A comprehensive evaluation method and evaluation system for heat tolerance of rice

Abstract

The application discloses a kind of rice heat tolerance comprehensive evaluation method and evaluation system, it is related to rice heat tolerance evaluation technical field, including: extracting heat tolerance evaluation index in rice heat tolerance evaluation instance data, according to the interactive relationship of each heat tolerance evaluation index and rice heat tolerance evaluation task to build interactive diagram, using graph neural network to filter key heat tolerance evaluation index;Using domain knowledge graph to carry out knowledge enhancement to key heat tolerance evaluation index, generate expansion heat tolerance evaluation index to supplement key heat tolerance evaluation index, and select the best key heat tolerance evaluation index combination by feature selection;Based on data reconstruction network, build rice heat tolerance comprehensive evaluation model, obtain the rice heat tolerance comprehensive evaluation result of the rice to be measured.The application combines key heat tolerance evaluation index and data reconstruction model to build rice heat tolerance evaluation model, realizes the automation and intelligentization of rice heat tolerance evaluation, improves rice heat tolerance evaluation precision.

Description

Technical Field

[0001] This invention relates to the field of rice heat resistance evaluation technology, and more specifically, to a comprehensive evaluation method and system for rice heat resistance. Background Technology

[0002] Rice is one of the world's most important food crops. With global warming, high-temperature stress has become one of the main factors affecting rice yield and quality. High temperatures can lead to problems such as stunted growth and development, pollen sterility, and poor grain filling, severely reducing yield. Therefore, studying the heat tolerance of rice and breeding heat-resistant varieties is of great significance for ensuring food security. High-temperature stress inhibits root development and tillering in rice, damages chloroplast structure, reduces photosynthetic efficiency, leads to stunted plants, and affects carbohydrate accumulation. In addition, high temperatures can cause pollen sterility, reduce seed setting rate, seriously affect yield, and reduce rice quality.

[0003] Traditional methods for evaluating rice heat tolerance mainly rely on field trials and manual observation. These methods involve planting rice under natural high-temperature conditions and observing its growth, development, and yield performance, as well as simulating high-temperature conditions in artificial climate chambers by controlling temperature and time to evaluate heat tolerance. Evaluation indicators typically include morphological indicators such as plant height, tiller number, and leaf area, as well as physiological and biochemical indicators such as chlorophyll content, photosynthetic rate, and antioxidant enzyme activity. However, these methods lack a scientifically comprehensive evaluation system for rice heat tolerance, leading to problems such as poor consistency in evaluation results and small sample sizes. Therefore, developing an intelligent heat tolerance evaluation system using artificial intelligence and big data technologies to achieve a comprehensive evaluation of rice heat tolerance is a pressing issue that needs to be addressed. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a comprehensive evaluation method and system for rice heat resistance. By combining key heat resistance evaluation indicators and a data reconstruction model, a rice heat resistance evaluation model is constructed, thereby achieving automation and intelligence in the evaluation of rice heat resistance.

[0005] The first aspect of this invention provides a method for comprehensive evaluation of heat resistance in rice, comprising the following steps: Retrieve and obtain rice heat resistance evaluation example data, extract heat resistance evaluation indicators from the rice heat resistance evaluation example data, and construct a heat resistance evaluation indicator set; An interaction graph is constructed based on the interaction relationship between each heat resistance evaluation index in the heat resistance evaluation index set and the rice heat resistance evaluation task. A graph neural network is used to perform representation learning on the interaction graph to screen key heat resistance evaluation indicators. The domain knowledge graph is used to enhance the knowledge of key heat resistance evaluation indicators, generate extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators, and select the best combination of key heat resistance evaluation indicators. A comprehensive evaluation model for rice heat resistance was constructed based on the optimal combination of key heat resistance evaluation indicators. The index parameter values ​​of the rice to be tested were collected as model inputs to obtain the comprehensive evaluation results of the rice heat resistance of the rice to be tested.

[0006] In this scheme, rice heat resistance evaluation instance data is retrieved, heat resistance evaluation indicators are extracted from the rice heat resistance evaluation instance data, and a heat resistance evaluation indicator set is constructed, specifically as follows: Using rice heat resistance evaluation as a keyword, relevant descriptive texts of rice heat resistance evaluation were obtained. A pre-trained BERT model was used for word embedding, and an attention mechanism was used to obtain the global context of different words in the relevant descriptive texts, and the corresponding global context embeddings were obtained. In the relevant descriptive text, semantic relative distance is obtained based on the positional information of different word vectors and keyword vectors. Words with a relative semantic distance less than a preset threshold are selected as local contexts. The local contexts are dynamically weighted to obtain local context embedding. The global context embedding and local context embedding are concatenated and imported into a fully connected layer to obtain the corresponding semantic feature sequence. The semantic feature sequence is used in combination with the IVF algorithm and the HNSW algorithm to construct a retrieval index. The HNSW structure is applied to the inverted index to generate a rice heat resistance evaluation index. The similarity between data is obtained by using the rice heat resistance evaluation index combined with the cosine distance metric function, and rice heat resistance evaluation instance data that meet the similarity criteria are obtained. The rice heat resistance evaluation instance data is then processed in a structured manner. Heat resistance evaluation indicators were extracted from the structured data of rice heat resistance evaluation examples, and a set of heat resistance evaluation indicators was constructed.

[0007] In this solution, a domain knowledge graph is used to enhance the knowledge of key heat resistance evaluation indicators, generating extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators. Specifically: Access the knowledge graph related to rice heat resistance evaluation, and use key heat resistance evaluation indicators to obtain the corresponding knowledge graph subgraphs in the domain knowledge graph; The problem input is "to obtain extended indicators of key heat resistance evaluation indicators". The problem is vectorized and features are extracted. A multi-head attention mechanism is used to obtain fine-grained information of the problem vectorization representation to obtain intermediate inference information. An intermediate indicator vector is generated based on intermediate inference information. Attention is combined with the intermediate indicator vector to obtain a final indicator vector. The final indicator vector is used to perform local path matching in the knowledge graph subgraph. The entity relationships in the local path are imported into the BERT model to obtain the local path embedding. The local path embedding is scored for similarity using the indicator vector. The entity score is updated using the similarity score. The entity scores of each hop are stacked, multiplied by the attention, and summed to obtain the final entity score. Based on the final entity score, entities that meet the standards are selected to generate extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators.

[0008] In this scheme, the optimal combination of key heat resistance evaluation indicators is selected as follows: The key heat resistance evaluation indicators were obtained and encoded. An initial population was constructed by randomly generating an initial combination of heat resistance evaluation indicators. The combination of heat resistance evaluation indicators was optimized by a genetic algorithm. A classification model was constructed using an SVM model. The feature vectors corresponding to the initial combination of heat resistance evaluation indicators were imported into the classification model to obtain the classification accuracy. The fitness function was constructed using the classification accuracy. The fitness of the heat resistance evaluation index combination is calculated according to the fitness function. A preset number of heat resistance evaluation index combinations are selected and replicated based on the fitness to expand the population. The selected heat resistance evaluation index combinations are then cross-recombined to change the coding of individual heat resistance evaluation indexes. The optimal solution is obtained by iteratively optimizing the population, and the best combination of key heat resistance evaluation indicators is output.

[0009] In this scheme, a comprehensive evaluation model for rice heat resistance is constructed based on the optimal combination of key heat resistance evaluation indicators, specifically as follows: Based on the combination of key heat resistance evaluation indicators, single-indicator evaluation data are extracted from the rice heat resistance evaluation example data, and data labels are generated using the rice heat resistance evaluation results. A comprehensive evaluation model for rice heat resistance was constructed using a conditional variational autoencoder. The model was trained using labeled single-index evaluation data. The latent variable distribution of the training samples was obtained through the encoder, and the data was reconstructed using the latent variable distribution through the decoder. By combining the labels of rice heat resistance evaluation results with latent variable distribution to train the discriminator, the predicted single index evaluation data under the specific rice heat resistance evaluation result labels generated by data reconstruction is obtained, and the discriminator is used to determine the data labels. The trained conditional variational autoencoder is obtained through alternating training. The predicted single-index evaluation data generated by the conditional variational autoencoder is imported into the multilayer perceptron to obtain the deviation from the test data. Based on the deviation, the label classification result of the rice heat resistance evaluation result is obtained. When the classification accuracy meets the preset requirements, the comprehensive evaluation model of rice heat resistance is output.

[0010] In this scheme, the index parameter values ​​of the rice to be tested are collected as model inputs to obtain the comprehensive evaluation results of the rice's heat resistance, specifically: The index parameter values ​​of the rice to be tested were collected using the optimal combination of key heat resistance evaluation indicators. The index parameter values ​​were then imported into the comprehensive evaluation model of rice heat resistance to obtain the estimated index parameter values ​​under each rice heat resistance evaluation result label. The residual between the predicted index parameter value and the index parameter value is obtained by multilayer perceptron, the label of the rice heat resistance evaluation result corresponding to the minimum residual is obtained, and the heat resistance evaluation result of the rice to be tested is output.

[0011] A second aspect of the present invention provides a comprehensive evaluation system for rice heat resistance, comprising: an evaluation index screening module, an evaluation index combination optimization module, a data acquisition module, a rice heat resistance evaluation module, and an evaluation result output module; The evaluation index screening module is responsible for extracting heat resistance evaluation indicators from rice heat resistance evaluation instance data, constructing an interaction graph based on the interaction relationship between each heat resistance evaluation indicator and the rice heat resistance evaluation task, using a graph neural network to perform representation learning on the interaction graph, and screening key heat resistance evaluation indicators. The evaluation index combination optimization module uses a domain knowledge graph to enhance the knowledge of key heat resistance evaluation indicators, constructs extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators, and selects the best combination of key heat resistance evaluation indicators through feature selection. The data acquisition module is responsible for collecting the index parameter values ​​of the rice to be tested based on the optimal combination of key heat resistance evaluation indicators. The rice heat resistance evaluation module constructs a comprehensive evaluation model for rice heat resistance based on a data reconstruction model combined with the optimal key heat resistance evaluation index combination, and imports index parameter values ​​to obtain the comprehensive evaluation results of the rice heat resistance of the rice to be tested. The evaluation result output module is responsible for outputting and visualizing the comprehensive evaluation results of the heat resistance of the rice to be tested in a predictive manner.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes graph neural networks to capture the complex relationships between heat tolerance evaluation indicators, identifying the key indicators that have the greatest impact on rice heat tolerance and improving the relevance of the evaluation. Combining key heat tolerance evaluation indicators and a data reconstruction model, the constructed rice heat tolerance evaluation model exhibits high accuracy and robustness. While ensuring the consistency of rice heat tolerance evaluation results, it reduces reliance on human experience, achieving automation and intelligence in rice heat tolerance evaluation. The rice heat tolerance evaluation model constructed in this invention is characterized by high efficiency, accuracy, and intelligence, providing a data foundation for multiple fields such as rice breeding, agricultural research, and agricultural production by acquiring precise rice heat tolerance evaluation data. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the accompanying 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.

[0014] Figure 1 A flowchart of a comprehensive evaluation method for rice heat resistance is shown; Figure 2 The flowchart illustrating the construction of a comprehensive evaluation model for rice heat resistance is shown in this embodiment. Figure 3 A block diagram of a comprehensive evaluation system for rice heat resistance is shown. Detailed Implementation

[0015] 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.

[0016] 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.

[0017] Figure 1 A flowchart of a comprehensive evaluation method for rice heat resistance is shown.

[0018] like Figure 1 As shown in the figure, this embodiment provides a comprehensive evaluation method for the heat resistance of rice, including: S102, retrieve and obtain rice heat resistance evaluation instance data, extract heat resistance evaluation indicators from the rice heat resistance evaluation instance data, and construct a heat resistance evaluation indicator set; S104, construct an interaction graph based on the interaction relationship between each heat resistance evaluation index in the heat resistance evaluation index set and the rice heat resistance evaluation task, and use a graph neural network to perform representation learning on the interaction graph to screen key heat resistance evaluation indicators. S106, Use domain knowledge graph to enhance the knowledge of key heat resistance evaluation indicators, generate extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators, and select the best combination of key heat resistance evaluation indicators. S108. A comprehensive evaluation model for rice heat resistance is constructed based on the optimal combination of key heat resistance evaluation indicators. The index parameter values ​​of the rice to be tested are collected as model inputs to obtain the comprehensive evaluation results of the rice heat resistance of the rice to be tested.

[0019] It should be noted that, using "rice heat resistance evaluation" as a keyword, relevant descriptive text for rice heat resistance evaluation is obtained. A pre-trained BERT model is used for word embedding, and an attention mechanism is used to obtain the global context of different words in the relevant descriptive text, thus obtaining the corresponding global context embedding. Semantic relative distance is obtained in the relevant descriptive text based on the positional information of different word vectors and keyword vectors. Words with a relative semantic distance less than a preset threshold are selected as local contexts. These local contexts are then dynamically weighted to obtain local context embeddings. The formula for calculating the semantic relative distance d is as follows: , These represent the positional information of the word vector and the keyword vector, respectively, with m representing the length of the keyword vector.

[0020] The global and local context embeddings are concatenated and imported into a fully connected layer to obtain the corresponding semantic feature sequence. This semantic feature sequence is then used in conjunction with the IVF and HNSW algorithms to construct a retrieval index. The HNSW structure is applied to the inverted index to generate an index for evaluating rice heat resistance. The IVF algorithm divides the feature space into several partitions, each represented by a centroid. For a query feature vector, the distance to all centroids is calculated, and an exact match search is performed from the partitions containing the n nearest centroids. The HNSW algorithm constructs a multi-level NSW structure. Starting from any node at the top level, it searches for the node closest to the query point in each level as the entry node for the next level, continuing until the bottom level. The HNSW algorithm is applied to the nearest neighbor centroid search of the IVF algorithm to achieve a combination of algorithms. The similarity between data is obtained by using a rice heat resistance evaluation index combined with a cosine distance metric function. The closer the distance value is to 0, the more it meets the preset standard. Rice heat resistance evaluation instance data that meets the similarity standard is obtained, and the rice heat resistance evaluation instance data is structured. The relevant heat resistance evaluation indicators are extracted from the structured rice heat resistance evaluation instance data to construct a heat resistance evaluation indicator set.

[0021] It should be noted that a domain knowledge graph is used to augment the key heat resistance evaluation indicators. This involves accessing a knowledge graph related to rice heat resistance evaluation and retrieving corresponding subgraphs from the domain knowledge graph for each key heat resistance evaluation indicator. The question "obtaining extended indicators for key heat resistance evaluation indicators" is used as input. Pre-trained networks such as BERT are used for question vectorization and semantic feature extraction. During each hop of reasoning, the question representation and the indicator vector from the previous hop are first processed through a multi-head attention mechanism to obtain fine-grained information of the question vectorization, resulting in intermediate reasoning information. The acquired attention weights are used to characterize which part of the current question is more important. Indicator vectors are generated based on different parts of the question to guide dynamic reasoning and path expansion. An intermediate indicator vector is generated based on the intermediate reasoning information. During each reasoning iteration, more attention is paid to the current information. A sigmoid activation function is used to output a random value, which is multiplied by historical reasoning information to filter redundant information from the previous hop. Indicator vectors are generated by combining past and present key reasoning information. Finally, the attention weights of each word vector in each question are combined with the intermediate indicator vectors to obtain the final indicator vector.

[0022] The final indicator vector is used to perform local path matching in the knowledge graph subgraph. Entity relationships within the local paths are imported into the BERT model to obtain local path embeddings. The indicator vectors are used to score the similarity of the local path embeddings, and the entity scores are updated using the similarity scores. The previous hop indicator vector and the local path embedding are multiplied and passed through a fully connected layer. The score is normalized using a sigmoid activation function. The entity score of the previous hop is multiplied by the local path score to obtain the entity score of the current hop. Attention weights for each hop are obtained based on all indicator vectors. The entity scores of each hop are stacked, multiplied by the attention weights of each hop, and summed to obtain the final entity score. Entities meeting the criteria are selected based on the final entity score to generate extended heat resistance evaluation indicators to supplement key heat resistance evaluation indicators. Preferably, knowledge enhancement is performed using the seed setting rate. Relevant knowledge is obtained through the knowledge graph, such as: obtaining the trend of seed setting rate changes under different temperature treatments; calculating the slope of the overall trend line (linear fitting); using the ratio of the overall slope of the tested rice variety to the overall slope of the control variety as a heat resistance coefficient; and judging the heat resistance of the tested variety based on the heat resistance coefficient. Therefore, the overall trend line of the change in the seed set rate is used as the extended heat resistance evaluation index of the seed set rate.

[0023] It should be noted that the supplemented key heat resistance evaluation indicators are encoded, and an initial population is constructed by randomly generating initial combinations of heat resistance evaluation indicators. A genetic algorithm is used to optimize these combinations, and an SVM model is used to construct a classification model. The feature vectors corresponding to the initial heat resistance evaluation indicator combinations are imported into the classification model, and an S-fold cross-validation strategy is employed to obtain the classification accuracy. The classification accuracy is then used to construct a fitness function. The fitness of the heat resistance evaluation indicator combinations is calculated based on this fitness function. A predetermined number of heat resistance evaluation indicator combinations are selected and replicated based on the fitness to expand the population and determine the optimization direction. The selected heat resistance evaluation indicator combinations are then cross-recombined, merging important heat resistance evaluation indicators to accelerate the optimization selection of feature combinations. Furthermore, the encoding of individual heat resistance evaluation indicators is modified to increase the possibility of new feature combinations and avoid local optima. Through iterative optimization of the population, the optimal solution is obtained when the planting criterion is met, indicating that the classification accuracy of the heat resistance evaluation indicator combination has reached an ideal state, and the optimal key heat resistance evaluation indicator combination is output.

[0024] Figure 2 The flowchart illustrating the construction of a comprehensive evaluation model for rice heat resistance is shown in this embodiment.

[0025] According to an embodiment of the present invention, a comprehensive evaluation model for rice heat resistance is constructed based on the optimal combination of key heat resistance evaluation indicators, specifically as follows: S202, based on the key heat resistance evaluation indicators in the optimal combination of key heat resistance evaluation indicators, extract single indicator evaluation data from the rice heat resistance evaluation example data, and use the rice heat resistance evaluation results to generate data labels. S204, a comprehensive evaluation model for rice heat resistance is constructed using a conditional variational autoencoder, the model is trained using labeled single-index evaluation data, the latent variable distribution of the training samples is obtained through the encoder, and the data is reconstructed using the latent variable distribution through the decoder. S206 combines the labels of rice heat resistance evaluation results with the latent variable distribution to train the discriminator, obtain the predicted single index evaluation data under the specific rice heat resistance evaluation result labels generated by data reconstruction, and use the discriminator to determine the data labels. The trained conditional variational autoencoder is obtained through alternating training. S208, the predicted single-index evaluation data generated by the conditional variational autoencoder is imported into the multilayer perceptron to obtain the deviation from the test data, and the rice heat resistance evaluation result label classification result is obtained according to the deviation. When the classification accuracy meets the preset requirements, the comprehensive evaluation model of rice heat resistance is output.

[0026] It should be noted that the Conditional Variational Autoencoder (CVA) decodes data by inputting the latent variables output by the encoder and the corresponding class labels into the decoder. During data generation, the class labels and the decoder control the type of generated data. A discriminator is introduced into the CVA to enhance the model's generation capability and promote the effective use of input class label information during data generation. The training and reconstruction process for the CVA includes training the CVA, using single-index evaluation data, and training the discriminator with reconstructed data. The generation training process involves generating predicted single-index evaluation data for a specific category using specific rice heat resistance evaluation result labels, and then using the discriminator to determine its category.

[0027] The optimal combination of key heat resistance evaluation indicators was used to collect the indicator parameter values ​​of the rice to be tested. These indicator parameter values ​​were then imported into a comprehensive rice heat resistance evaluation model to obtain the estimated indicator parameter values ​​under each rice heat resistance evaluation result label. A multilayer perceptron was used to obtain the residuals between the estimated indicator parameter values ​​and the actual indicator parameter values. The rice heat resistance evaluation result label corresponding to the smallest residual was obtained, and the heat resistance evaluation result of the rice to be tested was output. The preferred rice heat resistance evaluation result labels are divided into strong heat resistance, relatively strong heat resistance, general heat resistance, relatively weak heat resistance, and weak heat resistance. The number of levels can be flexibly modified according to actual conditions for phenotypic observation and description.

[0028] Figure 3 A block diagram of a comprehensive evaluation system for rice heat resistance is shown.

[0029] The second embodiment of the present invention provides a comprehensive evaluation system 3 for rice heat resistance, including: an evaluation index screening module 301, an evaluation index combination optimization module 302, a data acquisition module 303, a rice heat resistance evaluation module 304, and an evaluation result output module 305; The evaluation index screening module is responsible for extracting heat resistance evaluation indicators from rice heat resistance evaluation instance data, constructing an interaction graph based on the interaction relationship between each heat resistance evaluation indicator and the rice heat resistance evaluation task, using a graph neural network to perform representation learning on the interaction graph, and screening key heat resistance evaluation indicators. The evaluation index combination optimization module uses a domain knowledge graph to enhance the knowledge of key heat resistance evaluation indicators, constructs extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators, and selects the best combination of key heat resistance evaluation indicators through feature selection. The data acquisition module is responsible for collecting the index parameter values ​​of the rice to be tested based on the optimal combination of key heat resistance evaluation indicators. The rice heat resistance evaluation module constructs a comprehensive evaluation model for rice heat resistance based on a data reconstruction model combined with the optimal key heat resistance evaluation index combination, and imports index parameter values ​​to obtain the comprehensive evaluation results of the rice heat resistance of the rice to be tested. The evaluation result output module is responsible for outputting and visualizing the comprehensive evaluation results of the heat resistance of the rice to be tested in a predictive manner.

[0030] A third aspect of the present invention provides a computer-readable storage medium comprising a program for a comprehensive evaluation method of rice heat resistance, wherein when the program is executed by a processor, it implements the steps of a comprehensive evaluation method of rice heat resistance.

[0031] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple modules 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 indirect coupling or communication connection through some interfaces, devices, or modules, and can be electrical, mechanical, or other forms. Furthermore, in the various embodiments of the present invention, all functional modules can be integrated into one processing module, or each module can be a separate module, or two or more modules can be integrated into one module; the integrated modules can be implemented in hardware or in the form of hardware plus software functional modules.

[0032] 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.

[0033] Alternatively, if the integrated modules 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.

[0034] 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 changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A comprehensive evaluation method for the heat resistance of rice, characterized in that, Includes the following steps: Retrieve and obtain rice heat resistance evaluation example data, extract heat resistance evaluation indicators from the rice heat resistance evaluation example data, and construct a heat resistance evaluation indicator set; An interaction graph is constructed based on the interaction relationship between each heat resistance evaluation index in the heat resistance evaluation index set and the rice heat resistance evaluation task. A graph neural network is used to perform representation learning on the interaction graph to screen key heat resistance evaluation indicators. The domain knowledge graph is used to enhance the knowledge of key heat resistance evaluation indicators, generate extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators, and select the best combination of key heat resistance evaluation indicators. A comprehensive evaluation model for rice heat resistance was constructed based on the optimal combination of key heat resistance evaluation indicators. The index parameter values ​​of the rice to be tested were collected as model inputs to obtain the comprehensive evaluation results of the rice heat resistance of the rice to be tested. A comprehensive evaluation model for rice heat resistance is constructed based on the optimal combination of key heat resistance evaluation indicators, specifically as follows: Based on the combination of key heat resistance evaluation indicators, single-indicator evaluation data are extracted from the rice heat resistance evaluation example data, and data labels are generated using the rice heat resistance evaluation results. A comprehensive evaluation model for rice heat resistance was constructed using a conditional variational autoencoder. The model was trained using labeled single-index evaluation data. The latent variable distribution of the training samples was obtained through the encoder, and the data was reconstructed using the latent variable distribution through the decoder. By combining the labels of rice heat resistance evaluation results with latent variable distribution to train the discriminator, the predicted single index evaluation data under the specific rice heat resistance evaluation result labels generated by data reconstruction is obtained, and the discriminator is used to determine the data labels. The trained conditional variational autoencoder is obtained through alternating training. The predicted single-index evaluation data generated by the conditional variational autoencoder is imported into the multilayer perceptron to obtain the deviation from the test data. Based on the deviation, the label classification result of the rice heat resistance evaluation result is obtained. When the classification accuracy meets the preset requirements, the comprehensive evaluation model of rice heat resistance is output.

2. The method for comprehensive evaluation of rice heat resistance according to claim 1, characterized in that, Retrieve example data on rice heat resistance evaluation, extract heat resistance evaluation indicators from the example data, and construct a set of heat resistance evaluation indicators, specifically: Using rice heat resistance evaluation as a keyword, relevant descriptive texts of rice heat resistance evaluation were obtained. A pre-trained BERT model was used for word embedding, and an attention mechanism was used to obtain the global context of different words in the relevant descriptive texts, and the corresponding global context embeddings were obtained. In the relevant descriptive text, semantic relative distance is obtained based on the positional information of different word vectors and keyword vectors. Words with a relative semantic distance less than a preset threshold are selected as local contexts. The local contexts are dynamically weighted to obtain local context embedding. The global context embedding and local context embedding are concatenated and imported into a fully connected layer to obtain the corresponding semantic feature sequence. The semantic feature sequence is used in combination with the IVF algorithm and the HNSW algorithm to construct a retrieval index. The HNSW structure is applied to the inverted index to generate a rice heat resistance evaluation index. The similarity between data is obtained by using the rice heat resistance evaluation index combined with the cosine distance metric function, and rice heat resistance evaluation instance data that meet the similarity criteria are obtained. The rice heat resistance evaluation instance data is then processed in a structured manner. Heat resistance evaluation indicators were extracted from the structured data of rice heat resistance evaluation examples, and a set of heat resistance evaluation indicators was constructed.

3. The method for comprehensive evaluation of rice heat resistance according to claim 1, characterized in that, The key heat resistance evaluation indicators are augmented using a domain knowledge graph, generating extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators. Specifically: Access the knowledge graph related to rice heat resistance evaluation, and use key heat resistance evaluation indicators to obtain the corresponding knowledge graph subgraphs in the domain knowledge graph; The problem input is "to obtain extended indicators of key heat resistance evaluation indicators". The problem is vectorized and features are extracted. A multi-head attention mechanism is used to obtain fine-grained information of the problem vectorization to obtain intermediate inference information. An intermediate indicator vector is generated based on intermediate inference information. Attention is combined with the intermediate indicator vector to obtain a final indicator vector. The final indicator vector is used to perform local path matching in the knowledge graph subgraph. The entity relationships in the local path are imported into the BERT model to obtain the local path embedding. The local path embedding is scored for similarity using the indicator vector. The entity score is updated using the similarity score. The entity scores of each hop are stacked, multiplied by the attention weight of each hop, and summed to obtain the final entity score. Based on the final entity score, entities that meet the standard are selected to generate extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators.

4. The method for comprehensive evaluation of rice heat resistance according to claim 1, characterized in that, The optimal combination of key heat resistance evaluation indicators is selected as follows: The key heat resistance evaluation indicators were obtained and encoded. An initial population was constructed by randomly generating an initial combination of heat resistance evaluation indicators. The combination of heat resistance evaluation indicators was optimized by a genetic algorithm. A classification model was constructed using an SVM model. The feature vectors corresponding to the initial combination of heat resistance evaluation indicators were imported into the classification model to obtain the classification accuracy. The fitness function was constructed using the classification accuracy. The fitness of the heat resistance evaluation index combination is calculated according to the fitness function. A preset number of heat resistance evaluation index combinations are selected and replicated based on the fitness to expand the population. The selected heat resistance evaluation index combinations are then cross-recombined to change the coding of individual heat resistance evaluation indexes. The optimal solution is obtained by iteratively optimizing the population, and the best combination of key heat resistance evaluation indicators is output.

5. The method for comprehensive evaluation of rice heat resistance according to claim 1, characterized in that, The index parameter values ​​of the rice to be tested are collected as model inputs to obtain the comprehensive evaluation results of the rice's heat resistance, specifically: The index parameter values ​​of the rice to be tested were collected using the optimal combination of key heat resistance evaluation indicators. The index parameter values ​​were then imported into the comprehensive evaluation model of rice heat resistance to obtain the estimated index parameter values ​​under each rice heat resistance evaluation result label. The residual between the predicted index parameter value and the index parameter value is obtained by multilayer perceptron, the label of the rice heat resistance evaluation result corresponding to the minimum residual is obtained, and the heat resistance evaluation result of the rice to be tested is output.

6. A comprehensive evaluation system for rice heat resistance, characterized in that, The system for implementing the comprehensive evaluation method for rice heat resistance as described in any one of claims 1-5 includes: an evaluation index screening module, an evaluation index combination optimization module, a data acquisition module, a rice heat resistance evaluation module, and an evaluation result output module. The evaluation index screening module is responsible for extracting heat resistance evaluation indicators from rice heat resistance evaluation instance data, constructing an interaction graph based on the interaction relationship between each heat resistance evaluation indicator and the rice heat resistance evaluation task, using a graph neural network to perform representation learning on the interaction graph, and screening key heat resistance evaluation indicators. The evaluation index combination optimization module uses a domain knowledge graph to enhance the knowledge of key heat resistance evaluation indicators, constructs extended heat resistance evaluation indicators to supplement the key heat resistance evaluation indicators, and selects the best combination of key heat resistance evaluation indicators through feature selection. The data acquisition module is responsible for collecting the index parameter values ​​of the rice to be tested based on the optimal combination of key heat resistance evaluation indicators. The rice heat resistance evaluation module constructs a comprehensive evaluation model for rice heat resistance based on a data reconstruction model combined with the optimal key heat resistance evaluation index combination, and imports index parameter values ​​to obtain the comprehensive evaluation results of the rice heat resistance of the rice to be tested. The evaluation result output module is responsible for outputting and visualizing the comprehensive evaluation results of the heat resistance of the rice to be tested in a predictive manner.

Images