An adaptive regression modeling analysis method based on rice metabolic hyperspectral phenotype data

Abstract

The present application relates to the technical field of crop hyperspectral phenotype data modeling analysis, and discloses a self-adaptive regression modeling analysis method based on rice metabolic hyperspectral phenotype data, comprising the following steps: step one, collecting hyperspectral data of the wavelength range of a rice grain sample, segmenting the grain area based on grain morphological characteristics, and extracting features using a double-branch convolutional neural network model, wherein the double-branch convolutional neural network model comprises a spectral branch and a spatial branch, the spectral branch adopts a three-dimensional convolutional neural network architecture and is embedded with a residual dense block and a spectral attention mechanism, and the spatial branch adopts a two-dimensional convolutional neural network architecture and is embedded with a residual dense block and a spatial attention mechanism. The present application uses a bidirectional symmetric search algorithm to optimize the hyperparameters of the modeling algorithm, adaptively searches and selects the optimal hyperparameters of various algorithms according to different rice metabolic hyperspectral phenotype data, and enhances the performance of the model.

Description

Technical Field

[0001] This invention relates to the field of crop hyperspectral phenotypic data modeling and analysis technology, specifically an adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data. Background Technology

[0002] Rice plays a vital role in global food security, with over half the world's population relying on it as a staple food. To meet the demands of a continuously growing population, global rice production is projected to increase by 112 million tons by 2035. However, this goal faces significant challenges due to the continued depletion of arable land and irrigation water resources, as well as increasing uncertainty in climate conditions. From a crop breeding perspective, primary metabolites in rice play a crucial role in maintaining plant growth and development, while secondary metabolites help enhance the plant's resistance to stress and adverse conditions.

[0003] Phenotypic analysis is a key technique for systematically observing and analyzing the external morphological characteristics of organisms to identify and diagnose pathological effects. Plant phenotypic analysis provides a foundation for exploring the laws of plant life activities, revealing plant adaptive mechanisms, and improving traits such as crop yield. Hyperspectral imaging technology, due to its non-destructive and efficient detection capabilities, has become an important tool in modern phenotypic analysis. Typically, high-concentration macromolecules can be accurately predicted using limited spectral bands, while low-concentration small-molecule metabolites are more difficult to detect. Hyperspectral imaging technology can obtain continuous spectral information in the visible and infrared bands, making it easier to predict the content of low-molecular-weight metabolites in plants.

[0004] However, current hyperspectral phenotypic data processing still faces certain limitations. Most studies rely on manually selecting regression algorithms and using uniform hyperparameter settings for different datasets, leading to increased computational complexity and potentially lower model accuracy and insufficient generalization ability. Therefore, automating the selection of the most suitable regression model and optimizing hyperparameters has become a key challenge for improving the efficiency and accuracy of hyperspectral data analysis.

[0005] Therefore, the problems to be solved by the present invention include, but are not limited to, the fact that traditional rice metabolic hyperspectral phenotypic data generally rely on commercial software for modeling and analysis, which makes it difficult to fully extract information from the phenotypic data, and the model has poor generalization ability and lacks universality.

[0006] To address these issues, this invention proposes an adaptive regression modeling and analysis method based on hyperspectral phenotypic data of rice metabolism. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data, in order to solve the problems mentioned in the background.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data, comprising:

[0009] Step 1: Collect rice grain samples Hyperspectral data across a wavelength range was used to segment grain regions based on grain morphology features. A dual-branch convolutional neural network model was employed to extract features, comprising a spectral branch and a spatial branch. The spectral branch utilizes a three-dimensional convolutional neural network architecture and embeds residual dense blocks. The spectral attention mechanism uses a two-dimensional convolutional neural network architecture for the spatial branch and embeds residual dense blocks. The spatial attention mechanism uses a multi-head cross-attention fusion module to fuse the spectral branch output features as queries and the spatial branch output features as keys and values, resulting in the output. 3D fusion features;

[0010] Step 2: High-performance liquid chromatography (HPLC) is used to target and determine the presence of [specific substances] in rice grain samples from the same batch. The content of various metabolites was determined to obtain the true metabolite content values ​​of rice grain samples.

[0011] Step 3: Establish a metabolite content-hyperspectral feature association database, and construct a database based on metabolite content and fusion features. A bidirectional symmetric search algorithm is used to optimize the hyperparameters of various regression models and output the optimal hyperparameters for each regression model.

[0012] Step 4: Select the optimal regression model based on the evaluation of the model's generalization ability. Construct each regression model with the optimal output hyperparameters, calculate the difference in the coefficient of determination between the training set and the test set of each model, and select the optimal regression model as the final prediction model based on the difference in generalization ability.

[0013] Step 5: Using the extracted convolutional neural network features as surrogate phenotypes, a mixed linear model genome-wide association analysis is performed. Principal component analysis and phylogenetic matrix are introduced to control false positives and screen for initially significant loci. The measured metabolite content is used as the baseline phenotype for fine-tuning within the initially selected loci, employing regional analysis. A significance threshold was set for calibration, and candidate genes that were simultaneously located in the peak region of the phenotypic Manhattan plot were screened in combination with rice genome annotation information.

[0014] Preferably, the following is used in step three: There are several ways to build a regression model, including: lasso regression, support vector machine regression, random forest regression, etc. Nearest neighbor regression, decision tree regression, partial least squares regression, ridge regression, and extreme gradient boosting regression.

[0015] Preferably, in step three, the bidirectional symmetric search algorithm initializes the population and sets the proportions of explorers, collaborators, and scouts. It calculates the safety value of each individual based on the negative mean square error function and assigns individual roles according to the size of the safety value. The individual with the highest safety value is assigned as an explorer responsible for global search, the individual with the lowest safety value is assigned as a scout responsible for environmental scouting, and the remaining individuals are assigned as collaborators responsible for assisting in exploration.

[0016] Preferably, in step three, when an individual's position exceeds the search boundary, a reflection mechanism is used to process it. The difference between the individual's position and the boundary is calculated for reverse mapping, and perturbation noise is added to avoid boundary adhesion. Then, a truncation operation is performed to ensure that all individual positions are limited to the legal range of the search space.

[0017] Preferably, in step four, the criterion for selecting the optimal regression model is: when there exists a generalization difference less than 1 / 3... When selecting models, choose the model with the highest determination coefficient on the test set as the optimal model; when the generalization difference of all models is greater than or equal to... The model with the highest determination coefficient on the test set is directly selected as the optimal model.

[0018] Preferably, in step five, a regression model is established to screen samples with a determination coefficient greater than [value missing]. The characteristics of the surrogate phenotype are used as surrogate phenotypes to ensure that the surrogate phenotypes are highly correlated with the actual metabolite content for subsequent association analysis.

[0019] In step five, the fine positioning is performed upstream and downstream of the initially selected significant site. The study was conducted within a specific range, and regional variations were determined by calculating the number of local single nucleotide polymorphism sites. The threshold was adjusted, and sites that were significant in both surrogate and baseline phenotype genome-wide association analyses were selected as final candidate regions.

[0020] Preferably, the formula processed by the reflection mechanism is as follows:

[0021] ,

[0022] in, and For the upper and lower bounds of the search space, For minor noise, , Noise intensity factor;

[0023] Subsequent approval The operation ensures that the position of each entity is restricted to the legal area of ​​the search space. The specific formula for the operation is as follows: ,

[0024] in, The first in the population Individual in the first Location information in 3D space.

[0025] Preferably, in step three, a bidirectional symmetric search algorithm is used for hyperparameter optimization. The initialization of the bidirectional symmetric search algorithm specifically includes:

[0026] Set the population size as Each individual has Bit hyperparameters For the first In the nth iteration Individual in the first Location information in the dimension Population matrix;

[0027] The position of each individual in the population is represented as follows:

[0028] ,

[0029] Set the ratio of exploring individuals, cooperative individuals, and scout individuals, and calculate the safety value.

[0030] Preferably, the position update of the explored individual adopts real-time step size coefficient control. When the individual's safety value is greater than or equal to the product of the safety threshold and the maximum safety value, the step size calculation based on the search boundary is adopted. When the individual's safety value is less than the product of the safety threshold and the maximum safety value, the step size calculation based on the population centroid distance is adopted. A random variable that follows a heavy-tailed distribution is introduced to enhance the jump search capability.

[0031] The cooperative individual updates its position by monitoring the position changes of the exploring individual. It takes the current optimal position of the exploring individual as the target position and calculates the new position by combining random numbers that follow a normal distribution and a random assignment matrix. When the safety value of the cooperative individual is lower than the average safety value, it explores other areas. When the safety value of the cooperative individual is higher than the average safety value, it moves with the exploring individual.

[0032] The reconnaissance individual determines its movement direction based on a comparison between its current safety value and the average of the global best and worst safety values. It then calculates its position update using step size control parameters and random numbers following a standard normal distribution. When an individual's safety value is higher than the global average safety value, it moves toward the global best position. When an individual's safety value is lower than the global average safety value, it moves toward other individuals.

[0033] Preferably, in step three, a bidirectional symmetric search algorithm is used for hyperparameter optimization. In the bidirectional symmetric search algorithm, the position update formula for the explored individual is as follows:

[0034] ,

[0035] in, This is a warning value. For real-time security thresholds, For the first The real-time step size coefficient of an individual

[0036] The expression is: ,

[0037] in, This is the ranking decay factor.

[0038] ,

[0039] ,

[0040] As the initial threshold, As the current population centroid, The current global optimum is at the th The position of the dimension The worst case in the current global context is at the 1st position. The position of the dimension Let be a random variable that follows a heavy-tailed distribution;

[0041] when At that time, individuals can perform a wide range of search operations;

[0042] like ≥ This indicates that the individual is currently outside the safety threshold and needs to be moved to a safe area.

[0043] The cooperating individual updates its position by monitoring the position changes of the exploring individual. Once it detects that the exploring individual has found a suitable place, it will immediately leave its current position to assist the exploring individual.

[0044] The formula for updating the position of collaborative individuals is as follows:

[0045] ,

[0046] in, For random numbers that follow a normal distribution, To explore the optimal position occupied by the individual at present, for A matrix, wherein each element is randomly assigned a value. or ,and , for All of Matrix;

[0047] when At that time, the first with a lower safety value Collaborating individuals need to travel to other places to explore;

[0048] when ≤ At that time, cooperative individuals with higher safety values ​​will follow the exploring individuals;

[0049] The reconnaissance individual detects dangers in the environment, and the location update formula for the reconnaissance individual is as follows:

[0050] ,

[0051] in, These are step size control parameters. It is a random number. This is the current individual's safety value. and The current global best and worst safe values, It is the smallest constant;

[0052] when This indicates that the individual is currently at the edge of the population and needs to move towards the globally optimal position.

[0053] when At this time, the current individual needs to move closer to other individuals.

[0054] This invention provides an adaptive regression modeling and analysis method based on hyperspectral phenotypic data of rice metabolism. It has the following beneficial effects:

[0055] 1. This invention uses a bidirectional symmetric search algorithm to optimize the hyperparameters of the modeling algorithm. Based on different rice metabolic hyperspectral phenotypic data, it adaptively searches and selects the optimal hyperparameters of various algorithms to enhance the performance of the model.

[0056] 2. This invention introduces a bidirectional symmetric search mechanism, enabling the search entity to move towards the optimal solution during the search process and to perform a mirror-like reverse exploration on the other side of the search space, effectively avoiding getting trapped in local optima. This mechanism ensures the diversity of the search and allows for a comprehensive exploration of the solution space.

[0057] 3. This invention utilizes the training and test sets of eight regression models established after hyperparameter optimization. The difference is used to automatically select the optimal regression model and automatically output the optimal regression model, effectively improving modeling efficiency and model accuracy. Attached Figure Description

[0058] Figure 1 This is a flowchart of the present invention;

[0059] Figure 2 This is a flowchart of the process of the present invention. Detailed Implementation

[0060] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0061] The present invention will now be described in detail with reference to the accompanying drawings:

[0062] Example:

[0063] Please see the appendix Figure 1 and attached Figure 2 This invention provides an adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data, comprising:

[0064] Step 1: Collect hyperspectral data of rice and perform image processing to extract features, including the following steps:

[0065] 1. Select rice samples and conduct hyperspectral data acquisition in a dark room with halogen lamps as the sole light source to ensure the consistency of the light source environment;

[0066] 2. The rice hyperspectral data stored in BIL format was remodeled, and a binary map was obtained by threshold segmentation to extract the grain region;

[0067] 3. Based on the hyperspectral data and binary map of the reorganized rice sample, extract the hyperspectral data in the grain region, input it into a two-branch convolutional neural network model, and output 256-dimensional fusion features;

[0068] Step two involves determining the true content of the rice sample using high-performance liquid chromatography (HPLC) as a reference measurement. This step includes the following steps:

[0069] Rice samples were ground into powder, dissolved in solvent, and the content data of 887 rice grain metabolites were obtained by high performance liquid chromatography.

[0070] Step 3: Eight methods are used to perform regression modeling on the rice metabolic hyperspectral phenotypic dataset. A bidirectional symmetric search algorithm is used to optimize hyperparameters for different types of metabolites, and the optimal hyperparameters for various regression models are output. This includes the following steps:

[0071] 1. Integrate rice hyperspectral characteristic data with metabolite content data to obtain a dataset for modeling and analyzing rice metabolic hyperspectral phenotypic data;

[0072] 2. The following eight methods are used to build regression models: Lasso Regression, Support Vector Machine Regression, Random Forest Regression, Nearest neighbor regression, decision tree regression, partial least squares regression, ridge regression, and extreme gradient boosting regression;

[0073] 3. A bidirectional symmetric search algorithm is used to find the optimal hyperparameters for each regression model, with the number of individuals in the population set to be... The proportions of exploratory, cooperative, and scout individuals in the population are: , and Calculate the safety value of each individual and use a bidirectional symmetric search algorithm to output the optimal hyperparameter results of each regression model;

[0074] Step four, based on the optimal hyperparameters of various models obtained in step three, and according to the training and test sets of various regression models... The optimal regression model after hyperparameter tuning is selected by using the difference, and the optimal regression model is output as the final model, including the following steps:

[0075] 1. Calculate the metabolites under optimal hyperparameters on the training and test sets of various regression models. Difference, if there exists a difference less than The model consists of differences less than Selecting the test set from the model The largest model is selected as the optimal model, and the filtering process ends at this point.

[0076] 2. If there is no difference less than The model, select the test set The largest model is taken as the optimal model;

[0077] Step 5: After obtaining the optimal model, perform genome-wide association analysis on the features extracted by the convolutional neural network model and the metabolite content, including the following steps:

[0078] 1. Using features extracted by the convolutional neural network model as surrogate phenotypes, genome-wide association analysis was performed using a mixed linear model. Principal components and kinship matrices were introduced into the model to control false positives and screen for preliminary significant loci.

[0079] 2. Using the metabolite content determined by high-performance liquid chromatography as the baseline phenotype, fine localization was performed within the initially selected site range, with regional significance thresholds used. Correction; candidate genes were screened by combining rice genome annotation information, and genes that were all located in the peak region of the Manhattan plot in the phenotype were selected.

[0080] By constructing a convolutional neural network model with both spectral and spatial branches and an attention mechanism, deep fusion features in hyperspectral data are extracted. The advantage lies in its ability to fully exploit the rich information contained within phenotypic data. Compared to traditional methods that rely on commercial software for standard analysis, this customized deep learning model can simultaneously capture subtle differences in the spectral response and spatial texture features of rice grains. Furthermore, by enhancing the weight of key information through the attention mechanism, it extracts representative features, laying a solid data foundation for subsequent accurate modeling and addressing the problem of insufficient information extraction in traditional methods.

[0081] High-performance liquid chromatography (HPLC), the "gold standard" method, was used to accurately determine the true content of 887 metabolites in rice grains. The advantage is that it provides an accurate and reliable benchmark for the modeling and analysis process. Hyperspectral prediction models require supervised learning and performance validation using real-world labeled data. The precise metabolite content obtained through HPLC ensures that the model's training objectives are clear and consistent with biological reality, thus guaranteeing the practical significance and high reliability of the final model's predictions.

[0082] For eight different regression models, an innovative bidirectional symmetric search algorithm is employed for adaptive hyperparameter optimization. The advantages include: addressing the problems of traditional modeling and analysis relying on commercial software and lacking hyperparameter optimization capabilities, leading to poor model performance. By adaptively searching for optimal hyperparameters based on the characteristics of different rice metabolic data, the prediction accuracy of each candidate model is significantly improved; and the shortcomings of traditional search algorithms that easily get trapped in local optima are overcome. By introducing a bidirectional symmetric search mechanism, the algorithm can conduct a more comprehensive exploration within the solution space, effectively avoiding premature convergence and finding the globally optimal or near-globally optimal hyperparameter combination, ensuring the robustness of the model.

[0083] Establish an automated model selection criterion based on the performance difference between the training and test sets. The advantage lies in effectively improving modeling efficiency and the model's generalization ability. It replaces the tedious process of manually selecting and comparing different model results in traditional methods, achieving automatic output of the optimal model. Importantly, by using "generalization ability" as the core evaluation criterion, it prioritizes models that perform well on both the training and test sets, ensuring that the final model not only performs excellently on known data but also maintains high accuracy and stability when applied to new, unknown data, thus addressing the problems of poor generalization and lack of universality in traditional models.

[0084] Combining hyperspectral prediction models with genetic analysis for candidate gene discovery offers the advantage of enabling deep association analysis from high-throughput phenotypes to genotypes, thus expanding the application value of the technology. Deep features extracted by the model are used as "surrogate phenotypes" for initial screening, improving analytical efficiency; then, the actual content determined by high-performance liquid chromatography (HPLC) is used as a "benchmark phenotype" for fine-tuning, ensuring the accuracy of the results. This "initial screening-fine-tuning" strategy efficiently utilizes hyperspectral data and enhances the reliability of candidate genes through dual validation, providing strong technical support for molecular breeding and trait improvement in rice.

[0085] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data, characterized in that, include: Step 1: Collect rice grain samples Hyperspectral data across a wavelength range was used to segment grain regions based on grain morphology features. A dual-branch convolutional neural network model was employed to extract features, comprising a spectral branch and a spatial branch. The spectral branch utilized a three-dimensional convolutional neural network architecture and embedded residual dense blocks. The spectral attention mechanism uses a two-dimensional convolutional neural network architecture for the spatial branch and embeds residual dense blocks. The spatial attention mechanism uses a multi-head cross-attention fusion module to fuse the features output from the spectral branch and the spatial branch as keys and values, and outputs the fused features. Step 2: High performance liquid chromatography (HPLC) is used to target and determine the content of metabolites in rice grain samples from the same batch to obtain the true metabolite content values ​​of rice grain samples. Step 3: Establish a metabolite content-hyperspectral feature association database, and construct a database based on metabolite content and fusion features. A bidirectional symmetric search algorithm is used to optimize the hyperparameters of various regression models and output the optimal hyperparameters for each regression model. Step 4: Select the optimal regression model based on the evaluation of the model's generalization ability. Construct each regression model with the output optimal hyperparameters, calculate the difference between the coefficients of determination of each model's training set and test set, and select the optimal regression model as the final prediction model based on the difference. Step 5: Using the extracted convolutional neural network features as surrogate phenotypes, a mixed linear model genome-wide association analysis is performed. Principal component analysis and phylogenetic matrix are introduced to control false positives and screen for initially significant loci. The measured metabolite content is used as the baseline phenotype for fine-tuning within the initially selected loci, employing regional analysis. A significance threshold was set for calibration, and candidate genes that were simultaneously located in the peak region of the phenotypic Manhattan plot were screened in combination with rice genome annotation information.

2. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 1, characterized in that, The following is used in step three: There are several ways to build a regression model, including: lasso regression, support vector machine regression, random forest regression, etc. Nearest neighbor regression, decision tree regression, partial least squares regression, ridge regression, and extreme gradient boosting regression.

3. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 1, characterized in that, In step three, the bidirectional symmetric search algorithm initializes the population and sets the proportions of explorers, collaborators, and scouts. It calculates the safety value of each individual based on the negative mean square error function and assigns individual roles according to the size of the safety value. The individual with the highest safety value is assigned as an explorer responsible for global search, the individual with the lowest safety value is assigned as a scout responsible for environmental reconnaissance, and the remaining individuals are assigned as collaborators responsible for assisting in exploration.

4. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 3, characterized in that, In step three, when an individual's position exceeds the search boundary, a reflection mechanism is used. The difference between the individual's position and the boundary is calculated for reverse mapping, and perturbation noise is added to avoid boundary adhesion. Finally, a truncation operation is performed to ensure that all individual positions are limited to the legal range of the search space.

5. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 1, characterized in that, In step four, the criterion for selecting the optimal model is that when there exists a difference less than 1 / 3... When selecting a model, the model with the highest determination coefficient on the test set is chosen as the optimal model, provided that all the differences are greater than or equal to the optimal model. When the test set has the highest coefficient of determination, the model with the highest coefficient of determination is selected as the optimal model.

6. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 1, characterized in that, In step five, a regression model is established to compare the hyperspectral characteristics with metabolite content, and a coefficient of determination greater than [value missing] is selected. The characteristics of the surrogate phenotype are used as surrogate phenotypes to ensure that the surrogate phenotypes are highly correlated with the actual metabolite content for subsequent association analysis. In step five, the preliminary fine positioning is performed upstream and downstream of the initially selected significant site. The study was conducted within a specific range, and regional variations were determined by calculating the number of local single nucleotide polymorphism sites. A significance threshold was set for calibration, and sites that were significant in both surrogate and baseline phenotype genome-wide association analyses were selected as final candidate regions.

7. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 4, characterized in that, The formula processed by the reflection mechanism is as follows: ， in, and For the upper and lower bounds of the search space, For minor noise, , Noise intensity factor; Subsequent approval The operation ensures that the position of each entity is restricted to the legal area of ​​the search space. The specific formula for the operation is as follows: , in, The first in the population Individual in the first Location information in 3D space.

8. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 3, characterized in that, In step three, a bidirectional symmetric search algorithm is used for hyperparameter optimization. The initialization of the bidirectional symmetric search algorithm specifically includes: Set the population size as Each individual has Bit hyperparameters For the first In the nth iteration Individual in the first Location information in the dimension Population matrix; The position of each individual in the population is represented as follows: ， Set the ratio of exploring individuals, cooperative individuals, and scout individuals, and calculate the safety value.

9. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 8, characterized in that, The location update of the explored individuals is controlled by a real-time step size coefficient. When the individual's safety value is greater than or equal to the product of the safety threshold and the maximum safety value, the step size is calculated based on the search boundary. When the individual's safety value is less than the product of the safety threshold and the maximum safety value, the step size is calculated based on the distance between the population centroids. A random variable that follows a heavy-tailed distribution is introduced to enhance the jump search capability. The cooperative individual updates its position by monitoring the position changes of the exploring individual. It takes the current optimal position of the exploring individual as the target position and calculates the new position by combining random numbers that follow a normal distribution and a random assignment matrix. When the safety value of the cooperative individual is lower than the average safety value, it explores other areas. When the safety value of the cooperative individual is higher than the average safety value, it moves with the exploring individual. The reconnaissance individual determines its movement direction based on a comparison between its current safety value and the average of the global best and worst safety values. It then calculates its position update using step size control parameters and random numbers following a standard normal distribution. When an individual's safety value is higher than the global average safety value, it moves toward the global best position. When an individual's safety value is lower than the global average safety value, it moves toward other individuals.

10. The adaptive regression modeling and analysis method based on rice metabolic hyperspectral phenotypic data according to claim 9, characterized in that, In step three, a bidirectional symmetric search algorithm is used to optimize hyperparameters. The position update formula for the explored individual in this bidirectional symmetric search algorithm is as follows: ， in, This is a warning value. For real-time security thresholds, For the first The real-time step size coefficient of an individual The expression is: ,in, This is the ranking decay factor. , , As the initial threshold, As the current population centroid, The current global optimum is at the th The position of the dimension The worst case in the current global context is at the 1st position. The position of the dimension Let be a random variable that follows a heavy-tailed distribution; when At that time, individuals can perform a wide range of search operations; like ≥ This indicates that the individual is currently outside the safety threshold and needs to be moved to a safe area. During the exploration process, the collaborating individual will monitor the exploring individual, and once it detects that the exploring individual has found a suitable place, it will immediately leave its current position to assist the exploring individual; The formula for updating the position of collaborative individuals is as follows: ， in, For random numbers that follow a normal distribution, To explore the optimal position occupied by the individual at present, for A matrix, wherein each element is randomly assigned a value. or ,and , for All of Matrix; when At that time, the first with a lower safety value Collaborating individuals need to travel to other places to explore; when ≤ At that time, cooperative individuals with higher safety values ​​will follow the exploring individuals; The reconnaissance individual detects dangers in the environment, and the location update formula for the reconnaissance individual is as follows: ， in, These are step size control parameters. It is a random number. This is the current individual's safety value. and The current global best and worst safe values, It is the smallest constant; when This indicates that the individual is currently at the edge of the population and needs to move towards the globally optimal position. when At this time, the current individual needs to move closer to other individuals.

Images