Planting nutrition diagnosis method based on large model

Abstract

The invention provides a planting nutrition diagnosis method based on a large model, and the method comprises the steps: extracting the nutrient metabolism and morphological expression characteristics of a plant based on the growth and development law and environment variables of the plant, and constructing a plant nutrition metabolism model; according to the plant nutrition metabolism model, model parameters are adjusted based on multi-modal data of plant growth, and a multi-dimensional correlation model is constructed; obtaining a morphological picture and an environmental condition in the growth process of a target plant, and identifying the nutrition state of the target plant according to the multi-dimensional correlation model; and generating personalized precise fertilization suggestions according to the nutrition state of the target plant. According to the method, personalized precise fertilization suggestions are generated according to the nutrition state of the target plant, and the suggestions comprise functional fertilizer use suggestions, fertilization mode suggestions, fertilization dosage suggestions and fertilization time suggestions. Scientific guidance can be provided for agricultural production, a user is helped to achieve precise fertilization, crop growth is optimized, and the efficiency of agricultural production is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of plant nutrition management, and in particular relates to a planting nutrition diagnosis method based on a large model. Background Art

[0002] With the growth of global population and the increase of food demand, agricultural production is facing huge challenges. In order to improve the efficiency and sustainability of agricultural production, precision agriculture technology has received extensive attention and application. Precision agriculture uses advanced information technology and data analysis methods to achieve refined management of farmland, crops and environmental conditions, thereby optimizing the agricultural production process and improving crop yield and quality.

[0003] In precision agriculture, plant nutrition management is a key link. The growth, development and yield of plants are directly affected by their nutritional status. Traditional fertilization methods often lack specificity and scientificity, and are prone to cause insufficient or excessive nutrition, affecting the growth and yield of crops. Therefore, it is particularly important to develop a technology that can accurately diagnose the nutritional status of plants and generate personalized precision fertilization recommendations. Summary of the invention

[0004] In order to alleviate the above problems, the present application provides a method for intelligent diagnosis of planting nutrition based on a large model, comprising:

[0005] Based on the growth and development rules of plants and environmental variables, extract the characteristics of plant nutrient metabolism and morphological performance, and construct a plant nutrient metabolism model;

[0006] According to the plant nutrition metabolism model, model parameters are adjusted based on multimodal data of plant growth to construct a multidimensional correlation model;

[0007] Acquire morphological photos and environmental conditions of the target plant during its growth process, and identify the nutritional status of the target plant according to the multidimensional association model;

[0008] Generate personalized precision fertilization recommendations based on the nutritional status of the target plants.

[0009] Optionally, in the process of extracting crop nutrient metabolism and morphological characteristics based on plant growth and development laws and environmental variables, and constructing a plant nutrient metabolism model, deep learning technology is used to establish the nutrient metabolism pathways of the target plant at different growth and development stages, and the plant nutrient metabolism model is trained in combination with the morphological performance of the target plant to identify the plant's nutrient needs.

[0010] Optionally, the process of establishing the nutrient metabolism pathway of the target plant at different growth and development stages by deep learning technology, and training the plant nutrient metabolism model in combination with the morphological performance of the target plant to identify the plant nutrient requirements includes:

[0011] Collecting sample data about the target plant, the sample data including morphological photos at different growth stages, environmental data, and corresponding nutritional status data;

[0012] Use convolutional neural networks to extract plant morphological features from morphological photos and extract relevant features from environmental data and nutritional status data;

[0013] Convolutional layers, pooling layers, and fully connected layers are set up to integrate features extracted from morphological photos and environmental data for model training to identify the nutritional needs of plants;

[0014] Apply back-propagation algorithms and optimizers (such as Adam, SGD, etc.) to adjust the weights of the model to minimize the recognition error.

[0015] Optionally, according to the plant nutritional metabolism model, the model parameters are adjusted based on the multimodal data of plant growth. In the process of constructing the multidimensional correlation model, based on the multimodal data, plant metabolic characteristics, planting environment characteristics and fertilization behavior characteristics are extracted respectively, and the correlation between crop metabolic pathways and environmental variables is explored, and the model parameters are dynamically optimized through a reinforcement learning algorithm.

[0016] Optionally, based on the multimodal data, extracting plant metabolic characteristics, planting environment characteristics and fertilization behavior characteristics respectively, exploring the correlation between crop metabolic pathways and environmental variables, and dynamically optimizing model parameters through a reinforcement learning algorithm includes:

[0017] Extract plant metabolic features from plant metabolic data using principal component analysis (PCA) or self-organizing map (SOM), use sensor data as environmental features or screen planting environment features through feature selection methods, and screen data features of fertilization frequency and dosage changes as fertilization behavior features;

[0018] Combining decision tree algorithms (such as random forests) to process categorical features, neural network algorithms to process continuous features, and clustering algorithms to discover hidden patterns in data, building composite processing models to identify relationships between environmental variables and crop metabolic pathways;

[0019] Use cross-validation and other techniques to evaluate the generalization ability of the model, apply Q learning, deep Q network (DQN) or policy gradient method to strengthen the learning algorithm, and use time difference learning (TD Learning) method to update the multi-dimensional correlation model parameters to maximize long-term rewards and achieve dynamic optimization of model parameters.

[0020] Optionally, the process of updating the multidimensional association model parameters using the temporal difference learning method to maximize the long-term reward includes:

[0021] Initialize the weights of the neural network and the splitting rules of the decision tree, and use the initial data to train the multidimensional association model;

[0022] At each time step, the multidimensional association model takes an action based on the current state, then observes the next state and reward, and calculates the temporal difference error;

[0023] According to the time difference error, the gradient descent algorithm is used to update the parameters of the multidimensional correlation model and perform iterative training.

[0024] Optionally, morphological photographs and environmental conditions during the growth process of the target plant are obtained, and in the process of identifying the nutritional status of the target plant according to the multidimensional association model, large model reasoning is used to extract the morphological features of the morphological photographs based on the morphological photographs and input them into the multidimensional association model. The nutritional status of the target plant at the growth stage is identified by combining the morphological features and environmental conditions, and a plant nutritional status analysis result is generated.

[0025] Optionally, the process of using large model reasoning, based on the morphological photograph, extracting morphological features of the morphological photograph and inputting them into the multidimensional association model, combining morphological features and environmental conditions, identifying the nutritional status of the target plant at the growth stage, and generating a plant nutritional status analysis result includes:

[0026] Collect morphological photos of target plants at different growth stages, and pre-process the collected morphological photos to improve image quality and meet the input requirements of the multidimensional association model;

[0027] Using a convolutional neural network to automatically extract plant morphological features from morphological photos, and inputting the morphological features and environmental data into a trained multidimensional association model;

[0028] The output of the model is analyzed to generate plant nutritional status analysis results.

[0029] Optionally, in the process of generating personalized precise fertilization recommendations based on the nutritional status of the target plant, personalized precise fertilization recommendations are generated based on the nutritional status analysis results and the output of the multidimensional association model, combined with the target yield, regional historical planting data and the current status of the plant, based on the large model reasoning capability, and the personalized precise fertilization recommendations include functional fertilizer recommendations, fertilization method recommendations, fertilization amount recommendations and fertilization time recommendations.

[0030] Optionally, the process of generating personalized precise fertilization suggestions based on the large model reasoning capability according to the nutritional status analysis results and the output of the multidimensional association model, combined with the historical planting data of the region and the current status of the plant, wherein the personalized precise fertilization suggestions include functional fertilizer suggestions, fertilization method suggestions, fertilization amount suggestions and fertilization time suggestions, includes:

[0031] Based on the results of nutrient status analysis, the output of the multidimensional association model, the historical planting data of the region and the current status information of the target plants, feature engineering is performed on the integrated data to extract key features related to fertilization decision-making, including soil nutrient levels, crop growth stages, historical yield data and environmental conditions;

[0032] According to the key features of the fertilization decision, a corresponding fertilization strategy is selected to generate the personalized and precise fertilization recommendation.

[0033] The large-model-based intelligent diagnosis method for planting nutrition in this application is based on the growth and development laws of plants and environmental variables, extracts plant nutrient metabolism and morphological performance characteristics, and constructs a plant nutrient metabolism model; according to the plant nutrient metabolism model, adjusts model parameters around the multimodal data of plant growth, and constructs a multidimensional correlation model; obtains morphological photos and environmental conditions during the growth of the target plant, and identifies the nutritional status of the target plant according to the multidimensional correlation model; generates personalized and precise fertilization suggestions according to the nutritional status of the target plant. This application uses deep learning technology to establish the nutrient metabolism pathway of the target plant at different growth and development stages, and trains the plant nutrient metabolism model in combination with the morphological performance of the plant to identify the plant's nutritional needs. At the same time, the system uses multimodal data to extract plant metabolic characteristics, planting environment characteristics, and fertilization behavior characteristics, explores the correlation between crop metabolic pathways and environmental variables, and dynamically optimizes model parameters through reinforcement learning algorithms. In addition, the system can also generate personalized and precise fertilization suggestions based on the nutritional status of the target plant, including functional fertilizer suggestions, fertilization method suggestions, fertilization dosage suggestions, and fertilization time suggestions. It can provide scientific guidance for agricultural production, help users achieve precise fertilization, optimize crop growth and improve the efficiency of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This application provides a flow chart of a method for intelligent diagnosis of plant nutrition based on a large model. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0036] First embodiment

[0037] This application first provides a method for intelligent diagnosis of planting nutrition based on a large model. Figure 1 This application provides a flow chart of a method for intelligent diagnosis of plant nutrition based on a large model.

[0038] like Figure 1 As shown, in one embodiment, a method for intelligent diagnosis of planting nutrition based on a large model includes:

[0039] S10: Based on the growth and development rules of plants and environmental variables, extract the characteristics of plant nutrient metabolism and morphological performance, and construct a plant nutrient metabolism model;

[0040] S20: According to the plant nutrition metabolism model, adjusting model parameters based on multimodal data of plant growth, and constructing a multidimensional correlation model;

[0041] S30: Acquire morphological photos and environmental conditions of the target plant during its growth process, and identify the nutritional status of the target plant according to the multidimensional association model;

[0042] S40: Generate personalized precise fertilization suggestions based on the nutritional status of the target plant.

[0043] In recent years, with the rapid development of artificial intelligence technologies such as machine learning and deep learning, new possibilities have been provided for building intelligent plant nutrition diagnosis systems. By combining the growth and development laws, environmental variables and morphological performance of plants, a plant nutrition metabolism model can be constructed to identify plant nutrition needs. At the same time, using multimodal data acquisition and processing technology, plant metabolic characteristics, planting environment characteristics and fertilization behavior characteristics can be extracted to explore the correlation between crop metabolic pathways and environmental variables.

[0044] In this context, this embodiment proposes a method for intelligent diagnosis of planting nutrition based on a large model. This method uses deep learning technology to establish the nutrient metabolism pathways of target plants at different growth and development stages, and combines the morphological performance of plants to train plant nutrient metabolism models to identify plant nutritional needs. At the same time, the system uses multimodal data to extract plant metabolic characteristics, planting environment characteristics, and fertilization behavior characteristics, explores the correlation between crop metabolic pathways and environmental variables, and dynamically optimizes model parameters through reinforcement learning algorithms. In addition, personalized and precise fertilization recommendations can be generated based on the nutritional status of the target plant, including functional fertilizer recommendations, fertilization method recommendations, fertilization amount recommendations, and fertilization time recommendations.

[0045] Optionally, in the process of extracting crop nutrient metabolism and morphological characteristics based on plant growth and development laws and environmental variables, and constructing a plant nutrient metabolism model, deep learning technology is used to establish the nutrient metabolism pathway of the target plant at different growth and development stages, and the plant nutrient metabolism model is trained in combination with the morphological performance of the target plant to identify plant nutrient requirements.

[0046] Illustratively, a plant nutrient metabolism and morphological expression model is constructed based on the growth and development laws of plants and environmental variables (light, heat, water, soil, and temperature); the model uses deep learning technology to establish the nutrient metabolism pathways of plants at different growth and development stages, combined with plant morphological expressions (including branches, leaves, flowers, and fruits), thereby achieving the purpose of identifying plant nutritional needs.

[0047] Optionally, the process of establishing the nutrient metabolism pathway of the target plant at different growth and development stages by deep learning technology, and training the plant nutrient metabolism model in combination with the morphological performance of the target plant to identify the plant nutrient requirements includes:

[0048] Collecting sample data about the target plant, the sample data including morphological photos at different growth stages, environmental data, and corresponding nutritional status data;

[0049] Use convolutional neural networks to extract plant morphological features from morphological photos and extract relevant features from environmental data and nutritional status data;

[0050] Convolutional layers, pooling layers, and fully connected layers are set up to integrate features extracted from morphological photos and environmental data for model training to identify the nutritional needs of plants;

[0051] Apply back-propagation algorithms and optimizers (such as Adam, SGD, etc.) to adjust the weights of the model to minimize the recognition error.

[0052] For example, a large amount of sample data about the target plant can be collected, including morphological photos at different growth stages, environmental data (such as temperature, humidity, light intensity, soil composition, etc.), and corresponding nutritional status data. The image data is preprocessed, such as resizing, normalization, enhancement, etc., to meet the input requirements of the deep learning model. Then a deep learning model such as a convolutional neural network (CNN) is used to extract the morphological characteristics of the plant from the morphological photos, such as leaf size, color, shape, etc. Relevant features are extracted from the environmental data and nutritional status data, such as soil nutrient content, climatic conditions, etc. A deep learning model architecture is designed that can integrate the features extracted from the morphological photos and environmental data and identify the nutritional needs of the plant.

[0053] For example, the model may include multiple layers, such as convolutional layers, pooling layers, fully connected layers, etc., as well as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) for processing sequence data. The model is trained using the collected dataset. During the training process, the model learns how to identify the nutritional requirements of plants from the input features. Backpropagation algorithms and optimizers (such as Adam, SGD, etc.) are applied to adjust the weights of the model to minimize the recognition error.

[0054] For example, the validation set can be used to evaluate the performance of the model, such as accuracy, recall, F1 score and other indicators. The model is adjusted and optimized according to the evaluation results, such as adjusting the model architecture, hyperparameter tuning, etc. Through these steps, an accurate plant nutrition metabolism model is constructed, which can identify the nutritional needs of plants according to their morphological performance and environmental conditions, thereby providing scientific guidance for agricultural production.

[0055] Optionally, according to the plant nutritional metabolism model, the model parameters are adjusted based on the multimodal data of plant growth. In the process of constructing the multidimensional correlation model, based on the multimodal data, plant metabolic characteristics, planting environment characteristics and fertilization behavior characteristics are extracted respectively, and the correlation between crop metabolic pathways and environmental variables is explored, and the model parameters are dynamically optimized through a reinforcement learning algorithm.

[0056] Exemplarily, based on multimodal data, a multidimensional correlation model between "plant metabolism-planting environment-fertilization behavior" is constructed; the correlation model automatically mines the implicit correlation between plant metabolic pathways and environmental variables through the general large model reasoning capability, and dynamically optimizes the model parameters through the reinforcement learning algorithm.

[0057] Optionally, based on the multimodal data, extracting plant metabolic characteristics, planting environment characteristics and fertilization behavior characteristics respectively, exploring the correlation between crop metabolic pathways and environmental variables, and dynamically optimizing model parameters through a reinforcement learning algorithm includes:

[0058] Extract plant metabolic features from plant metabolic data using principal component analysis (PCA) or self-organizing map (SOM), use sensor data as environmental features or screen planting environment features through feature selection methods, and screen data features of fertilization frequency and dosage changes as fertilization behavior features;

[0059] Combining decision tree algorithms (such as random forests) to process categorical features, neural network algorithms to process continuous features, and clustering algorithms to discover hidden patterns in the data, building composite processing models to make predictions based on the relationship between environmental variables and crop metabolic pathways;

[0060] Use cross-validation and other techniques to evaluate the generalization ability of the model, apply Q learning, deep Q network (DQN) or policy gradient method to strengthen the learning algorithm, and use time difference learning (TD Learning) method to update the multi-dimensional correlation model parameters to maximize long-term rewards and achieve dynamic optimization of model parameters.

[0061] For example, multimodal data can be collected, including plant metabolic data (such as nutrient content, growth rate, etc.), planting environment data (such as temperature, humidity, light intensity, soil composition, etc.), and fertilization behavior data (such as fertilization type, dosage, time, etc.). Different types of data are integrated and synchronized to ensure the time series consistency of the data. Metabolic features are then extracted from the plant metabolic data using appropriate methods, such as using principal component analysis (PCA) or self-organizing map (SOM). Environmental features are extracted from the planting environment data, such as using sensor data directly as features or screening important features through feature selection methods. Fertilization features are extracted from fertilization behavior data, such as fertilization frequency, dosage changes, etc. A model is constructed using machine learning or deep learning techniques to explore the association between crop metabolic pathways and environmental variables. This can be a composite model to combine multiple algorithms to process different types of data. Reinforcement learning algorithms, such as Q learning, deep Q network (DQN), or policy gradient methods, are applied to dynamically optimize model parameters. A reward mechanism can also be designed to give positive rewards when the fertilization recommendations identified by the model lead to improved plant growth; otherwise, negative rewards are given.

[0062] For example, the model is trained in a simulated environment or in an actual planting environment. The model learns the best fertilization scheme by continuously trying different fertilization strategies and observing the results. Methods such as temporal difference learning (TD Learning) are used to update model parameters to maximize long-term rewards.

[0063] The reinforcement learning algorithm of this embodiment can help dynamically optimize model parameters to adapt to changing environmental conditions and crop growth requirements, thereby achieving more accurate plant nutrition management and fertilization recommendations.

[0064] Optionally, the process of updating the multidimensional association model parameters using the temporal difference learning method to maximize the long-term reward includes:

[0065] Initialize the weights of the neural network and the splitting rules of the decision tree, and use the initial data to train the multidimensional association model;

[0066] At each time step, the multidimensional association model takes an action based on the current state, then observes the next state and reward, and calculates the temporal difference error;

[0067] According to the time difference error, the gradient descent algorithm is used to update the parameters of the multidimensional correlation model and perform iterative training.

[0068] Temporal Difference Learning (TD Learning) is a reinforcement learning algorithm used to update model parameters to maximize long-term rewards. Exemplarily, before starting the learning process, the parameters of the model need to be initialized. These parameters can be weights of a neural network, splitting rules of a decision tree, etc., depending on the type of model used. Collect some initial data, which can be the state of the environment, the actions taken, and the corresponding rewards. This data is used to train the initial model.

[0069] During the model training process, the model is trained using the initial data. The goal of the model is to learn how to take the best action based on the current state of the environment to obtain the maximum reward. The core idea of ​​TD Learning is to use temporal difference (TemporalDifference) to update the model parameters. Temporal difference refers to the difference between the current state and the next state. At each time step, the model takes an action based on the current state and then observes the next state and reward. The model uses the temporal difference error (TD Error) to update the parameters. The temporal difference error is defined as the difference between the predicted value of the current state and the predicted value of the next state plus the actual reward.

[0070] Exemplarily, based on the temporal difference error, an optimization algorithm (such as gradient descent) is used to update the model parameters. The update goal is to maximize the long-term reward, that is, the total reward obtained in future time steps. Repeat the above steps to improve the performance of the model by continuously updating the model parameters. In each iteration, the model uses new data to update the parameters to adapt to the changing environment. Regularly evaluate the performance of the model to ensure that it can accurately identify the nutritional needs of crops under different environmental conditions. Adjust and optimize the model based on the evaluation results to improve its performance.

[0071] In this embodiment, by using methods such as time difference learning to update model parameters, accurate identification of plant nutritional needs can be achieved and scientific guidance can be provided for agricultural production.

[0072] Optionally, morphological photographs and environmental conditions during the growth process of the target plant are obtained, and in the process of identifying the nutritional status of the target plant according to the multidimensional association model, large model reasoning is used to extract morphological features of the morphological photographs based on the morphological photographs and input them into the multidimensional association model. The nutritional status of the target plant at the growth stage is automatically identified in combination with the morphological features and environmental conditions, and it is determined whether the plant has nutritional deficiency, nutritional excess or nutritional imbalance, and the plant nutritional status analysis results are generated.

[0073] By using large-scale model reasoning and based on the input of morphological photos and multi-dimensional association models, the nutritional status of plants at growth stages such as buds, shoots, flowers, and fruits can be automatically identified. By combining morphological characteristics and environmental conditions, it is determined whether the plants have nutritional deficiencies, excesses, or other nutritional imbalances, and a plant nutritional status analysis report can be generated.

[0074] Optionally, the process of using large model reasoning, based on the morphological photos, extracting morphological features of the morphological photos and inputting them into the multidimensional association model, combining morphological features and environmental conditions, automatically identifying the nutritional status of the target plant at the growth stage, judging whether the plant has nutritional deficiency, nutritional excess or nutritional imbalance, and generating plant nutritional status analysis results includes:

[0075] Collect morphological photos of target plants at different growth stages, and pre-process the collected morphological photos to improve image quality and meet the input requirements of the multidimensional association model;

[0076] Using a convolutional neural network to automatically extract plant morphological features from morphological photos, and inputting the morphological features and environmental data into a trained multidimensional association model;

[0077] The output of the model is analyzed to generate plant nutritional status analysis results.

[0078] For example, morphological photos of target plants at different growth stages are collected, and these photos should cover various environmental conditions and nutritional status. At the same time, the environmental condition data corresponding to each photo is collected, such as temperature, humidity, light intensity, soil composition, etc. The collected morphological photos are preprocessed, including resizing, noise removal, contrast enhancement, etc., to improve image quality and meet the input requirements of subsequent models.

[0079] For example, a deep learning model (such as a convolutional neural network (CNN)) is used to automatically extract plant morphological features from morphological photos, such as leaf size, color, shape, texture, etc. These features will serve as one of the inputs of the multidimensional association model.

[0080] Exemplarily, the collected environmental condition data can be cleaned and standardized to ensure the consistency and quality of the data. A multidimensional association model is constructed that can identify the nutritional status of plants by combining plant morphological characteristics and environmental condition data. This model can be an algorithm based on machine learning or deep learning, such as a random forest, a support vector machine, or a neural network. The multidimensional association model can be trained using labeled data sets. These labels can be given by experts based on the actual nutritional status of the plant, such as nutritional deficiency, nutritional excess, or nutritional balance. During the training process, the model can learn how to extract information from morphological characteristics and environmental conditions to accurately identify the nutritional status of the plant.

[0081] For new morphological photos and environmental condition data, the trained multidimensional association model is used for inference. The model automatically identifies the growth stage and nutritional status of the target plant based on the input morphological characteristics and environmental conditions. According to the output of the model, it is analyzed whether the plant has nutritional deficiency, nutritional excess or nutritional imbalance. Finally, the plant nutritional status analysis results are generated, including specific nutritional problems and their severity. This can realize the automatic identification and analysis of the nutritional status of the target plant during its growth process, thus providing valuable information for precision agriculture.

[0082] Optionally, in the process of generating personalized precise fertilization recommendations based on the nutritional status of the target plant, personalized precise fertilization recommendations are generated based on the nutritional status analysis results and the output of the multidimensional association model, combined with the target yield, regional historical planting data and the current status of the plant, based on the large model reasoning capability, and the personalized precise fertilization recommendations include functional fertilizer recommendations, fertilization method recommendations, fertilization amount recommendations and fertilization time recommendations.

[0083] For example, based on the output of the nutritional status analysis report and the multidimensional association model, combined with the target yield, regional historical planting data and the current status of the plants, personalized and precise fertilization recommendations are generated based on the reasoning capability of the large model, including: functional fertilizer recommendations (to promote shoot growth, flowering, and fruit growth, etc.), fertilization methods (foliar fertilizer and ground fertilizer), fertilizer amount (including number of times) and fertilization time.

[0084] Optionally, the process of generating personalized precise fertilization suggestions based on the large model reasoning capability according to the nutritional status analysis results and the output of the multidimensional association model, combined with the target yield, regional historical planting data and the current state of the plant, wherein the personalized precise fertilization suggestions include functional fertilizer suggestions, fertilization method suggestions, fertilization amount suggestions and fertilization time suggestions, includes:

[0085] Integrate the results of nutrient status analysis, the output of the multidimensional correlation model, target yield requirements, regional historical planting data and current status information of target plants, perform feature engineering on the integrated data, and extract key features for fertilization decision-making, including soil nutrient levels, crop growth stages, historical yield data and environmental conditions;

[0086] According to the key features of the fertilization decision, a corresponding fertilization strategy is selected to generate the personalized and precise fertilization recommendation.

[0087] For example, the results of nutrient status analysis, the output of multidimensional association models, target yield requirements, regional historical planting data, and plant current status information are integrated. These data provide comprehensive background information for generating personalized fertilization recommendations. Feature engineering can be performed on the integrated data to extract key features related to fertilization decisions, such as soil nutrient levels, crop growth stages, historical yield data, environmental conditions, etc.

[0088] For example, a large-scale model-based fertilization recommendation model can be built, which can learn how to generate personalized fertilization recommendations based on the nutritional status and related characteristics of the plant. This model can be based on machine learning or deep learning algorithms, such as random forests, neural networks, or reinforcement learning models. The fertilization recommendation model is then trained using historical data. This data should include past fertilization strategies, corresponding plant growth responses, and yield results. During the training process, the model learns how to identify the best fertilization plan based on the input features.

[0089] For example, for new plant nutrition status data, a trained model is used to generate personalized fertilization recommendations. The model outputs fertilization recommendations based on input features such as soil nutrient levels, crop growth stage, historical yield data, environmental conditions, etc.

[0090] For example, the fertilization recommendations output by the model are detailed into specific fertilization strategies, including:

[0091] Functional fertilizer recommendations: Determine what type of fertilizer to use (such as nitrogen fertilizer, phosphorus fertilizer, potassium fertilizer, etc.).

[0092] Fertilization method recommendations: Determine the method of fertilization (such as foliar spray, soil application, etc.).

[0093] Fertilizer dosage recommendations: Determine the specific amount of fertilizer to be applied, which may include how much fertilizer to apply each time and how often to apply it.

[0094] Fertilizer Timing Recommendations: Determine the best time to apply fertilizer, which may be based on the crop's growth stage and environmental conditions.

[0095] For example, an independent test set can be used to verify the accuracy of fertilization recommendations to ensure that they can effectively increase crop yields in practical applications. The model can then be adjusted and optimized based on the verification results to improve the accuracy and practicality of the fertilization recommendations. Furthermore, the generated fertilization recommendations can be provided to users (such as farmers or agricultural experts) and their feedback can be collected. Based on user feedback and actual application results, the model and recommendations are iteratively improved. In this embodiment, personalized and precise fertilization recommendations based on large model reasoning capabilities can be generated to optimize plant growth and improve the efficiency of agricultural production.

[0096] Second embodiment

[0097] On the basis of the above embodiments, the present application also provides a planting nutrition intelligent identification and diagnosis system based on a universal large model.

[0098] The system builds a "plant nutrition intelligent diagnosis and analysis model" based on the "plant metabolic nutrition law and the mechanism of the environment (light, heat, water, soil, temperature, etc.)", including:

[0099] Under the influence of different environments (light, heat, water, soil, etc.), a "plant nutrient metabolism and morphological performance (shoots, leaves, flowers, fruits, etc.)" model is constructed to identify the nutritional needs of plants at different growth and development stages;

[0100] Obtain morphological photos of plants at different stages, fertilization data, and corresponding environmental data (temperature, soil moisture, rainfall, etc.), and establish a multi-dimensional correlation model between "plant metabolism, planting environment, and fertilization";

[0101] Based on the general large model reasoning ability and data, combined with the association model, it can automatically identify and judge the nutritional status of "plant buds, shoots, flowers, and fruits" at different stages, improve the recognition rate of nutritional status, and lay the foundation for precise fertilization.

[0102] Exemplarily, the planting nutrition intelligent identification and diagnosis system based on the general large model may include:

[0103] Plant nutrition metabolism modeling module: Based on the growth and development rules of plants and environmental variables (light, heat, water, soil, temperature), a plant nutrient metabolism and morphological performance model is constructed; the model uses deep learning technology to establish the nutrient metabolism pathways of plants at different growth and development stages, and combines plant morphological performance (including branches, leaves, flowers, and fruits) to identify plant nutritional needs;

[0104] Multimodal data acquisition and processing module, including:

[0105] a) an image acquisition unit, used to obtain high-resolution morphological photos of plants at various growth stages;

[0106] b) Environmental data collection unit, used to collect key data in the planting environment, including temperature, light intensity, soil moisture, rainfall and humidity;

[0107] c) Fertilization data collection unit, used to record the type, amount and time of fertilization;

[0108] Multidimensional association model building module: Based on multimodal data, a multidimensional association model between "plant metabolism-planting environment-fertilization behavior" is constructed; the association model automatically mines the implicit association between plant metabolic pathways and environmental variables through the general large model reasoning ability, and dynamically optimizes model parameters through reinforcement learning algorithms;

[0109] Intelligent nutritional status diagnosis module: Utilizes large-scale model reasoning, based on the input of morphological photos and multi-dimensional association models, automatically identifies the nutritional status of plants at growth stages such as buds, shoots, flowers, and fruits; combines morphological characteristics and environmental conditions to determine whether plants have nutritional deficiencies, excesses, or other nutritional imbalances, and generates a plant nutritional status analysis report.

[0110] Precision fertilization recommendation generation module: Based on the output of the nutritional status analysis report and the multidimensional correlation model, combined with the target yield, regional historical planting data and the current status of the plants, it uses large-scale model reasoning capabilities to generate personalized precision fertilization recommendations, including: functional fertilizer recommendations (to promote shoot growth, flowering, and fruit growth, etc.), fertilization methods (foliar fertilizer and ground fertilizer), fertilizer amount (including the number of fertilization times) and fertilization time.

[0111] The present invention discloses a method and system for intelligent diagnosis of plant nutrition based on a large model. Based on the growth and development rules of plants and environmental variables, the plant nutrient metabolism and morphological performance characteristics are extracted to construct a plant nutrient metabolism model; according to the plant nutrient metabolism model, the model parameters are adjusted based on the multimodal data of plant growth to construct a multidimensional correlation model; the morphological photos and environmental conditions of the target plant during its growth are obtained, and the nutritional status of the target plant is identified according to the multidimensional correlation model; and personalized and precise fertilization suggestions are generated according to the nutritional status of the target plant. The present invention establishes the nutrient metabolism pathway of the target plant at different growth and development stages through deep learning technology, and trains the plant nutrient metabolism model in combination with the morphological performance of the plant to identify the plant's nutritional needs. At the same time, the system uses multimodal data to extract plant metabolic characteristics, planting environment characteristics, and fertilization behavior characteristics, explores the correlation between crop metabolic pathways and environmental variables, and dynamically optimizes model parameters through reinforcement learning algorithms. In addition, the system can also generate personalized and precise fertilization suggestions based on the nutritional status of the target plant, including functional fertilizer suggestions, fertilization method suggestions, fertilization dosage suggestions, and fertilization time suggestions. It can provide scientific guidance for agricultural production, help users achieve precise fertilization, optimize crop growth and improve the efficiency of agricultural production.

[0112] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A method for diagnosing plant nutrition based on a large model, characterized in that: include: Based on the growth and development rules of plants and environmental variables, extract the characteristics of plant nutrient metabolism and morphological performance, and construct a plant nutrient metabolism model; According to the plant nutrition metabolism model, model parameters are adjusted based on multimodal data of plant growth to construct a multidimensional correlation model; Acquire morphological photos and environmental conditions of the target plant during its growth process, and identify the nutritional status of the target plant according to the multidimensional association model; Generate personalized precision fertilization recommendations based on the nutritional status of the target plants.

2. A method for diagnosing plant nutrition based on a large model according to claim 1, characterized in that: In the process of extracting crop nutrient metabolism and morphological characteristics based on plant growth and development laws and environmental variables and constructing a plant nutrient metabolism model, the nutrient metabolism pathways of the target plant at different growth and development stages are established through deep learning technology, and the plant nutrient metabolism model is trained to identify plant nutrient requirements in combination with the morphological performance of the target plant.

3. A method for diagnosing plant nutrition based on a large model according to claim 2, characterized in that: The process of establishing the nutrient metabolism pathway of the target plant at different growth and development stages by deep learning technology, and training the plant nutrient metabolism model in combination with the morphological performance of the target plant to identify the plant nutrient requirements includes: Collecting sample data about the target plant, the sample data including morphological photos at different growth stages, environmental data, and corresponding nutritional status data; Use convolutional neural networks to extract plant morphological features from morphological photos and extract relevant features from environmental data and nutritional status data; Convolutional layers, pooling layers, and fully connected layers are set up to integrate features extracted from morphological photos and environmental data for model training to identify the nutritional needs of plants; A back-propagation algorithm and an optimizer are applied to adjust the weights of the model to minimize the recognition error.

4. A method for diagnosing plant nutrition based on a large model according to claim 3, characterized in that: According to the plant nutritional metabolism model, the model parameters are adjusted based on the multimodal data of plant growth. In the process of constructing the multidimensional correlation model, the plant metabolic characteristics, planting environment characteristics and fertilization behavior characteristics are extracted based on the multimodal data, the correlation between plant growth and development data and environmental variables is explored, and the model parameters are dynamically optimized through the reinforcement learning algorithm.

5. A method for diagnosing plant nutrition based on a large model according to claim 4, characterized in that: Based on the multimodal data, plant growth and development characteristics, planting environment characteristics and fertilization behavior characteristics are extracted respectively, and the correlation between crop metabolic pathways and environmental variables is explored. The process of dynamically optimizing model parameters through a reinforcement learning algorithm includes: Extract plant metabolic characteristics from plant growth and development data using principal component analysis or self-organizing mapping, use sensor data as environmental characteristics or screen planting environment characteristics through feature selection methods, and screen data characteristics of fertilization frequency and dosage changes as fertilization behavior characteristics; Combining decision tree algorithms to process categorical features, neural network algorithms to process continuous features, and clustering algorithms to discover hidden patterns in data, a composite processing model was constructed to identify the relationship between environmental variables and crop metabolic pathways; Use techniques such as cross-validation to evaluate the generalization ability of the model, apply Q-learning, deep Q-network or policy gradient method to strengthen the learning algorithm, and use time difference learning method to update the parameters of multi-dimensional correlation model to maximize long-term rewards and realize dynamic optimization of model parameters.

6. A method for diagnosing plant nutrition based on a large model according to claim 5, characterized in that: The process of updating the multi-dimensional association model parameters using the temporal difference learning method to maximize the long-term reward includes: Initialize the weights of the neural network and the splitting rules of the decision tree, and use the initial data to train the multidimensional association model; At each time step, the multidimensional association model takes an action based on the current state, then observes the next state and reward, and calculates the temporal difference error; According to the time difference error, the gradient descent algorithm is used to update the parameters of the multidimensional correlation model and perform iterative training.

7. A method for diagnosing plant nutrition based on a large model according to claim 6, characterized in that: The morphological photographs and environmental conditions during the growth process of the target plant are obtained. In the process of identifying the nutritional status of the target plant according to the multidimensional association model, large model reasoning is used to extract the morphological features of the morphological photographs based on the morphological photographs and input them into the multidimensional association model. The nutritional status of the target plant at the growth stage is identified by combining the morphological features and environmental conditions, and a plant nutritional status analysis result is generated.

8. A method for diagnosing plant nutrition based on a large model according to claim 7, characterized in that: The process of using large model reasoning, extracting morphological features of the morphological photos based on the morphological photos and inputting them into the multidimensional association model, combining morphological features and environmental conditions, identifying the nutritional status of the target plant at the growth stage, and generating plant nutritional status analysis results includes: Collect morphological photos of target plants at different growth stages, and pre-process the collected morphological photos to improve image quality and meet the input requirements of the multidimensional association model; Using a convolutional neural network to automatically extract plant morphological features from morphological photos, and inputting the morphological features and environmental data into a trained multidimensional association model; The output of the model is analyzed to generate plant nutritional status analysis results.

9. A method for diagnosing plant nutrition based on a large model according to any one of claims 1 to 8, characterized in that: In the process of generating personalized and precise fertilization suggestions according to the nutritional status of the target plant, personalized and precise fertilization suggestions are generated based on the nutritional status analysis results and the output of the multidimensional association model, combined with the target yield, regional historical planting data and the current status of the plant, based on the large model reasoning capability. The personalized and precise fertilization suggestions include functional fertilizer recommendations, fertilization method recommendations, fertilization amount recommendations and fertilization time recommendations.

10. A method for diagnosing plant nutrition based on a large model according to claim 9, characterized in that: The process of generating personalized precise fertilization suggestions based on the large model reasoning capability according to the nutritional status analysis results and the output of the multidimensional association model, combined with the target yield, regional historical planting data and the current state of the plant, wherein the personalized precise fertilization suggestions include functional fertilizer suggestions, fertilization method suggestions, fertilization amount suggestions and fertilization time suggestions includes: Integrate the results of nutrient status analysis, the output of the multidimensional correlation model, target yield requirements, regional historical planting data and current status information of target plants, perform feature engineering on the integrated data, and extract key features for fertilization decision-making, including soil nutrient levels, crop growth stages, historical yield data and environmental conditions; According to the key features of the fertilization decision, a corresponding fertilization strategy is selected to generate the personalized and precise fertilization recommendation.

Images