Forage germplasm resource evaluation method based on deep learning

Abstract

The invention relates to the technical field of forage grass germplasm, and discloses a forage grass germplasm resource evaluation method based on deep learning, comprising the following steps: S1, collecting related data of forage grass germplasm resources from multiple channels, including but not limited to morphological characteristics, physiological indexes and other data of forage grass, ensuring diversity and representativeness of the data, forage grass germplasm resources of different types, geographical sources and growing environments are covered. According to the forage grass germplasm resource evaluation method based on deep learning, by increasing forage grass germplasm data sources, increasing scientific research institutions, colleges and national germplasm resource libraries and collecting scientific research germplasm resource data, a large amount of crop forage grass germplasm and wild forage grass germplasm information is included, forage grass germplasm data sources are increased, and the quality of the forage grass germplasm resources is improved. The characteristics of multiple dimensions such as morphology, physiology, genetics and the like of the pasture are considered at the same time, comprehensive and systematic evaluation is carried out, and a more comprehensive and comprehensive resource evaluation result is provided by carrying out normalization processing on the obtained data.

Description

Technical Field

[0001] The present invention relates to the field of forage germplasm technology, and specifically to a forage germplasm resource evaluation method based on deep learning. Background Art

[0002] With the continuous development of agricultural and pasture research, the evaluation of forage germplasm resources has become increasingly important. Traditional forage germplasm resource evaluation methods mainly rely on manual observation and measurement, which is not only inefficient but also easily affected by subjective factors, resulting in limited accuracy and reliability of evaluation results. In recent years, with the advancement of biotechnology, a large amount of data related to forage germplasm resources has been generated, including morphological characteristics, physiological characteristics, genetic information, etc., which increases the difficulty of integrating and analyzing these multi-source heterogeneous data, resulting in some defects in the original evaluation methods, such as: The existing forage germplasm resource evaluation system often focuses on a single indicator or characteristic and is unable to comprehensively consider the interaction and influence of multiple factors. In addition, the existing model has insufficient training data, incomplete feature extraction, and single evaluation indicators, resulting in low accuracy and reliability of the evaluation results.

[0003] In response to the above problems, it is urgently necessary to carry out innovative designs based on the original forage germplasm resource evaluation. Summary of the invention

[0004] The purpose of the present invention is to provide a forage germplasm resource evaluation method based on deep learning, so as to solve the problem that the existing forage germplasm resource evaluation system mentioned in the above background technology often focuses on a single indicator or characteristic, cannot comprehensively consider the interaction and influence of multiple factors, and the existing model training data is insufficient, feature extraction is not comprehensive, and the evaluation index is single, resulting in low accuracy and reliability of the evaluation results.

[0005] To achieve the above object, the present invention provides the following technical solution: a forage germplasm resource evaluation method based on deep learning, comprising the following evaluation method: S1. Collect relevant data on forage germplasm resources from multiple channels, including but not limited to data on morphological characteristics and physiological indicators of forage, to ensure the diversity and representativeness of the data, covering forage germplasm resources of different types, geographical origins and growth environments; S2. Clean the data collected in S1 above, remove noise, errors and duplicate data, annotate the data, and then standardize or normalize the data so that data with different features are comparable in value; S3. For the features of grass texture, shape, etc. in the image data of grass morphology in S1 above, a convolutional neural network method is used to automatically extract features, and for the gene sequence data, sequence embedding is used to convert it into a feature vector suitable for deep learning model processing; S4. Divide the data preprocessed in S2 and the data after feature extraction in S3 into a training set, a validation set, and a test set, use the training set data to train the selected deep learning model, and optimize the model performance by adjusting the parameters of the model; S5. Use the test set data in S4 to evaluate the trained model, analyze the evaluation results of the model, and consider adjusting the model structure, increasing the amount of data, or improving the training method according to performance requirements; S6 inputs the data of forage germplasm resources to be evaluated into the optimized model to obtain the evaluation results, analyzes and interprets the evaluation results, and provides a decision-making basis for the protection, utilization and breeding of forage germplasm resources.

[0006] The above technical solution is adopted to facilitate the acquisition of corresponding forage germplasm data sources from multiple aspects and increase the data sources for model training.

[0007] Preferably, the sources of the forage germplasm resource data in S1 include: scientific research institutions and universities, and national germplasm resource banks.

[0008] The above technical solution is adopted to facilitate the training and processing within the model by increasing the data source of forage germplasm resources.

[0009] Preferably, the forage germplasm resource data includes a summary of the survey, collection and identification of forage germplasm resources in multiple geographical areas.

[0010] With the above technical solution, as researchers collect and identify different types of forage in multiple geographical areas, forage data covering a wide geographical area is constructed.

[0011] Preferably, in the step of labeling the forage data in S2, the forage germplasm resources are classified or labeled according to specific evaluation criteria.

[0012] By adopting the above technical solution, the forage data is uniformly labeled and standardized.

[0013] Preferably, the deep learning model training work in S4 includes: data preparation, model selection, definition of model architecture, definition of loss function and optimizer, training model and evaluation model, model early stopping and model fine-tuning.

[0014] The above technical solution is used to adjust the learning rate and regularization parameters according to the evaluation results as the deep learning model is trained, thereby improving the model processing effect.

[0015] Preferably, the forage germplasm resources in the data preparation for the deep learning model training include images, feature descriptions, genetic data, and are preprocessed and normalized to form a data set, and the data set is divided into 70% training set, 15% validation set, and 15% test set.

[0016] The above technical solution is adopted to divide the data set for adjustment and processing in coordination with the model data.

[0017] Preferably, in the model selection for the deep learning model training, a convolutional neural network is used for image data, and a recurrent neural network is used for sequence data, and the defined model architecture after model selection determines the number of layers, the number of neurons, and the activation function to define the model architecture.

[0018] Using the above technical solution, corresponding convolutional neural networks and recurrent neural networks are selected to process different data types respectively.

[0019] Preferably, the deep learning model training defines a loss function and an optimizer, selects a suitable loss function according to the evaluation task, cross entropy loss is commonly used for classification tasks, and mean square error loss is commonly used for regression tasks, selects an optimizer, and sets parameters such as the learning rate.

[0020] Using the above technical solution, the deep learning model is further optimized through the stochastic gradient descent of the optimizer and the loss function definition.

[0021] Preferably, in the training model and evaluation model of the deep learning model training, the training set data is input into the model for iterative training. After each training batch, the error is calculated according to the loss function, and the parameters of the model are updated by the optimizer. During the training process, the validation set is regularly used to evaluate the performance of the model, and the learning rate, regularization parameters, etc. are adjusted according to the evaluation results.

[0022] By adopting the above technical solution, the number of iterative training of the deep learning model is increased, thereby improving the performance of the estimation model.

[0023] Preferably, early stopping of the model trained by the deep learning model is used to stop training when the validation set no longer improves after a certain number of rounds to prevent overfitting, and then fine-tune the model according to the results of the validation set, such as increasing the number of layers and adjusting the number of neurons to achieve model fine-tuning.

[0024] By adopting the above technical solution, the overfitting of the model can be effectively avoided by adding early stopping for the validation set.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: the forage germplasm resource evaluation method based on deep learning: 1. By increasing the sources of forage germplasm data, including a large amount of crop forage germplasm and wild forage germplasm information, by increasing scientific research institutions, universities, and national germplasm resource banks, increasing the sources of forage germplasm data, and considering the characteristics of forage in multiple dimensions such as morphology, physiology, and genetics, a comprehensive and systematic evaluation is conducted to provide more comprehensive and comprehensive resource assessment results, providing a scientific basis for the collection, preservation, and innovative use of forage germplasm resources, and contributing to a more reasonable allocation of resources and the formulation of management strategies; 2. By normalizing the acquired data and unifying the data labeling effects, the deep learning model can process and analyze large amounts of complex data, thereby more accurately evaluating the various characteristics and attributes of forage germplasm resources, reducing human errors and subjectivity. At the same time, the model training needs to be gradually iterated and increased, which can not only update the data and models in real time, but also adapt to the dynamic situations such as changes in the forage growth environment and the cultivation of new varieties, maintain the timeliness and effectiveness of the evaluation, and provide reliable scientific support for decision-making in agricultural production, ecological protection and other fields, and ensure the scientificity and rationality of related work. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of the steps of the forage germplasm resource evaluation method of the present invention; Figure 2 It is a brief schematic diagram of the forage germplasm resource evaluation method of the present invention; Figure 3 This is a schematic diagram of the deep learning model training process of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] See also Figure 1-3 The present invention provides a technical solution: a forage germplasm resource evaluation method based on deep learning, including the following evaluation methods: S1. Collect relevant data on forage germplasm resources from multiple channels, including but not limited to data on morphological characteristics and physiological indicators of forage, to ensure the diversity and representativeness of the data, covering forage germplasm resources of different types, geographical origins and growth environments; The sources of forage germplasm resource data in S1 include: scientific research institutions and universities, and national germplasm resource banks. The forage germplasm resource data include the data summary of the investigation, collection and identification of forage germplasm resources in multiple geographical regions; S2. Clean the data collected in S1 above, remove noise, errors and duplicate data, annotate the data, and then standardize or normalize the data so that data with different features are comparable in value; In the process of labeling the forage data in S2, the forage germplasm resources are classified or labeled according to specific evaluation criteria to determine their quality; S3. For the features of grass texture, shape, etc. in the image data of grass morphology in S1 above, a convolutional neural network method is used to automatically extract features, and for the gene sequence data, sequence embedding is used to convert it into a feature vector suitable for deep learning model processing; S4. Divide the data preprocessed in S2 and the data after feature extraction in S3 into a training set, a validation set, and a test set, use the training set data to train the selected deep learning model, and optimize the model performance by adjusting the parameters of the model; S5. Use the test set data in S4 to evaluate the trained model, analyze the evaluation results of the model, and consider adjusting the model structure, increasing the amount of data, or improving the training method according to performance requirements; S6 inputs the data of forage germplasm resources to be evaluated into the optimized model to obtain the evaluation results, analyzes and interprets the evaluation results, and provides a decision-making basis for the protection, utilization and breeding of forage germplasm resources; In conjunction with the accompanying drawings Figure 1-3As shown, when in use, scientific research institutions, universities, and national germplasm resource banks are used to increase the extensive acquisition of forage germplasm resource data. After the camera equipment is used to collect images, the measurement time and frequency of growth data are coordinated with the detection instruments, and the genetic data are obtained in conjunction with the instruments in the laboratory, the forage plant height, branching (tillering), leaf shape, leaf area, biomass, seed yield, stress resistance, photosynthetic efficiency, protein content, ADF, NDF, stress resistance and gene sequence data are obtained to further enrich the diversity and representativeness of the database in the evaluation method, covering forage germplasm resources of different types, geographical origins and growth environments. By cleaning the different collected data, removing noise, errors and duplicate data, the forage germplasm resources are classified or labeled according to specific evaluation standards to achieve standardization or normalization of the data so that different characteristics The data are numerically comparable. After feature extraction, the image data is sent to the convolutional neural network to automatically extract features. The gene sequence data is sent to the sequence embedding technology to convert it into a feature vector suitable for deep learning model processing. The processed data is then sorted and divided into data sets based on each forage grass as a unit. The data sets are then separated into training sets, validation sets, and test sets. The selected deep learning model is trained using the training set data, and the model performance is optimized by adjusting the model parameters. After the deep learning model is trained, the forage grass data to be evaluated is input to obtain the evaluation results. The evaluation results are analyzed and interpreted to provide a decision-making basis for the protection, utilization, and breeding of forage grass germplasm resources, provide reliable scientific support for decision-making in agricultural production, ecological protection and other fields, and ensure the scientificity and rationality of related work. The deep learning model training work in S4 includes: data preparation, model selection, defining model architecture, defining loss function and optimizer, training model and evaluating model, model early stopping and model fine-tuning.

[0029] The forage germplasm resources in the data preparation for deep learning model training include images, feature descriptions, and genetic data, which are preprocessed and normalized to form a data set, and the data set is divided into 70% training set, 15% validation set, and 15% test set.

[0030] In the model selection of deep learning model training, convolutional neural networks are used for image data, recurrent neural networks are used for sequence data, and the defined model architecture after model selection determines the number of layers, number of neurons, and activation function of the model to define the model architecture.

[0031] Define loss function and optimizer for deep learning model training, select the appropriate loss function according to the evaluation task, cross entropy loss is commonly used for classification tasks, mean square error loss is commonly used for regression tasks, select optimizer, and set parameters such as learning rate.

[0032] In the training model and evaluation model of deep learning model training, the training set data is input into the model for iterative training. After each training batch, the error is calculated according to the loss function, and the model parameters are updated through the optimizer. During the training process, the validation set is regularly used to evaluate the performance of the model, and the learning rate, regularization parameters, etc. are adjusted according to the evaluation results.

[0033] Early stopping of deep learning model training is used to stop training when the validation set no longer improves after a certain number of rounds to prevent overfitting. Then, the model is fine-tuned based on the results of the validation set, such as increasing the number of layers and adjusting the number of neurons to achieve model fine-tuning. In conjunction with the accompanying drawings Figure 3 As shown, the appearance image data of the forage during the use of the deep learning model is processed by a convolutional neural network to quickly confirm the spatial characteristics of the forage. The gene sequence uses sequence data to capture the long-term dependency in the sequence. After different processing and calculation of different forage data, the processing effect is increased, thereby realizing the preliminary classification of forage germplasm resources, and the acquired forage data is normalized to form a complete data set. Whenever the training model completes the data processing, the data is merged into the data set again to expand the data set content, to cooperate with the deep learning model training, and to iterate the deep learning model. The loss function definition in the deep learning model training process selects a suitable loss function according to the evaluation task. The classification task often uses the cross entropy loss, and the regression task often uses the mean square error loss. The optimizer is selected and the learning rate and other parameters are set. The optimizer used is stochastic gradient descent, and the learning rate and other parameters are set. The model parameters are adjusted so that the model can accurately predict the quality and value of forage germplasm resources. During the training process, the validation set is regularly used to evaluate the performance of the model, such as accuracy, recall rate, F1 value, etc., and then the learning rate, regularization parameters, etc. are adjusted according to the evaluation results. When the performance of the validation set no longer improves after a certain number of rounds, training is stopped to prevent overfitting and avoid problems with the model data. At the same time, according to the results of the validation set, the number of model layers is increased and the number of neurons is fine-tuned. The trained model is finally evaluated using the test set to determine the performance and generalization ability of the model. Finally, the model with good performance is saved for evaluation of forage germplasm resources in practical applications. When used, forage germplasm data is directly input to evaluate forage germplasm. By integrating multi-source data and deep learning algorithms, the characteristics of forage germplasm resources can be extracted more comprehensively and accurately, thereby improving the accuracy of the evaluation results. Compared with traditional methods, deep learning can quickly process and analyze large amounts of data, greatly shortening the time required for evaluation and improving work efficiency.

[0034] Working principle: When using this deep learning-based forage germplasm resource evaluation method, first, various databases and researchers collect diversified forage germplasm information resources in different geographical areas, and then uniformly process the collected data. After image data feature extraction, the data is further normalized to facilitate the deep learning model to divide and process the data, and to facilitate the learning and adjustment of the convolutional neural network and recurrent neural network within the deep learning model. After the model training is completed, the acquired data is input into the model evaluation, and the obtained model evaluation data is used to evaluate the quality of the forage, its ability to adapt to the environment, its potential economic value, etc., to facilitate the evaluation of the forage germplasm.

[0035] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A forage germplasm resource evaluation method based on deep learning, characterized in that: The following evaluation methods are included: S1. Collect relevant data on forage germplasm resources from multiple channels, including but not limited to data on morphological characteristics and physiological indicators of forage, to ensure the diversity and representativeness of the data, covering forage germplasm resources of different types, geographical origins and growth environments; S2. Clean the data collected in S1 above, remove noise, errors and duplicate data, annotate the data, and then standardize or normalize the data so that data with different features are comparable in value; S3. For the features of grass texture, shape, etc. in the image data of grass morphology in S1 above, a convolutional neural network method is used to automatically extract features, and for the gene sequence data, sequence embedding is used to convert it into a feature vector suitable for deep learning model processing; S4. Divide the preprocessed data in S2 and the data after feature extraction in S3 into a training set, a validation set, and a test set, use the training set data to train the selected deep learning model, and optimize the model performance by adjusting the parameters of the model; S5. Use the test set data in S4 to evaluate the trained model, analyze the evaluation results of the model, and consider adjusting the model structure, increasing the amount of data, or improving the training method according to performance requirements; S6 inputs the data of forage germplasm resources to be evaluated into the optimized model to obtain the evaluation results, analyzes and interprets the evaluation results, and provides a decision-making basis for the protection, utilization and breeding of forage germplasm resources.

2. The forage germplasm resource evaluation method based on deep learning according to claim 1, characterized in that: The sources of the forage germplasm resource data in S1 include: scientific research institutions and universities, and national germplasm resource banks.

3. The forage germplasm resource evaluation method based on deep learning according to claim 2, characterized in that: The forage germplasm resource data includes a summary of the survey, collection and identification of forage germplasm resources in multiple geographical areas.

4. The forage germplasm resource evaluation method based on deep learning according to claim 1, characterized in that: In the step S2 of labeling the forage data, the forage germplasm resources are classified or labeled according to specific evaluation criteria.

5. The forage germplasm resource evaluation method based on deep learning according to claim 1, characterized in that: The deep learning model training work in S4 includes: data preparation, model selection, definition of model architecture, definition of loss function and optimizer, model training and model evaluation, model early stopping and model fine-tuning.

6. The forage germplasm resource evaluation method based on deep learning according to claim 5, characterized in that: The forage germplasm resources in the data preparation for the deep learning model training include images, feature descriptions, gene data, and preprocessed and normalized to form a data set, and the data set is divided into 70% training set, 15% validation set, and 15% test set.

7. The forage germplasm resource evaluation method based on deep learning according to claim 5, characterized in that: In the model selection of the deep learning model training, the convolutional neural network is used for image data, and the recurrent neural network is used for sequence data, and the defined model architecture after the model selection determines the number of layers, the number of neurons, and the activation function to define the model architecture.

8. The forage germplasm resource evaluation method based on deep learning according to claim 5, characterized in that: The deep learning model training defines a loss function and an optimizer, selects a suitable loss function according to the evaluation task, cross entropy loss is commonly used for classification tasks, mean square error loss is commonly used for regression tasks, selects an optimizer, and sets parameters such as the learning rate.

9. The forage germplasm resource evaluation method based on deep learning according to claim 5, characterized in that: In the training model and evaluation model of the deep learning model training, the training set data is input into the model for iterative training. After each training batch, the error is calculated according to the loss function, and the parameters of the model are updated through the optimizer. During the training process, the validation set is regularly used to evaluate the performance of the model, and the learning rate, regularization parameters, etc. are adjusted according to the evaluation results.

10. The forage germplasm resource evaluation method based on deep learning according to claim 5, characterized in that: The early stopping of the model trained by the deep learning model is used to stop training when the validation set no longer improves after a certain number of rounds to prevent overfitting. Subsequently, the model is fine-tuned according to the results of the validation set, such as increasing the number of layers and adjusting the number of neurons to achieve model fine-tuning.

Images