Method for predicting gonad maturity of procambarus clarkii based on machine learning model

Abstract

The invention discloses a method for predicting gonad maturity of procambarus clarkii based on a machine learning model, and belongs to the field of aquaculture breeding. According to the method, five phenotypic data including the body weight, the head and chest shell width, the body length, the abdominal node width and the cheliped width of the procambarus clarkii are collected in a lossless mode, three base models including a Gaussian naive Bayes classifier, a decision tree classifier and an OneClass support vector machine are combined, an ovary weight prediction model is constructed through meta-learning weighted integration, and the sexual maturity state is judged with 0.38 g as a threshold value. Through verification, the prediction accuracy rates of the model on mature female shrimps and immature female shrimps are 82.4% and 92.3% respectively. Compared with a traditional dissection method, the method has the advantages that non-invasive detection is achieved, parent loss is avoided, dynamic monitoring is supported, sexual maturity individuals are accurately screened, synchronous mating and breeding are achieved, the growth consistency of offspring and the specification uniformity of commercial shrimps are remarkably improved, and a key technical support is provided for large-scale and efficient breeding of procambarus clarkii offspring seeds.

Description

Technical Field

[0001] The present invention belongs to the field of aquaculture breeding, and relates to a method for predicting the gonadal maturity of Procambarus clarkii based on a machine learning model, which is applicable to the synchronous mating management of parents in large-scale farms. Background Art

[0002] Procambarus clarkii ( Procambarus clarkii ), commonly known as crayfish, is an important freshwater aquaculture variety in China. With its advantages of fast growth rate, delicious meat, rich nutrition, etc., the domestic consumption demand and aquaculture scale have increased rapidly in recent years. With the industrial upgrade, the market's requirements for the consistency of commercial shrimp specifications and the stability of quality have been increasing day by day. Commercial shrimp with uniform specifications enjoy a significant premium in the market due to high processing efficiency and stable meat yield, while batches with mixed specifications often face the dilemma of low-price sales or unsalable due to large individual differences, especially prominent during the concentrated listing period, directly affecting the aquaculture economic benefits.

[0003] The core cause of the specification difference of commercial shrimp lies in the asynchrony of the growth cycle of shrimp seedlings. On the premise of standardizing the aquaculture environment and nutritional conditions, this asynchrony is mainly due to the dispersion of the seedling emergence time. The gonadal development of crayfish parents is asynchronous, and the time difference in reaching sexual maturity leads to batch-by-batch mating and spawning, and the hatching time of fertilized eggs shows a stepped distribution. The seedlings hatched earlier often form a specification suppression on the late seedlings during listing due to their longer growth cycle, ultimately resulting in the differentiation of individual specifications of commercial shrimp in the same batch. This phenomenon is particularly prominent in large-scale aquaculture, restricting the possibility of batch synchronous listing and increasing the difficulty of aquaculture cycle regulation. The key to solving this problem lies in accurately screening out gonadally mature parents to ensure synchronous mating timing.

[0004] Currently, the screening of sexually mature parents mainly relies on the anatomical observation method, which directly observes the gonadal state by destroying the cephalothorax carapace of female shrimps, and has the following technical defects: First, the screening process causes irreversible loss of high-quality parents, exacerbating the waste of germplasm resources; second, it can only judge the instantaneous maturity state and cannot dynamically track the gonadal development process; third, the manual detection efficiency is low and it is difficult to meet the requirements of large-scale aquaculture. Although the empirical judgment method based on body color and morphological characteristics has been applied, its judgment accuracy and stability are insufficient due to environmental interference and individual variation.

[0005] In view of the above technical bottlenecks, the present invention establishes an association model between gonadal maturity and phenotypic data through machine learning to achieve non-invasive in-vivo detection and break through the limitations of traditional methods. This technology not only avoids parent loss, but also can dynamically monitor the sexual maturity process, provides key technical support for the synchronization of seedling batches, and ultimately significantly improves the specification uniformity of commercial shrimp and aquaculture benefits, helping the crayfish industry to upgrade towards an intensive and large-scale sustainable direction. Summary of the Invention

[0006] The object of the present invention is to provide a non-invasive, efficient and accurate method for predicting the gonadal maturity of Procambarus clarkii, so as to overcome the problems of parental death and waste of aquaculture resources caused by anatomical examination in the prior art.

[0007] To achieve the above object, the applicant selected apparent morphological data that contribute greatly to ovarian weight and can be obtained without damage, and constructed a machine learning model for predicting the gonadal maturity of Procambarus clarkii through training. After verification, the prediction accuracy of this model for female shrimps with mature gonads is 82.4%, and the prediction accuracy for female shrimps with immature gonads is 92.3%, with an overall accuracy rate of 86.7%. The machine learning models adopted in the present invention, such as Gaussian Naive Bayes classifier and Decision Tree classifier, can effectively handle complex non-linear relationships and adapt to different sample feature distributions; OneClass Support Vector Machine can still provide relatively accurate prediction results in the case of sample imbalance. In addition, the meta-learning integration method not only improves the overall prediction accuracy, but also significantly reduces prediction errors, thereby improving the stability and reliability of the model.

[0008] The more specific technical solution is as follows: A method for predicting the gonadal maturity of Procambarus clarkii based on a machine learning model, comprising the following steps: 1) Collect morphological data of Procambarus clarkii, and screen out morphological features that contribute greatly to ovarian weight and can be obtained without damage; 2) Use a machine learning model to train and perform weighted integration on the morphological features to obtain a machine learning model for predicting the gonadal maturity of Procambarus clarkii; 3) Use the machine learning model to predict the gonadal maturity of Procambarus clarkii.

[0009] Among them, the morphological features screened out in step 1) include body weight, carapace width, body length, abdominal segment width and cheliped width.

[0010] Among them, the machine learning model includes Gaussian Naive Bayes classifier (GaussianNB), Decision Tree classifier (DecisionTreeClassifier) and OneClass Support Vector Machine (OneClassSVM).

[0011] Further, the multiple machine learning models are weighted and integrated through a meta-learning integration method, and the integration formula is: Om stacking=0.11· f GNB (X) +0.61· f DT (X)+0.28· f SVM (X) In the formula, Om stacking is the ovarian weight after meta - learning weighted integration, f GNB (X) , f DT (X) , f SVM (X) are the predicted values of Gaussian Naive Bayes, decision tree and OneClass support vector machine respectively.

[0012] If the predicted ovarian weight ≥ 0.38 g, it is determined that the gonad is mature, and the output result is "1" (mature); if the predicted ovarian weight < 0.38 g, it is determined that the gonad is immature, and the output result is "0" (immature).

[0013] The present invention further provides a system for predicting the gonad maturity of Procambarus clarkii, including: A data acquisition module for obtaining the morphological data of Procambarus clarkii; A feature selection module for screening out morphological features that contribute greatly to ovarian weight based on a random forest model; A model training module for training a Gaussian Naive Bayes classifier, a decision tree classifier and a OneClass support vector machine; A meta - learning integration module for integrating the prediction results of multiple models through a weighted formula; A prediction module for outputting the determination result of gonad maturity, and taking ovarian weight ≥ 0.38 g as the maturity standard.

[0014] Furthermore, the system further includes a model evaluation module for calculating the accuracy, Kappa coefficient, precision, recall rate and F1 - score of the model, and generating a classification result comparison table.

[0015] The present invention also provides a computer - readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the method for predicting the gonad maturity of Procambarus clarkii as described above.

[0016] The beneficial effects of the present invention are: The present invention first proposes a non-invasive method for judging the gonadal maturity of crayfish, which not only helps to improve the breeding efficiency and ensure the survival rate of parental shrimps, but also provides strong technical support for crayfish breeding. The modeling data used in the present invention is not only simple to measure and easy to obtain, but also highly correlated with gonadal maturity. The selected machine learning model and weighted integration method are beneficial to improving the stability and reliability of the model. The present invention has good application prospects in crayfish breeding and cultivation. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 It is a schematic flow chart for the construction of a prediction model; Figure 2 It is a data relationship diagram of ovarian weight distribution and maturity; Figure 3 It is a distribution diagram of phenotypic feature importance; Figure 4 It is a diagram of the appearance of mature and immature ovaries of Procambarus clarkii. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The technical solutions of the present invention will be further described in detail below with reference to specific embodiments.

[0018] This embodiment provides a method for predicting the gonadal maturity of Procambarus clarkii through a machine learning model. Briefly, the method includes: collecting and standardizing the morphological feature data of Procambarus clarkii; establishing a machine learning model using the feature data; and using the prediction model to predict the gonadal maturity of Procambarus clarkii. The specific flow chart is as Figure 1 shown Example 1 Construction of a Prediction Model for Gonadal Maturity of Procambarus clarkii 1. Data collection Collect the phenotypic data of 600 Procambarus clarkii, and obtain the body length (l), body weight (m), cheliped length (chl), cheliped weight (chm), cheliped width (chw), total length (fl), carapace length (hl), carapace width (hw), abdominal segment width (asw), tail fan length (tfl), and tail fan width (tfw) data of each shrimp. The ovarian weight (om) and ovarian development stage (t) of each sample are also recorded. These data have undergone strict quality control and standardization processing to ensure the accuracy of the data. The units of the collected data are centimeters (cm) and grams (g). Table 1 shows the processed data of some shrimps.

[0019] Table 1. Partial Phenotypic Data of Procambarus clarkii

[0020] 2. Feature selection As Figure 2As shown in the figure, the data analysis of the ovarian weight and gonadal maturity of 600 Procambarus clarkii found that when the ovarian weight reached 0.38 g or more, most of the ovaries had reached the maturity stage of stage IV or stage V. Therefore, the criterion for judging whether the gonads of Procambarus clarkii are mature is that the ovarian weight ≥ 0.38 g.

[0021] Next, through feature selection, the features that are most important for predicting ovarian weight are analyzed and selected. The random forest model (RandomForestClassifier) and the feature importance method (model.feature_importances) are used to calculate the importance of each feature. Then, a feature importance ranking graph is drawn through matplotlib to show the importance of each feature in the form of a horizontal bar chart. As Figure 3 shown, the feature importance analysis shows that body weight (m), carapace width (hw), abdominal segment width (asw), cheliped weight (chm), and cheliped width (chw) are the five features that have the greatest impact on predicting ovarian weight. Among them, only the measurement of "cheliped weight (chm)" requires cutting off the cheliped, which violates the principle of non-destructive testing. Therefore, "body length (l)" is selected to replace "cheliped weight (chm)", and five features are established for the training and prediction of the subsequent model.

[0022] 3. Dataset division The dataset is randomly divided into a training set (80%) and a test set (20%) for model training and validation. When splitting the data, ensure that the training set and the test set are consistent in the distribution of morphological features to avoid the impact of sample imbalance on model performance.

[0023] 4. Model training and prediction The machine learning methods tested in the present invention include logistic regression (LogisticRegression), random forest (RandomForestClassifier), support vector classification (SVC), linear support vector classification (LinearSVC), kernel support vector classification (NuSVC), K-nearest neighbor (KNeighborsClassifier), Gaussian naive Bayes (GaussianNB), perceptron (Perceptron), decision tree (DecisionTreeClassifier), stochastic gradient descent classification (SGDClassifier), gradient boosting classification (GradientBoostingClassifier), multi-layer perceptron classification (MLPClassifier), etc.

[0024] Use grid search on a single model to search for the optimal parameters of each model. Adopt two model integration methods: VotingClassifier and MetaLearningClassifier, and use recursive search for various model combination schemes for integration. Select the model combination with the optimal accuracy after integration to build the final model.

[0025] 5. Model evaluation: To verify the prediction effect of the model, use the following classification evaluation metrics to evaluate the model: Accuracy: Measure the overall prediction accuracy of the model on the test set; Kappa coefficient: Used to evaluate the consistency of the model prediction and eliminate the influence of accidental factors; Precision: Used to measure the accuracy of the model in positive class prediction; Recall: Measure the detection rate of the model in all positive class samples; F1 Score: A comprehensive evaluation metric that combines precision and recall.

[0026] The performance of each model is shown in Table 2. Each model has advantages under specific metrics. The Gaussian Naive Bayes classifier reaches 0.90 in the precision of immature samples, but its overall accuracy is only 60.94%; the Decision Tree classifier performs outstandingly in recall, with a recall rate of up to 0.99 for immature samples and an overall accuracy of 83.99%. OneClass Support Vector Machine performs well in both recall and F1 score, with an F1 score (average value) reaching 0.52 and an accuracy of 83.55%.

[0027] Table 2. Model result evaluation table

[0028] The comparison results of the Voting Ensemble model and the Meta-Learning Ensemble model are shown in Table 3. The Voting Ensemble model correctly predicted 16 mature samples in the prediction of mature samples, but misclassified 5 mature samples as immature and misjudged 99 immature samples as mature. While the Meta-Learning Ensemble method, while improving the overall prediction accuracy, significantly reduces prediction errors, especially in mature samples, correctly predicting 18 mature samples, misjudging only 6, and reducing the misjudgment of immature samples to 97. Overall, the Meta-Learning Ensemble method has obvious advantages in both considering prediction accuracy and reducing the misjudgment rate, especially in the classification of mature samples.

[0029] Table 3. Comparison of prediction results of two models

[0030] Finally, a meta - learning integration was carried out using the Gaussian Naive Bayes classifier, the decision tree classifier, and the OneClass support vector machine. The overall accuracy was increased to 84.88%, ranking among the top in all models, and the Kappa coefficient was 0.2131, showing high stability and consistency.

[0031] The formula is as follows: Om stacking=0.11· f GNB (X) +0.61· f DT (X) +0.28· f SVM (X) In the formula, Om stacking is the ovarian weight after meta - learning weighted integration, f GNB (X) is the ovarian weight predicted by the Gaussian Naive Bayes model, f DT (X) is the ovarian weight predicted by the decision tree model, f SVM (X) is the ovarian weight predicted by the OneClass support vector machine (SVM) model.

[0032] Result determination: If the predicted ovarian weight ≥ 0.38 g, it is determined as gonadal maturity (stage IV - V).

[0033] Example 2 Using the model to predict the gonadal maturity of Procambarus clarkii The sexual maturity of 30 female shrimps was predicted, and the specific steps are as follows: 1) Measure the body length (l), body weight (m), chela width (chw), carapace width (hw), and abdominal segment width (asw) of female Procambarus clarkii; 2) Input the measured body surface phenotype data into the model to predict the ovarian weight, and the output result is "1" for maturity and "0" for immaturity; 3) Dissect 30 female shrimps, pry open the carapace from the back, and observe the ovarian color. The ovaries of female shrimps are white or yellow before maturity and brown after maturity. As Figure 4As shown, the ovary is within the blue frame. The immature shrimp is on the left, with a pale yellow ovary, and the mature shrimp is on the right, with a brown ovary. Record the maturity of each shrimp according to the color of the ovary. If it is mature, record it as "1", and if it is immature, record it as "0".

[0034] The results are shown in Table 4, which records the data of 30 shrimps. "Actual" is the maturity of the shrimp judged by actual dissection, and "predict" is the result obtained by model prediction. Among them, the predictions for shrimp No. 2, No. 9, No. 24, and No. 27 are incorrect. There are 17 mature female shrimps in total, and the model predicts 14 of them, with an accuracy rate of 82.4%. There are 13 immature female shrimps, and the model predicts 12 of them, with an accuracy rate of 92.3%. The comprehensive evaluation shows that the accuracy rate of this model reaches 86.7%.

[0035] Table 4 Prediction Results

Claims

1. A method for predicting gonadal maturity of Procambarus clarkii based on a machine learning model, characterized in that: The following steps are involved: 1) Collect morphological data of Procambarus clarkii and select morphological features that contribute greatly to ovarian weight and can be obtained non-destructively; 2) using a machine learning model to train the morphological features and perform weighted integration to obtain a machine learning model for predicting gonad maturity of Procambarus clarkii; 3) Use the machine learning model to predict the gonad maturity of Procambarus clarkii.

2. The method according to claim 1, characterized in that: The morphological characteristics include body weight, carapace width, body length, abdominal segment width and chelipeds width.

3. The method according to claim 1, characterized in that: The machine learning models include Gaussian naive Bayes classifier (GaussianNB), decision tree classifier (DecisionTreeClassifier) ​​and OneClass support vector machine (OneClassSVM).

4. The method according to claim 3, characterized in that: The multiple machine learning models are weighted integrated by a meta-learning integration method, and the integration formula is: Om stacking=0.11· f GNB (X) +0.61· f DT (X) +0.28· f SVM (X) in, Om stacking is the ovarian weight after meta-learning weighted integration, f GNB (X) , f DT (X) , f SVM (X) They are the predicted values ​​of Gaussian Naive Bayes, Decision Tree and OneClass Support Vector Machine respectively.

5. The method according to claim 4, characterized in that: If the predicted ovarian weight is ≥0.38g, the gonads are determined to be mature, and the output result is "1" (mature); if the predicted ovarian weight is <0.38g, the gonads are determined to be immature, and the output result is "0" (immature).

6. A system for predicting gonadal maturity of Procambarus clarkii, characterized in that: include: A data acquisition module, used to obtain the morphological data of Procambarus clarkii; The feature selection module screens out morphological features that contribute most to ovarian weight based on the random forest model; Model training module, used to train Gaussian Naive Bayes classifier, decision tree classifier and OneClass support vector machine; Meta-learning integration module, which integrates the prediction results of multiple models through weighted formulas; The prediction module outputs the gonad maturity determination results and uses ovarian weight ≥ 0.38g as the maturity standard.

7. The system according to claim 6, characterized in that The system also includes a model evaluation module for calculating the accuracy, Kappa coefficient, precision, recall and F1 score of the model and generating a classification result comparison table.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method for predicting gonad maturity of Procambarus clarkii according to any one of claims 1 to 5 is implemented.

Images