial phenomics network.

[0031] The Node2Vec algorithm is used to learn the low-dimensional vector representation of the nodes in the initial phenomics network. The transition probabilities between nodes are defined. Node embedding vectors are learned by generating node sequences through biased random walks. Finally, the t-SNE dimensionality reduction algorithm is used to map the high-dimensional node embeddings to a two-dimensional space to generate the prawn phenomics network.

[0032] In this solution, the step of analyzing the growth path of the giant freshwater prawn (Macrobrachium rosenbergii) during the farming process using the prawn farming database, identifying abnormal events during the farming period and associating them with corresponding treatment plans, and generating a prawn growth status network specifically includes:

[0033] A database of giant freshwater prawns was obtained. The time-series phenotypic measurement data and corresponding environmental parameter sequences of all giant freshwater prawns were extracted from the database. After preprocessing, a nonlinear mixed-effects model was used, with individuals as random effects, to fit the standard growth curve of giant freshwater prawns and establish an individual growth curve model describing the normal growth pattern.

[0034] Based on the established individual growth curve model, the residual sequence between the actual measured value and the model predicted value of each individual is calculated, and wavelet transform analysis is performed on the residual sequence to extract time-frequency domain features and generate a set of individual growth trajectories of prawns; combined with the environmental parameter sequence, the isolated forest algorithm is used to detect abnormal events in the growth process, and the occurrence time, event characteristics, duration and abnormal intensity index of abnormal events are recorded to generate a set of abnormal growth event features;

[0035] Based on the feature set of abnormal growth events, the identified abnormal event points are matched with the control log records in the aquaculture control dataset by timestamp. If a certain abnormal event point has a manually recorded control operation, the corresponding abnormal event is associated with the control measures to form an abnormal event-control plan association pair. Otherwise, if it is an environmental self-fluctuation event, an abnormal event-self-fluctuation association pair is generated and marked as an event that does not require control.

[0036] Based on the obtained association pairs, an anomaly-handling mapping knowledge base is constructed with the association order of anomaly event characteristics, environmental parameter characteristics, and handling solutions. It includes anomaly event types, environmental parameter characteristics, handling measures taken, and the effects after handling.

[0037] The growth trajectory features of each individual prawn are extracted from the prawn individual growth trajectory set to generate several growth trajectory feature sequences with time-series attributes. Based on the anomaly-handling mapping knowledge base, the prawn individual identifier corresponding to the abnormal event and the timestamp of the abnormal event are merged into the corresponding growth trajectory feature sequence to generate a sequence representing the growth status of each prawn individual in the farming process.

[0038] The cosine similarity and Mahalanobis distance between each growth status sequence are calculated separately, and weighted average and normalization are performed to generate a sequence merging index. This index is then compared with a preset threshold. If the index is greater than the preset threshold, the difference between the two growth status sequences is calculated to identify the difference sequence segments. New sequence branches are generated in the difference sequence segments for sequence merging. Through repeated iterative merging steps, the prawn growth status network is finally generated.

[0039] In this scheme, the construction of a prawn phenotypic knowledge graph by combining the prawn phenomics network and the prawn growth status network, and the use of the prawn phenotypic knowledge graph to construct a prawn claw-to-body ratio prediction model, specifically includes:

[0040] Obtain the phenomics network and growth status network of the prawn. Extract all phenotypic feature nodes and node attribute information from the prawn phenomics network, including the correlation strength between growth characteristics and chelicerate ratio, heritability estimate and individual breeding value. Extract abnormal event nodes, treatment plan nodes and their correlation relationships from the prawn growth status network.

[0041] Based on the extracted phenotypic feature nodes, the phenotypic change sequence during the growth process of individual prawns is used as the node connection order to generate the phenotypic node sequence main axis. The phenotypic node sequence main axis represents all phenotypic change patterns of individual prawns during the growth process. The correlation between abnormal event nodes and treatment plan nodes is used as the connection basis to generate node connection edges to connect abnormal event nodes and treatment plan nodes, thus generating node sequence sub-axis.

[0042] Based on the node sequence sub-axis, several tree-like node sequences are obtained by merging with the node sequence main axis using the trigger event stamp of the abnormal event as the index. The tree-like node sequences represent the phenotypic change patterns and growth status of individual prawns during the farming process. Finally, a prawn phenotypic knowledge graph is constructed based on all the tree-like node sequences.

[0043] The TransE knowledge identifier learning algorithm is used to learn low-dimensional vector representations of entities and relationships in the phenotypic knowledge graph of the giant freshwater prawn and construct a training dataset. A giant freshwater prawn claw-to-body ratio prediction model is built based on a graph neural network architecture, and the model is trained using the training dataset. Hyperparameters are optimized through cross-validation and grid search optimization, and finally, a giant freshwater prawn claw-to-body ratio prediction model that meets the expectations is obtained.

[0044] In this scheme, the process of collecting early phenotypic data from the proposed Macrobrachium rosenbergii population, making breeding decisions using the Macrobrachium rosenbergii claw-to-body ratio prediction model, and recommending farming pathways for selected individuals using the Macrobrachium rosenbergii phenotypic knowledge graph, specifically includes:

[0045] When raising the population of giant freshwater prawns to be bred, morphological phenotypic data of each individual is acquired through an automated image acquisition system, and environmental parameters of the breeding environment are monitored and collected simultaneously to obtain breeding environment monitoring data.

[0046] The collected early phenotypic data and aquaculture environment monitoring data are preprocessed and standardized to form early monitoring feature vectors of individuals to be selected for breeding. The early monitoring feature vectors are input into a pre-trained prawn claw-to-body ratio prediction model. Through multi-layer nonlinear transformation and feature extraction within the model, the predicted value of the claw-to-body ratio after adulthood and the corresponding prediction confidence interval are output.

[0047] Based on the predicted chelicerae ratio of all individuals to be bred, they are sorted according to the predicted chelicerae ratio. Combined with the prediction confidence interval and the preset selection intensity threshold, individuals of prawns whose chelicerae ratio meets the preset expectation are selected as candidate parents, and a prawn breeding recommendation table is generated.

[0048] Based on the recommended breeding table for freshwater prawns, and combined with the prawn phenotypic knowledge graph, the breeding path is analyzed. Using the early monitoring feature vectors and pedigree information of the individuals to be bred as query conditions, historical individual nodes with similar phenotypic characteristics and genetic backgrounds are retrieved in the knowledge graph. The optimal environmental parameter range, successful control measures sequence, and final growth performance indicators experienced by the historical individual nodes in the complete breeding cycle are obtained, and similar breeding cases are generated.

[0049] Based on similar breeding cases retrieved, case reasoning technology is used to generate personalized breeding path recommendation schemes for target recommended breeding individuals, including environmental parameter control targets, feeding strategies, density management requirements, and expected abnormal events and corresponding measures for different growth stages;

[0050] The prawn breeding recommendation form is linked with the personalized aquaculture path recommendation plan to generate a final prawn breeding decision report, which is then pushed out.

[0051] In this plan, the acquisition of multimodal monitoring data of selected individuals during the breeding process is analyzed to determine whether the actual growth trajectory or environmental parameters deviate from the expected path. If deviations are found, a correction plan is developed and implemented. Specifically, this includes:

[0052] Real-time acquisition of multimodal monitoring data of selected individuals during the breeding process, including time-series environmental parameters collected by sensor arrays and individual phenotypic measurement data acquired periodically by image acquisition system;

[0053] The multimodal monitoring data is compared with the expected value of the preset breeding path recommended for the individual in the prawn phenotypic knowledge graph. The similarity distance between the actual growth sequence and the expected growth sequence is calculated using the dynamic time warping algorithm, and the deviation threshold is set based on the statistical process control method.

[0054] When the actual growth trajectory or environmental parameters are detected to continuously deviate from the expected path and exceed the deviation threshold, an early warning mechanism is triggered; based on the deviation characteristics, the abnormal-treatment mapping knowledge base in the prawn growth status network is queried to match historical similar deviation patterns and their corresponding effective treatment solutions; at the same time, the genetic background information and phenotypic feature association rules of the individual are retrieved from the prawn phenotypic knowledge graph to assess the potential impact of specific treatment measures on the genetic strain.

[0055] Based on the combined effects of historical treatment plans and individual genetic characteristics, a personalized correction plan is generated for the current deviation. The correction plan includes environmental parameter adjustment strategies, suggestions for adjusting feeding and management measures, and expected recovery trajectory. The correction plan is then pushed to the aquaculture management system and an early warning is triggered simultaneously to guide manual or automatic control operations.

[0056] After implementing the corrective measures, we continue to monitor the growth response of individuals and changes in environmental parameters, compare the actual recovery trajectory with the expected recovery trajectory, and evaluate the effectiveness of the corrective measures. We use the deviation event, the corrective measures taken, and the final effect as new knowledge samples, and feed them back to update the prawn growth status network and prawn phenotypic knowledge graph to achieve continuous optimization and learning of the knowledge base.

[0057] In another aspect, the present invention provides a computer-readable storage medium comprising a program for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis. When the program for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis is executed by a processor, it implements the steps of the method for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis as described in any of the preceding claims. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples 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 according to these drawings without creative effort.

[0059] Figure 1 is a flowchart of the first method of a method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to an embodiment of the present invention.

[0060] Figure 2 is a flowchart of the second method of a method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to an embodiment of the present invention.

[0061] Figure 3 is a flowchart of the third method of a method for predicting and breeding the claw ratio of giant freshwater prawn based on phenomics analysis according to an embodiment of the present invention.

[0062] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0063] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0064] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0065] Figure 1 is a flowchart of the first method of a method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to an embodiment of the present invention.

[0066] As shown in Figure 1, this invention provides a first method flowchart for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis, including:

[0067] S102, Select a number of giant freshwater prawns for multi-generation farming, record the pedigree data of each individual during the farming process, and collect phenotypic data and farming management data of giant freshwater prawns at different farming stages to construct a giant freshwater prawn farming database.

[0068] S104. Phenotypic characteristics of prawns at different farming stages are extracted from the prawn farming database, the relationship between individual growth indicators and chelicerate ratio is analyzed, heritability is assessed, and a prawn phenomics network is constructed.

[0069] S106, Analyze the growth path of giant freshwater prawns during the farming process using the prawn farming database, identify abnormal events during the farming period and associate them with corresponding treatment plans, and generate a prawn growth status network.

[0070] S108, Combine the above-mentioned prawn phenomics network and prawn growth status network to construct a prawn phenotypic knowledge graph, and use the prawn phenotypic knowledge graph to construct a prawn claw-body ratio prediction model.

[0071] S110, Collect early phenotypic data of the giant freshwater prawn population to be bred, make bred selection decisions through the giant freshwater prawn claw-to-body ratio prediction model, and recommend farming paths for selected individuals using the giant freshwater prawn phenotypic knowledge graph.

[0072] S112: Obtain multimodal monitoring data of selected individuals during the breeding process, analyze whether the actual growth trajectory or environmental parameters deviate from the expected path, and if there is a deviation, formulate a correction plan and push it out.

[0073] Furthermore, in a preferred embodiment of the present invention, the step of selecting several giant freshwater prawns for multi-generational culture, recording the pedigree data of each individual during the culture process, and collecting prawn phenotypic data and culture management data at different culture stages to construct a giant freshwater prawn culture database specifically includes:

[0074] Several giant freshwater prawns with chelicerae ratios meeting the desired threshold were selected from existing culture ponds to form a first-generation culture population. Each giant freshwater prawn parent and offspring was assigned a unique electronic identifier, and a giant freshwater prawn culture pedigree was constructed through multiple generations of culture.

[0075] During the aquaculture process, an array of aquaculture monitoring sensors is deployed to periodically collect individual image data of prawns at different growth stages, environmental data of prawn farming ponds, and prawn management data to generate a raw multimodal dataset.

[0076] The obtained original multimodal dataset is preprocessed, and the collected individual images of prawns are imported into an image segmentation model built with a convolutional neural network to obtain individual prawn segmentation images. The morphological phenotypic data of the individuals are calculated according to the preset pixel-actual size conversion coefficient and bound to the corresponding individual ID and collection timestamp to generate a prawn phenotypic dataset.

[0077] For the environmental data and management data of prawn farming ponds, an outlier monitoring algorithm is used to clean the data, identify and remove abnormal measurement values, extract the environmental data fluctuation characteristics before and after the event point based on the timestamp of the management event recorded in the farming log, and generate a management tag for each event. The management tag includes management measures, changes in environmental parameters before and after management, and prawn status data, and finally generates a farming management dataset with event semantics.

[0078] A database of giant freshwater prawn (Gastropoda rosenbergii) farming was constructed by combining the pedigree of giant freshwater prawn farming, prawn phenotypic datasets, and farming management datasets.

[0079] It should be noted that, within existing aquaculture ponds, individuals with superior phenotypic potential are selected based on clearly defined phenotypic criteria—namely, the cheliped-to-body ratio (the ratio of cheliped length to body length) meeting a pre-set expected threshold—to form the core basic breeding population. To ensure the uniqueness of subsequent data traceability and individual identification, each selected parent and its offspring are assigned a unique electronic identifier. Throughout the multi-generational aquaculture process, an integrated monitoring sensor array is systematically deployed in the aquaculture environment. This array periodically captures multimodal raw data: images of individual prawns at different growth stages (e.g., larval, postlarval, and adult stages) are acquired using high-definition industrial cameras; environmental parameters in the aquaculture ponds are continuously recorded using water quality sensors (e.g., temperature, pH, dissolved oxygen, and ammonia nitrogen sensors); and all human intervention operations (e.g., feeding, water changes, pond separation, medication, etc.) are recorded through a digital log system, collectively forming the raw multimodal dataset. Preprocessing of the above raw dataset is a crucial step in ensuring data quality and usability. Individual image data is imported into an image segmentation model pre-trained using a convolutional neural network (such as U-Net or Mask R-CNN architecture). This model can accurately identify and segment key morphological parts of the shrimp in the image, such as the body outline and claws. After segmentation, precise morphological phenotypic data, such as body length and claw length, are automatically calculated based on pixel-to-actual-size conversion coefficients determined in advance using a calibration board. These quantified data are then strictly bound to the corresponding unique individual ID and collection timestamp, thereby generating a structured shrimp phenotypic dataset with temporal characteristics.

[0080] For continuously monitored environmental data and manually recorded control data, outlier monitoring algorithms (such as isolated forest or Z-Score algorithms) are used for data cleaning to identify and remove abnormal measurements caused by momentary sensor failures or human recording errors, ensuring the reliability of the data sequence. Furthermore, based on the timestamps of control events recorded in the aquaculture logs, environmental data fluctuation characteristics (such as temperature change curves and dissolved oxygen recovery rates) within a specific time window before and after the event are extracted, and a structured control tag is generated for each event. This tag not only records the control measures themselves but also encapsulates the dynamic changes in environmental parameters before and after control, as well as the observed prawn population status data at that time (such as feeding behavior and activity levels), ultimately generating an aquaculture control dataset rich in contextual semantics. Finally, through data fusion technology, the three types of processed data—the prawn pedigree revealing genetic associations, the prawn phenotypic dataset recording individual growth history, and the aquaculture control dataset describing environmental and management interventions—are integrated using individual IDs and timestamps as key associations to construct a spatiotemporally synchronized, structurally standardized, and information-complete prawn aquaculture database.

[0081] Furthermore, in a preferred embodiment of the present invention, the step of extracting phenotypic characteristics of prawns at different farming stages from the prawn farming database and analyzing the relationship between individual growth indicators and chelicerae ratio specifically includes:

[0082] Obtain the prawn farming database, extract complete time-series phenotypic data of all prawn individuals at various time points from the prawn farming database, generate prawn phenotypic measurement sequences, and use a sliding window of fixed time length to segment the prawn phenotypic measurement sequences to generate several prawn phenotypic measurement subsequences.

[0083] For each phenotypic measurement subsequence of freshwater prawns, the first-order difference between adjacent time points is calculated as the instantaneous growth rate characteristic, and the difference between the first-order difference sequences is calculated as the second-order difference to characterize the growth acceleration.

[0084] Meanwhile, cubic spline interpolation or polynomial fitting is performed on the complete growth curve of each phenotypic index, the coefficients of the fitting function are extracted as shape features describing the overall growth pattern, and the statistical features of each sequence are calculated, including mean, variance, skewness and kurtosis, and finally the phenotypic derived feature sequence of prawn is obtained.

[0085] A multimodal feature set characterizing individual growth dynamics was constructed by integrating the phenotypic measurement sequence and the phenotypic derived feature sequence of the prawn. The association analysis was performed by combining the final measured chelicerae ratio of each prawn individual, and the maximum mutual information coefficient and Pearson correlation coefficient between each feature and the chelicerae ratio were calculated.

[0086] Subsequently, a random forest regression model was invoked, with the final measured chelicerae ratio as the target variable and the phenotypic measurement features and extended phenotypic features of the prawn as decision variables. The importance score of each feature was calculated through 10-fold cross-validation. The key growth features and weight coefficients that are significantly related to the chelicerae ratio were analyzed. A Gaussian process regression model was established by selecting key growth features to fit the nonlinear mapping relationship between phenotypic features and target traits, and finally, a set of association rules between phenotypic features and chelicerae ratio was generated.

[0087] It should be noted that complete phenotypic measurement data of all individuals throughout the entire farming cycle were extracted from the established prawn farming database, forming a prawn phenotypic measurement sequence for each individual. To capture the dynamic characteristics of different growth stages, the sequence was segmented using a sliding window of fixed time length, generating a series of prawn phenotypic measurement subsequences that reflect local growth characteristics. For each phenotypic measurement subsequence, its instantaneous growth rate was quantified by calculating the first-order difference between measurements at adjacent time points. The second-order difference was then calculated from the first-order difference sequence to characterize the acceleration change pattern of growth. Simultaneously, for the complete growth trajectory of each phenotypic index (such as body length and cheliped length), cubic spline interpolation was used for smoothing or polynomial functions were used for trend approximation. The coefficients of the fitted function were extracted as shape features describing the overall growth morphology. Furthermore, statistical characteristics of each sequence were calculated, including the mean reflecting the average level, variance representing the fluctuation range, skewness indicating the symmetry of the distribution, and kurtosis measuring the steepness of the distribution, collectively constituting the prawn phenotypic derived feature sequence. By integrating the original measurement sequences and derived feature sequences, a multimodal feature set capable of comprehensively characterizing individual growth dynamics was constructed. This feature set was then correlated with the final measured chelicerae ratio for each individual. The maximum mutual information coefficient was used to capture the nonlinear correlation strength between the feature and the target trait, while the Pearson correlation coefficient was calculated to measure the degree of linear correlation, thus initially screening out feature indicators closely related to the chelicerae ratio. Based on the above analysis results, a random forest regression model was used for in-depth feature screening and modeling. Using the final chelicerae ratio as the target variable and all phenotypic measurement features and derived features as decision variables, ten-fold cross-validation was used to evaluate the importance score of each feature, thereby identifying key growth features significantly related to the chelicerae ratio and their corresponding weight coefficients. Finally, based on the screened key features, a Gaussian process regression model was established to accurately fit the complex nonlinear mapping relationship between phenotypic features and the target trait. This model can provide an estimate of the uncertainty of the prediction results, ultimately generating a set of phenotypic feature-chelicerae ratio association rules containing feature-trait association rules and their confidence levels, providing a quantitative basis for subsequent genetic evaluation and breeding decisions.

[0088] Furthermore, in a preferred embodiment of the present invention, the step of performing heritability assessment and constructing a prawn phenomics network specifically includes:

[0089] Obtain the phenotypic feature-chelicer ratio association rule set, extract key growth features and weight coefficients that are significantly related to the chelicer ratio from the phenotypic feature-chelicer ratio association rule set to construct a fixed-effects design matrix, extract the pedigree table from the prawn farming database, and use the pedigree table to construct a kinship matrix of all prawn individuals;

[0090] Based on the fixed effects design matrix and kinship matrix, the additive genetic effects of individual prawns are treated as random effects, and the kinship matrix is ​​defined as the variance-covariance matrix prior of the random effects to construct a mixed linear model based on an animal model.

[0091] The restricted maximum likelihood method was used to fit the mixed linear model. The variance components were solved iteratively by the EM algorithm to obtain the additive genetic variance component and residual variance component of the chelate ratio trait. The heritability estimate was obtained by calculating the ratio of additive genetic variance to phenotypic variance. The breeding value of each individual was solved by the BLUP method.

[0092] Based on the phenotypic feature-chelicere ratio association rule set, key growth features are extracted as network nodes. The dynamic time warping algorithm is used to calculate the phenotypic distance matrix between individuals to represent the asynchronous growth trajectory of individual prawns. The graphicallasso algorithm is used to estimate the conditional independence relationship between features and to construct a sparse Gaussian graph model.

[0093] The sparse Gaussian graph model is defined as the basic topology of the phenomics network, where nodes represent phenotypic features and edge weights represent the conditional correlation coefficients between features. The association strength between each phenotypic feature and the cheliceroid ratio is extracted as the initial attribute of the node through the phenotypic feature-cheliceroid ratio association rule set. The heritability estimate is used as the global attribute of the network, and the individual breeding value is used as the additional attribute of the corresponding individual phenotypic node to generate the initial phenomics network.

[0094] The Node2Vec algorithm is used to learn the low-dimensional vector representation of the nodes in the initial phenomics network. The transition probabilities between nodes are defined. Node embedding vectors are learned by generating node sequences through biased random walks. Finally, the t-SNE dimensionality reduction algorithm is used to map the high-dimensional node embeddings to a two-dimensional space to generate the prawn phenomics network.

[0095] It should be noted that, utilizing the previously obtained phenotypic feature-chelicerae ratio association rule set, key growth features significantly correlated with the chelicerae ratio and their weight coefficients were extracted to construct a fixed-effects design matrix for genetic assessment. Simultaneously, a complete pedigree information table was extracted from the prawn farming database. Based on the pedigree data, a kinship coefficient calculation rule was used to construct a kinship matrix for all individuals, which quantifies the genetic similarity among individuals within the population. Based on the fixed-effects design matrix and the kinship matrix, a mixed linear model based on an animal model was constructed. In this model, the selected key phenotypic features were used as fixed effects, the additive genetic effects of individuals were used as random effects, and the kinship matrix was defined as the variance-covariance matrix prior of the random effects to accurately characterize the genetic correlation structure. The restricted maximum likelihood method was used to fit the model, and the variance components were iteratively solved using the EM algorithm to obtain the additive genetic variance component and residual variance component of the chelicerae ratio trait. By calculating the ratio of additive genetic variance to total phenotypic variance, an accurate estimate of the heritability of the trait is obtained. At the same time, the optimal linear unbiased prediction method is used to solve the breeding value of each individual, providing a quantitative basis for subsequent breeding.

[0096] Based on the obtained genetic parameters, a phenomics network is further constructed to achieve visualization and analysis of multidimensional data. Key growth features are extracted from the association rule set as network nodes, and the dynamic time warping algorithm is used to calculate the phenotypic distance matrix between individuals, effectively representing the similarity of different individuals in asynchronous growth trajectories. Simultaneously, the graphicallasso algorithm is used to estimate the conditional independence relationships between features, constructing a sparse Gaussian graph model to reveal the direct correlations between features. The sparse Gaussian graph model is used as the basic network topology, where nodes represent phenotypic features and edge weights represent the conditional correlation coefficients between features. The association strength between each feature and the chelicerate ratio is extracted from the association rule set as node attributes, the heritability estimate is used as the global attribute of the network, and the individual breeding value is used as the additional attribute of the corresponding individual phenotypic node, generating an initial phenomics network rich in multidimensional information. Finally, the Node2Vec algorithm is used to learn the low-dimensional vector representation of network nodes, a biased random walk strategy is used to generate node sequences, the Skip-gram model is used to learn node embedding vectors, and the t-SNE dimensionality reduction algorithm is used to map the high-dimensional embeddings to a two-dimensional space, ultimately generating a visualized prawn phenomics network. This network effectively integrates the relationships between phenotypic traits, genetic parameter information, and individual breeding values, providing an intuitive, multi-dimensional analysis tool for breeding decisions.

[0097] Furthermore, in a preferred embodiment of the present invention, the step of analyzing the growth path of the giant freshwater prawn during the farming process using the prawn farming database, identifying abnormal events during the farming period and associating them with corresponding treatment plans, and generating a prawn growth status network specifically includes:

[0098] A database of giant freshwater prawns was obtained. The time-series phenotypic measurement data and corresponding environmental parameter sequences of all giant freshwater prawns were extracted from the database. After preprocessing, a nonlinear mixed-effects model was used, with individuals as random effects, to fit the standard growth curve of giant freshwater prawns and establish an individual growth curve model describing the normal growth pattern.

[0099] Based on the established individual growth curve model, the residual sequence between the actual measured value and the model predicted value of each individual is calculated, and wavelet transform analysis is performed on the residual sequence to extract time-frequency domain features and generate a set of individual growth trajectories of prawns; combined with the environmental parameter sequence, the isolated forest algorithm is used to detect abnormal events in the growth process, and the occurrence time, event characteristics, duration and abnormal intensity index of abnormal events are recorded to generate a set of abnormal growth event features;

[0100] Based on the feature set of abnormal growth events, the identified abnormal event points are matched with the control log records in the aquaculture control dataset by timestamp. If a certain abnormal event point has a manually recorded control operation, the corresponding abnormal event is associated with the control measures to form an abnormal event-control plan association pair. Otherwise, if it is an environmental self-fluctuation event, an abnormal event-self-fluctuation association pair is generated and marked as an event that does not require control.

[0101] Based on the obtained association pairs, an anomaly-handling mapping knowledge base is constructed with the association order of anomaly event characteristics, environmental parameter characteristics, and handling solutions. It includes anomaly event types, environmental parameter characteristics, handling measures taken, and the effects after handling.

[0102] The growth trajectory features of each individual prawn are extracted from the prawn individual growth trajectory set to generate several growth trajectory feature sequences with time-series attributes. Based on the anomaly-handling mapping knowledge base, the prawn individual identifier corresponding to the abnormal event and the timestamp of the abnormal event are merged into the corresponding growth trajectory feature sequence to generate a sequence representing the growth status of each prawn individual in the farming process.

[0103] The cosine similarity and Mahalanobis distance between each growth status sequence are calculated separately, and weighted average and normalization are performed to generate a sequence merging index. This index is then compared with a preset threshold. If the index is greater than the preset threshold, the difference between the two growth status sequences is calculated to identify the difference sequence segments. New sequence branches are generated in the difference sequence segments for sequence merging. Through repeated iterative merging steps, the prawn growth status network is finally generated.

[0104] It should be noted that, firstly, time-series phenotypic measurement data (such as body length, claw length, weight, etc.) and their corresponding environmental parameter sequences (such as water temperature, dissolved oxygen, pH, etc.) of all individuals were extracted from the established giant freshwater prawn (Gastropoda rosenbergii) farming database. After preprocessing (including missing value imputation, outlier removal, and standardization), a nonlinear mixed-effects model was used for modeling. This model uses individuals as random effect terms, which can capture growth variations among individuals and fit a standard growth curve of giant freshwater prawns, thereby establishing an individual growth curve model describing the normal growth pattern and providing a benchmark reference for subsequent analysis. Based on this model, the residual sequence between the actual measured values ​​and the model predicted values ​​of each individual was calculated. These residuals reflect the degree to which the individual's growth deviates from the normal pattern. Subsequently, wavelet transform analysis was performed on the residual sequence to extract time-frequency domain features (such as energy distribution and frequency components), thereby generating a set of individual growth trajectories of giant freshwater prawns. This set quantifies the dynamic growth characteristics of each individual. Simultaneously, by combining environmental parameter sequences, the isolated forest algorithm is used to detect abnormal events during the growth process, identify anomalous time segments that significantly deviate from the environmental pattern, and record the occurrence time, event characteristics (such as sudden temperature changes, dissolved oxygen anomalies), duration, and intensity of these anomalous events, forming an anomalous growth event feature set. Next, these identified anomalous event points are matched with the management log records in the aquaculture management dataset using timestamps: if a manually recorded management operation (such as oxygenation, water change, or medication) exists for an anomalous event point, the corresponding anomalous event is associated with the management measure, forming an "abnormal event-treatment plan" association pair; otherwise, it is marked as an environmental self-fluctuation event, generating an "abnormal event-self-fluctuation" association pair and designated as an event requiring no treatment. Based on these association pairs, an anomalous-treatment mapping knowledge base is constructed, with the anomalous event characteristics, environmental parameter characteristics, and treatment plans as the association order. This knowledge base includes the anomalous event type, environmental parameter characteristics, the treatment measures taken, and the effects after treatment (such as recovery time and growth response), providing historical experience support for subsequent decision-making.

[0105] Furthermore, growth trajectory features (such as growth rate, acceleration, and fluctuation patterns) of individual prawns are extracted from the prawn growth trajectory set to generate several growth trajectory feature sequences with temporal attributes. Then, based on the anomaly-disposal mapping knowledge base, the prawn individual identifiers and occurrence timestamps corresponding to abnormal events are merged into the corresponding growth trajectory feature sequences, thereby generating sequences representing the complete growth status of each prawn individual during the farming process. These sequences integrate multi-dimensional information of normal growth, abnormal events, and disposal responses. Finally, the cosine similarity (measuring sequence shape similarity) and Mahalanobis distance (measuring sequence statistical distribution differences) between each growth status sequence are calculated, and weighted averages and normalization are performed to generate a sequence merging index. This index is compared with a preset threshold. If it is greater than the threshold, the difference between the two growth status sequences is calculated, and the difference sequence segments (such as event response segments) are identified. New sequence branches are generated in the difference sequence segments for sequence merging. By repeating this merging step iteratively, a prawn growth status network is finally generated. This network visually demonstrates the similarity of different individuals or groups during the growth process, the impact path of abnormal events, and the effectiveness of disposal measures, providing data-driven insights for optimizing farming strategies.

[0106] Furthermore, in a preferred embodiment of the present invention, the process of collecting early phenotypic data from the Macrobrachium rosenbergii population to be bred, making breding decisions using the Macrobrachium rosenbergii claw-to-body ratio prediction model, and recommending farming paths for selected individuals using the Macrobrachium rosenbergii phenotypic knowledge graph specifically includes:

[0107] When raising the population of giant freshwater prawns to be bred, morphological phenotypic data of each individual is acquired through an automated image acquisition system, and environmental parameters of the breeding environment are monitored and collected simultaneously to obtain breeding environment monitoring data.

[0108] The collected early phenotypic data and aquaculture environment monitoring data are preprocessed and standardized to form early monitoring feature vectors of individuals to be selected for breeding. The early monitoring feature vectors are input into a pre-trained prawn claw-to-body ratio prediction model. Through multi-layer nonlinear transformation and feature extraction within the model, the predicted value of the claw-to-body ratio after adulthood and the corresponding prediction confidence interval are output.

[0109] Based on the predicted chelicerae ratio of all individuals to be bred, they are sorted according to the predicted chelicerae ratio. Combined with the prediction confidence interval and the preset selection intensity threshold, individuals of prawns whose chelicerae ratio meets the preset expectation are selected as candidate parents, and a prawn breeding recommendation table is generated.

[0110] Based on the recommended breeding table for freshwater prawns, and combined with the prawn phenotypic knowledge graph, the breeding path is analyzed. Using the early monitoring feature vectors and pedigree information of the individuals to be bred as query conditions, historical individual nodes with similar phenotypic characteristics and genetic backgrounds are retrieved in the knowledge graph. The optimal environmental parameter range, successful control measures sequence, and final growth performance indicators experienced by the historical individual nodes in the complete breeding cycle are obtained, and similar breeding cases are generated.

[0111] Based on similar breeding cases retrieved, case reasoning technology is used to generate personalized breeding path recommendation schemes for target recommended breeding individuals, including environmental parameter control targets, feeding strategies, density management requirements, and expected abnormal events and corresponding measures for different growth stages;

[0112] The prawn breeding recommendation form is linked with the personalized aquaculture path recommendation plan to generate a final prawn breeding decision report, which is then pushed out.

[0113] It should be noted that when raising the selected population, the first step is to periodically acquire key morphological phenotypic data for each individual, including body length, claw length, and weight, using an automated image acquisition system (typically containing high-definition industrial cameras and image processing units). Simultaneously, an IoT sensor network is used to monitor and collect environmental parameters (such as water temperature, pH, dissolved oxygen, and ammonia nitrogen concentration), obtaining environmental monitoring data corresponding to the phenotypic data in time and space. This early, multi-source, heterogeneous data is preprocessed and standardized, including data cleaning, outlier removal, missing value imputation, and feature scaling, forming standardized early monitoring feature vectors for the selected individuals. This provides high-quality input for subsequent predictive analysis. The early monitoring feature vectors are then input into a pre-trained prawn claw-to-body ratio prediction model. Through multi-layered nonlinear transformations and feature extraction mechanisms within the model, the model outputs the predicted claw-to-body ratio of the adult individual and its corresponding confidence interval. This confidence interval quantifies the degree of uncertainty in the prediction result. Based on the prediction results of all individuals to be bred, they are sorted according to the predicted chelicer-to-body ratio. Combined with the prediction confidence interval and the preset selection intensity threshold (such as the selection ratio of the top 10%), excellent individuals with chelicer-to-body ratios that meet the preset expected values ​​are selected as candidate parents, thereby generating a prawn breeding recommendation table containing individual number, predicted value and priority.

[0114] Subsequently, based on the breeding recommendation table, and combined with the prawn phenotypic knowledge graph, in-depth aquaculture path analysis was conducted. Using early monitoring feature vectors and pedigree information (such as family origin and parental traits) of the individuals to be bred as query conditions, historical individual nodes with similar phenotypic characteristics and genetic backgrounds were retrieved from the knowledge graph. This yielded the optimal environmental parameter ranges experienced by these historical individuals throughout their complete aquaculture cycle, the successful control measures sequences, and the final growth performance indicators achieved, thereby generating a valuable set of similar aquaculture case studies. Based on the retrieved similar aquaculture case studies, case-based reasoning technology was used to generate personalized aquaculture path recommendation schemes for the target recommended breeding individuals. These schemes detailed the environmental parameter control targets for different growth stages (such as suitable temperature ranges and dissolved oxygen thresholds), feeding strategies (feed type, feeding frequency and amount), density management requirements (aquaculture density adjustment scheme), and anticipated abnormal events (such as stress responses and disease risks) and corresponding countermeasures, forming a complete and precise aquaculture guidance scheme. Finally, the prawn breeding recommendation form is linked and integrated with the personalized aquaculture path recommendation scheme to generate a structured final prawn breeding decision report, which is delivered to aquaculture managers through a visual interface or automated push system, providing data-driven decision support for actual breeding operations.

[0115] Figure 2 is a flowchart of the second method of a method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to an embodiment of the present invention.

[0116] As shown in Figure 2, this invention provides a second method flowchart for predicting and breeding the claw-body ratio of Macrobrachium rosenbergii based on phenomics analysis, including:

[0117] S202, obtain the prawn phenomics network and the prawn growth status network, extract all phenotypic feature nodes and node attribute information from the prawn phenomics network, including the correlation strength between growth characteristics and chelicerate ratio, heritability estimate and individual breeding value, and extract abnormal event nodes, treatment plan nodes and their correlation relationships from the prawn growth status network.

[0118] S204, Based on the extracted phenotypic feature nodes, the phenotypic change sequence during the growth process of individual prawns is used as the node connection order to generate the phenotypic node sequence main axis, which represents all phenotypic change patterns of individual prawns during the growth process; the correlation between abnormal event nodes and treatment plan nodes is used as the connection basis to generate node connection edges to connect abnormal event nodes and treatment plan nodes, thereby generating node sequence sub-axis;

[0119] S206, according to the node sequence sub-axis, the node sequence trigger event stamp is used as the index to merge with the node sequence main axis to obtain several tree-like node sequences. The tree-like node sequences represent the phenotypic change patterns and growth status of individual prawns during the farming process. Finally, a prawn phenotypic knowledge graph is constructed based on all the tree-like node sequences.

[0120] S208, the TransE knowledge identifier learning algorithm is used to learn low-dimensional vector representations of entities and relationships in the phenotypic knowledge graph of the prawn and construct a training dataset. A prawn claw-to-body ratio prediction model is built based on a graph neural network architecture, and the model is trained using the training dataset. Hyperparameters are optimized through cross-validation and grid search optimization, and finally a prawn claw-to-body ratio prediction model that meets the expectations is obtained.

[0121] It should be noted that in constructing the phenotypic knowledge graph of the giant freshwater prawn, the first step is to integrate two key data sources: the giant freshwater prawn phenomics network and the giant freshwater prawn growth status network. From the phenomics network, all phenotypic feature nodes and their attribute information are extracted, including key parameters such as the correlation strength between each growth characteristic and the chelicerae ratio, heritability estimates, and individual breeding values. Simultaneously, from the growth status network, anomalous event nodes, treatment plan nodes, and their relationships are extracted. These nodes and relationships constitute the basic elements of the knowledge graph. Based on the extracted phenotypic feature nodes, according to the temporal relationship of phenotypic changes during the growth of individual giant freshwater prawns, the nodes are connected in a time-series manner to form the phenotypic node sequence main axis. This main axis fully represents the phenotypic change pattern of an individual throughout its entire growth cycle. Simultaneously, based on the causal relationship between anomalous event nodes and treatment plan nodes, sequence sub-axes connecting these nodes are generated, forming an event-treatment association network. The network structures along these two axes are then merged. Using the timestamps of anomalous events as index points, the node sequence sub-axes and the phenotypic node sequence main axis are spatiotemporally associated and merged to generate a tree-like node sequence with a branching structure. This tree-like structure visually demonstrates the correspondence between phenotypic changes, abnormal events, and corresponding interventions during individual growth. Ultimately, by integrating the tree-like node sequences of all individuals, a complete phenotypic knowledge graph of the giant freshwater prawn (Macrobrachium rosenbergii) is constructed. After the knowledge graph is built, knowledge representation learning algorithms such as TransE are used to learn low-dimensional vector representations of entities and relationships in the graph, transforming discrete graph information into continuous vector space representations. A training dataset is constructed based on the learned vector representations, and a cheliceroid ratio prediction model is built using a graph neural network architecture. This model updates node representations by aggregating node neighborhood information, effectively capturing the complex correlation between phenotypic features, abnormal events, and cheliceroid ratio. Finally, the model hyperparameters are optimized using methods such as cross-validation and grid search, resulting in a high-performance cheliceroid ratio prediction model that provides reliable technical support for the precise breeding of giant freshwater prawns.

[0122] The continuous monitoring of the recommended and selected giant freshwater prawns during the farming process involves analyzing the prawn growth data to determine if it deviates from the preset farming expectations, formulating corrective measures, and issuing early warnings. Specifically, this includes:

[0123] Figure 3 is a flowchart of the third method of a method for predicting and breeding the claw ratio of giant freshwater prawn based on phenomics analysis according to an embodiment of the present invention.

[0124] As shown in Figure 3, this invention provides a third method flowchart for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis, including:

[0125] S302, real-time acquisition of multimodal monitoring data of selected individuals during the breeding process, including time-series environmental parameters collected by sensor array and individual phenotypic measurement data periodically acquired by image acquisition system;

[0126] S304, compare the multimodal monitoring data with the expected value of the preset breeding path recommended for the individual in the prawn phenotypic knowledge graph, use the dynamic time warping algorithm to calculate the similarity distance between the actual growth sequence and the expected growth sequence, and set the deviation threshold based on the statistical process control method.

[0127] S306, when the actual growth trajectory or environmental parameters are detected to continuously deviate from the expected path and exceed the deviation threshold, an early warning mechanism is triggered; based on the deviation characteristics, the abnormal-treatment mapping knowledge base in the prawn growth status network is queried to match historical similar deviation patterns and their corresponding effective treatment solutions; at the same time, the genetic background information and phenotypic feature association rules of the individual in the prawn phenotypic knowledge graph are retrieved to assess the potential impact of specific treatment measures on the genetic strain;

[0128] S308, taking into account the effects of historical treatment plans and individual genetic characteristics, generates a personalized correction plan for the current deviation. The correction plan includes environmental parameter adjustment strategies, suggestions for adjusting feeding and management measures, and expected recovery trajectory. The correction plan is pushed to the breeding management system and an early warning prompt is triggered simultaneously to guide manual or automatic execution of control operations.

[0129] S310: After implementing the correction plan, continue to monitor the growth response of individuals and changes in environmental parameters, compare the actual recovery trajectory with the expected recovery trajectory, and evaluate the implementation effect of the correction plan; use the deviation event, the corrective measures taken, and the final effect as new knowledge samples, and feed them back to update the prawn growth status network and prawn phenotypic knowledge graph to achieve continuous optimization and learning of the knowledge base, forming a closed-loop intelligent control system of monitoring-early warning-decision-feedback.

[0130] It should be noted that during the intelligent breeding process of giant freshwater prawns, deviations from the recommended breeding path may still occur. Corrections are necessary to avoid breeding errors. A deployed sensor array and image acquisition system acquire multimodal monitoring data of selected individuals in real time, including time-series environmental parameters such as water temperature, dissolved oxygen, and pH, as well as individual phenotypic measurements such as body length and claw length. This data is continuously compared with the preset personalized breeding path expectations. A dynamic time warping algorithm is used to calculate the similarity distance between the actual growth sequence and the expected growth sequence to accurately assess the degree of alignment with the growth trajectory. A deviation threshold is set based on statistical process control methods to establish a quantitative standard for growth monitoring. When the actual growth trajectory or environmental parameters continuously deviate from the expected path and exceed the deviation threshold, an early warning mechanism is automatically triggered. Based on the specific characteristics of the deviation, the system intelligently queries the anomaly-treatment mapping knowledge base in the giant freshwater prawn growth status network to match historical cases with similar deviation patterns and their corresponding effective treatment solutions. Simultaneously, the genetic background information and phenotypic association rules of the individual were retrieved from the prawn phenotypic knowledge graph. A comprehensive analysis of the potential impact and applicability of specific treatment measures on this genetic strain was conducted to ensure the targeted nature of the recommended program. Based on the combined effects of historical treatment programs and the individual's genetic characteristics, a personalized correction plan was generated for the current deviation. This plan included strategies for adjusting environmental parameters, specific suggestions for adjusting feeding and management measures, and a scientific prediction of the expected recovery trajectory. The correction plan was promptly pushed to the aquaculture management system, simultaneously triggering early warning prompts to guide staff in manual adjustments or directly triggering automated equipment to perform control operations. After the correction plan was implemented, the individual's growth response and changes in environmental parameters continued to be closely monitored. The actual recovery trajectory was compared with the expected recovery trajectory to evaluate the effectiveness of the correction plan. Finally, the characteristics of this deviation event, the corrective measures taken, and the final effect were used as new knowledge samples and fed back to update the prawn growth status network and the prawn phenotypic knowledge graph. This allowed the knowledge base to continuously accumulate practical experience, achieving self-improvement and continuous optimization of the system, forming a complete monitoring-early warning-decision-feedback closed-loop intelligent control system, thereby improving the accuracy and intelligence level of prawn breeding.

[0131] In another aspect, the present invention provides a computer-readable storage medium comprising a program for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis. When the program for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis is executed by a processor, it implements the steps of the method for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis as described in any of the preceding claims.

[0132] This invention provides a method for predicting and breeding *Macrobrachium rosenbergii* (Giant River Prawn) chelicerae ratio based on phenomics analysis. By constructing a phenotypic knowledge graph of the prawn, it analyzes the correlation between chelicerae ratio and growth indicators during the prawn's growth process. Simultaneously, by combining relevant aquaculture data, it further reveals the growth scenarios of certain *Macrobrachium rosenbergii* prawns that meet preset desired chelicerae ratios. This is based on different growth trends and states of *Macrobrachium rosenbergii*, thus characterizing their growth patterns. It is possible that certain types of *Macrobrachium rosenbergii* prawns exhibit growth curves different from other types during their full rearing to maturity, and may be prone to abnormal events (diseases or stress) at certain stages of growth. Therefore, it provides personalized aquaculture path recommendations, which, while providing breeding decisions, further improve the corresponding survival rate. Simultaneously, it analyzes whether the aquaculture process deviates from the recommended path and develops and pushes corrective measures, thereby shortening the prawn breeding cycle and improving the accuracy and efficiency of breeding.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0134] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0135] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0136] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0138] 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 the claims.

Claims

1. A method for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis, characterized in that, include: A number of giant freshwater prawns were selected for multi-generation culture. Pedigree data for each individual was recorded during the culture process, and phenotypic data and culture management data were collected at different culture stages to construct a giant freshwater prawn culture database. Phenotypic characteristics of giant freshwater prawns at different culture stages were extracted from this database, the relationship between individual growth indicators and chelicerae ratio was analyzed, and heritability was assessed to construct a giant freshwater prawn phenomics network. The growth path of giant freshwater prawns during the culture process was analyzed using the database, abnormal events during the culture period were identified, and corresponding treatment plans were associated to generate a giant freshwater prawn growth status network. A phenotypic knowledge graph of giant freshwater prawns is constructed by combining the prawn phenomics network and the prawn growth status network, and a prawn chelicerae ratio prediction model is built using the prawn phenotypic knowledge graph. Early phenotypic data are collected from the giant freshwater prawn population to be bred, and bred selection decisions are made using the prawn chelicerae ratio prediction model. The prawn phenotypic knowledge graph is used to recommend aquaculture paths for selected individuals. Multimodal monitoring data of selected individuals during the aquaculture process is obtained, and the actual growth trajectory or environmental parameters are analyzed to see if they deviate from the expected path. If deviations exist, a correction plan is developed and implemented. Heritability assessment and construction of a phenomics network for *Macrobrachium spp.* include: obtaining a set of association rules between phenotypic features and chelicerae ratios; extracting key growth features significantly correlated with chelicerae ratios and their weighting coefficients from the set of rules to construct a fixed-effects design matrix; extracting pedigree tables from a *Macrobrachium spp.* farming database; and constructing a phylogenetic matrix for all *Macrobrachium spp.* individuals using the pedigree tables. Based on the fixed-effects design matrix and the phylogenetic matrix, the additive genetic effects of *Macrobrachium spp.* individuals are treated as random effects, and the phylogenetic matrix is ​​defined as the variance-covariance matrix prior of the random effects. A dynamic genetic network is then constructed. A mixed linear model of the phenotypic model was constructed. The restricted maximum likelihood method was used to fit the mixed linear model, and the variance components were iteratively solved using the EM algorithm to obtain the additive genetic variance component and residual variance component of the chelicerate ratio trait. The heritability estimate was obtained by calculating the ratio of the additive genetic variance to the phenotypic variance, and the breeding value of each individual was solved using the BLUP method. Key growth features were extracted as network nodes based on the phenotypic feature-chelicerate ratio association rule set. The dynamic time warping algorithm was used to calculate the phenotypic distance matrix between individuals to characterize the asynchronous growth trajectory of individual prawns. The graphical lasso algorithm was used to estimate the conditional independence relationship between features, and a sparse Gaussian graph model was constructed. The sparse Gaussian graph model was defined as the basic topology of the phenotypic network, where nodes represent phenotypic features and edge weights represent the conditional correlation coefficients between features. The association strength between each phenotypic feature and the chelicerate ratio was extracted from the phenotypic feature-chelicerate ratio association rule set as the initial attribute of the node. The heritability estimate was used as the global attribute of the network, and the individual breeding value was used as the additional attribute of the corresponding individual phenotypic node, generating the initial phenotypic network.The Node2Vec algorithm is used to learn the low-dimensional vector representations of the nodes in the initial phenomics network. Transition probabilities between nodes are defined, and node embedding vectors are learned by generating node sequences through biased random walks. Finally, the t-SNE dimensionality reduction algorithm is used to map the high-dimensional node embeddings to a two-dimensional space, generating the prawn phenomics network.

2. The method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to claim 1, characterized in that, The process involves selecting several giant freshwater prawns for multi-generational culture, recording the pedigree data of each individual during the culture process, and collecting phenotypic data and culture management data at different stages of culture to construct a giant freshwater prawn culture database. Specifically, this includes: selecting several giant freshwater prawns with chelicerae ratios meeting a desired threshold from existing culture ponds as a first-generation culture population; assigning a unique electronic identifier to each parent and offspring of the giant freshwater prawns entering the pond; constructing a giant freshwater prawn culture pedigree through multi-generational culture; periodically collecting individual image data of giant freshwater prawns at different growth stages, environmental data of the culture ponds, and culture management data of giant freshwater prawns using a deployed culture monitoring sensor array to generate a raw multimodal dataset; preprocessing the obtained raw multimodal dataset; and importing the collected individual image data of giant freshwater prawns into a convolutional neural network-based... The framework-built image segmentation model acquires segmented images of individual giant freshwater prawns. Morphological phenotypic data of each individual is calculated based on a preset pixel-to-actual-size conversion coefficient and bound to the corresponding individual ID and acquisition timestamp to generate a giant freshwater prawn phenotypic dataset. For the environmental data of the giant freshwater prawn farming ponds and the giant freshwater prawn management data, an outlier monitoring algorithm is used for data cleaning to identify and remove abnormal measurements. Based on the timestamps of management events recorded in the farming logs, environmental data fluctuation characteristics before and after the event are extracted, and a management label is generated for each event. This management label includes management measures, changes in environmental parameters before and after management, and giant freshwater prawn status data, ultimately generating a farming management dataset with event semantics. A giant freshwater prawn farming database is constructed by combining the giant freshwater prawn farming pedigree, the giant freshwater prawn phenotypic dataset, and the farming management dataset.

3. The method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to claim 1, characterized in that, The process of extracting phenotypic characteristics of prawns at different farming stages from the prawn farming database and analyzing the relationship between individual growth indicators and chelicerae ratio specifically includes: acquiring the prawn farming database; extracting complete time-series phenotypic data of all prawn individuals at various time points from the database; generating prawn phenotypic measurement sequences; segmenting the prawn phenotypic measurement sequences using a sliding window of fixed time length to generate several prawn phenotypic measurement subsequences; calculating the first-order difference between adjacent time points for each prawn phenotypic measurement subsequence as the instantaneous growth rate feature, and calculating the difference between the first-order difference sequences as the second-order difference to characterize growth acceleration; simultaneously, performing cubic spline interpolation or polynomial fitting on the complete growth curve of each phenotypic indicator, extracting the coefficients of the fitting function as shape features describing the overall growth pattern, and calculating the statistical characteristics of each sequence. The phenotypic characteristics of the prawns, including mean, variance, skewness, and kurtosis, were analyzed to obtain a sequence of derived phenotypic features. A multimodal feature set representing the dynamic growth pattern of individual prawns was constructed by integrating the prawn phenotypic measurement sequence and the prawn phenotypic derived feature sequence. Correlation analysis was performed using the final measured chelicerae ratio for each prawn individual to calculate the maximum mutual information coefficient and Pearson correlation coefficient between each feature and the chelicerae ratio. Subsequently, a random forest regression model was used, with the final measured chelicerae ratio as the target variable and the prawn phenotypic measurement features and extended prawn phenotypic features as decision variables. The importance score of each feature was calculated using 10-fold cross-validation. Key growth features significantly correlated with the chelicerae ratio and their weight coefficients were analyzed. A Gaussian process regression model was established using the selected key growth features to fit the nonlinear mapping relationship between phenotypic features and target traits, ultimately generating a set of phenotypic feature-chelicerae ratio association rules.

4. The method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to claim 1, characterized in that, The process of analyzing the growth path of giant freshwater prawns (Giant River prawns) during the farming process using the giant freshwater prawn farming database, identifying abnormal events during the farming period and associating them with corresponding treatment plans, and generating a giant freshwater prawn growth trend network specifically includes: acquiring the giant freshwater prawn farming database; extracting time-series phenotypic measurement data and corresponding environmental parameter sequences of all giant freshwater prawn individuals from the database; performing preprocessing and then using a nonlinear mixed-effects model, with individuals as random effects, to fit the standard growth curve of giant freshwater prawns and establish an individual growth curve model describing the normal growth pattern; based on the established individual growth curve model, calculating the residual sequence between the actual measured value and the model predicted value of each individual, and performing wavelet transform analysis on the residual sequence to extract time-frequency domain features and generate a set of giant freshwater prawn individual growth trajectories; combining the environmental parameter sequences with the isolated forest algorithm to detect abnormal events during the growth process, and recording the occurrence time, event characteristics, duration, and abnormal intensity index of abnormal events to generate an abnormal growth event feature set; matching the identified abnormal event points with the control log records in the farming control dataset based on the timestamps of the abnormal growth event feature set; if a certain abnormal event point has a manually recorded control operation, then the corresponding abnormal event... The system associates events with control measures to form anomaly event-disposal plan association pair; otherwise, it identifies environmental self-fluctuation events as anomaly event-self-fluctuation association pair and marks them as events requiring no action. Based on the obtained association pairs, anomaly-disposal mapping knowledge base is constructed with the association order of anomaly event characteristics, environmental parameter characteristics, and disposal plans. This knowledge base includes the anomaly event type, environmental parameter characteristics, the disposal measures taken, and the effects after disposal. The system extracts the growth trajectory features of each individual prawn from the prawn individual growth trajectory set, generating several growth trajectory feature sequences with temporal attributes. Based on the anomaly-disposal mapping knowledge base, the system merges the prawn individual identifier corresponding to the anomaly event and the timestamp of the anomaly event into the corresponding growth trajectory feature sequences to generate a sequence representing the growth status of each prawn individual during the farming process. The system calculates the cosine similarity and Mahalanobis distance between each growth status sequence, performs weighted averaging and normalization to generate a sequence merging index, and compares it with a preset threshold. If the index is greater than the preset threshold, the system calculates the difference between the two growth status sequences to identify the difference sequence segments. New sequence branches are generated in the difference sequence segments for sequence merging. Through repeated iterative merging steps, the system finally generates a prawn growth status network.

5. The method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to claim 1, characterized in that, The process of constructing a phenotypic knowledge graph of *Macrobrachium spp.* by combining the phenotypic network and the growth status network of *Macrobrachium spp.*, and then using this phenotypic knowledge graph to construct a cheliceroid ratio prediction model, specifically includes: acquiring the phenotypic network and the growth status network of *Macrobrachium spp.*; extracting all phenotypic feature nodes and node attribute information from the phenotypic network, including the correlation strength between growth characteristics and cheliceroid ratio, heritability estimates, and individual breeding values; extracting anomalous event nodes, treatment plan nodes, and their correlations from the growth status network; generating a phenotypic node sequence main axis based on the extracted phenotypic feature nodes, with the phenotypic change sequence during the growth process of the individual *Macrobrachium spp.* as the node connection order, whereby the phenotypic node sequence main axis represents all phenotypic change patterns of the individual *Macrobrachium spp.* during the growth process; and determining the correlation between anomalous event nodes and treatment plan nodes. As a connection basis, node connection edges are generated to connect abnormal event nodes and handling plan nodes, generating node sequence sub-axis. Based on the node sequence sub-axis, the node sequence trigger event stamp is used as an index to merge with the node sequence main axis to obtain several tree-like node sequences. The tree-like node sequences represent the phenotypic change patterns and growth status of individual prawns during the farming process. Finally, a prawn phenotypic knowledge graph is constructed based on all tree-like node sequences. The TransE knowledge identifier learning algorithm is used to learn low-dimensional vector representations of entities and relationships in the prawn phenotypic knowledge graph and construct a training dataset. A prawn claw-to-body ratio prediction model is built based on a graph neural network architecture, and the model is trained using the training dataset. Hyperparameters are optimized through cross-validation and grid search optimization to finally obtain a prawn claw-to-body ratio prediction model that meets the expectations.

6. The method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to claim 1, characterized in that, The process involves collecting early phenotypic data from the proposed Macrobrachium rosenbergii population, making breeding decisions using the prawn claw-to-body ratio prediction model, and recommending breeding paths for selected individuals using the prawn phenotypic knowledge graph. Specifically, this includes: acquiring morphological phenotypic data for each individual using an automated image acquisition system while simultaneously monitoring and collecting environmental parameters to obtain environmental monitoring data; preprocessing and standardizing the collected early phenotypic data and environmental monitoring data to form early monitoring feature vectors for the selected individuals; inputting these early monitoring feature vectors into a pre-trained prawn claw-to-body ratio prediction model, which, through multi-layer nonlinear transformations and feature extraction, outputs predicted claw-to-body ratios for adult individuals and corresponding confidence intervals; and sorting the individuals based on their predicted claw-to-body ratios, combining the predicted confidence intervals with preset selection criteria. By selecting an intensity threshold, individuals with chelicerae ratios meeting preset expectations are selected as candidate parents, generating a prawn breeding recommendation table. Based on this table, aquaculture pathway analysis is performed using a prawn phenotypic knowledge graph. Early monitoring feature vectors and pedigree information of the individuals to be bred are used as query criteria to retrieve historical individual nodes with similar phenotypic characteristics and genetic backgrounds from the knowledge graph. This yields the optimal environmental parameter ranges experienced by each historical node throughout its entire aquaculture cycle, successful management sequence sequences, and final growth performance indicators, generating similar aquaculture cases. Based on these similar cases, case-based reasoning technology is used to generate personalized aquaculture pathway recommendations for the target individuals, including environmental parameter control targets at different growth stages, feeding strategies, density management requirements, and anticipated abnormal events and corresponding countermeasures. The prawn breeding recommendation table is then linked with the personalized aquaculture pathway recommendations to generate a final prawn breeding decision report, which is then pushed out.

7. The method for predicting and breeding the claw ratio of Macrobrachium rosenbergii based on phenomics analysis according to claim 1, characterized in that, The process involves acquiring multimodal monitoring data of selected individuals during the rearing process, analyzing whether the actual growth trajectory or environmental parameters deviate from the expected path, and formulating and pushing correction plans if deviations are found. Specifically, this includes: real-time acquisition of multimodal monitoring data of selected individuals during the rearing process, including time-series environmental parameters collected by a sensor array and individual phenotypic measurement data periodically acquired by an image acquisition system; comparing the multimodal monitoring data with the expected value of the preset rearing path recommended for the individual in the prawn phenotypic knowledge graph, calculating the similarity distance between the actual growth sequence and the expected growth sequence using a dynamic time warping algorithm, and setting a deviation threshold based on a statistical process control method; triggering an early warning mechanism when the actual growth trajectory or environmental parameters continuously deviate from the expected path and exceed the deviation threshold; and querying the anomaly-treatment mapping knowledge base in the prawn growth status network based on the deviation characteristics to match historical similar deviation patterns. The study analyzes the data and corresponding effective treatment plans. Simultaneously, it retrieves the genetic background information and phenotypic association rules of the deviating individuals from the prawn phenotypic knowledge graph to assess the potential impact of specific treatment measures on the genetic strain to which the deviating individuals belong. By integrating the effects of historical treatment plans and individual genetic characteristics, a personalized correction plan is generated for the current deviation. This correction plan includes environmental parameter adjustment strategies, suggestions for adjusting feeding and management measures, and an expected recovery trajectory. The correction plan is then pushed to the aquaculture management system, triggering a warning prompt to guide manual or automatic control operations. After implementing the correction plan, the study continues to monitor the individual's growth response and changes in environmental parameters, comparing the actual recovery trajectory with the expected recovery trajectory to evaluate the effectiveness of the correction plan. Finally, the study uses the deviation event, the corrective measures taken, and the final effect as new knowledge samples, updating the prawn growth status network and the prawn phenotypic knowledge graph to achieve continuous optimization and learning of the knowledge base.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a program for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis. When the program is executed by a processor, it implements the steps of the method for predicting and breeding the claw-to-body ratio of Macrobrachium rosenbergii based on phenomics analysis as described in any one of claims 1 to 7.

Images