Wheat disease efficient prediction method and system based on deep learning

Abstract

The invention discloses a wheat disease efficient prediction method and system based on deep learning, and the method combines a bidirectional long-short term memory network (BiLSTM) and a Transform attention mechanism, and comprises the steps: collecting meteorological and historical disease data, and carrying out the data preprocessing and feature extraction; constructing a wheat disease knowledge graph; the BiLSTM is used for processing time sequence data, and the time sequence data is output as the input of Transform, so that the attention on key features is enhanced; and finally, training the model, predicting new data, and providing decision suggestions. The method has the advantages that information related to disease occurrence is comprehensively captured, and the prediction accuracy is improved; key information is screened through an attention mechanism, and the training efficiency is improved; real-time data prediction is supported, and the intelligent level of agricultural management is enhanced. The method has good generalization ability, and a scientific and effective disease management tool is provided for agricultural producers.

Description

Technical Field

[0001] The present invention relates to the technical field of wheat disease prediction, and in particular to a wheat disease prediction method based on knowledge graph, Transformer attention mechanism and bidirectional long short-term memory network. Background Art

[0002] With the modernization of agriculture and the advancement of science and technology, wheat disease prediction has become an important research field to increase crop yields and ensure food security. Existing disease prediction technologies are mainly divided into the following categories, each of which has its advantages and disadvantages.

[0003] First, human subjective predictions or statistical-based methods rely on expert experience and are highly subjective, which may lead to large prediction errors and make it difficult to adapt to complex environmental changes.

[0004] Secondly, prediction methods based on general neural networks. Although these methods can handle certain nonlinear relationships, they perform poorly in capturing the causal relationship between wheat diseases and environmental factors, resulting in limited prediction accuracy.

[0005] Third, detection technologies based on molecular biology. Although these methods, such as PCR, can detect pathogens, they are complex to operate and costly, lack real-time performance, and cannot accurately assess the extent of disease occurrence.

[0006] Fourth, the prediction method based on LSTM (Long Short-Term Memory Network). LSTM can capture the long-term dependencies of time series data, but its unidirectionality leads to insufficient information utilization, relatively limited expression ability, and inability to fully integrate the relevant factors of the previous and next time periods. It is difficult to fully learn and generalize complex sequence patterns. In addition, LSTM usually uses relatively few regularization methods, such as L2 regularization and Dropout. Although these methods can reduce overfitting to a certain extent, the effect is limited and overfitting is still prone to occur. Summary of the invention

[0007] In view of the defects of the prior art, the present invention provides an efficient prediction method for wheat diseases based on deep learning. This method can not only consider historical and future information at the same time, but also effectively filter out key information through the attention mechanism, improve the accuracy and efficiency of prediction, enhance the performance and generalization ability of the model, and provide a more scientific and practical solution for the prediction of wheat diseases.

[0008] In order to achieve the above invention object, the technical solution adopted by the present invention is as follows:

[0009] An efficient prediction method for wheat diseases based on deep learning includes the following steps:

[0010] S1: Data collection, collect meteorological data and historical disease data related to wheat diseases, and collect meteorological data in real time by deploying a sensor network in the farmland.

[0011] S2: Data preprocessing: cleaning, normalizing and feature engineering the collected data to extract features related to wheat diseases and label wheat disease images.

[0012] S3: Knowledge graph construction: Based on the knowledge of traditional agricultural experts and literature, a knowledge graph of wheat diseases is established to form a comprehensive disease information database.

[0013] S4: Model construction, including:

[0014] S4.1: BiLSTM layer, used to process time series data and learn the characteristics and trends of wheat growth;

[0015] S4.2: Transformer attention mechanism layer, based on the output of the BiLSTM layer, calculates the attention weight of each time step to highlight the information that contributes more to disease prediction.

[0016] S5: Model training: Use the preprocessed data to train the constructed model and use the loss function and optimization algorithm to adjust the model parameters.

[0017] S6: Prediction and decision-making: input new data into the trained model to obtain the predicted probability or category of wheat disease occurrence and provide corresponding decision-making recommendations.

[0018] Furthermore, the meteorological data collection is specifically as follows:

[0019] Arrange the sensor network, including temperature, humidity, light and wind speed sensors, set the data recording frequency, and ensure that the sensor network can cover different plots in the study area.

[0020] The historical disease data collection specifically includes: obtaining historical records of wheat diseases from agricultural databases or research institutions, including disease types, occurrence time and location, and collecting relevant disease image data to ensure that different diseases and their development stages are covered.

[0021] Furthermore, the data cleaning includes: checking missing values ​​and outliers in the data set, filling missing values ​​using interpolation or the mean of adjacent time periods, and deleting obviously erroneous or illogical data records.

[0022] The normalization process includes: normalizing all meteorological data and features.

[0023] The feature engineering includes: calculating the statistical characteristics of each meteorological variable, including: mean, standard deviation, maximum value and minimum value.

[0024] Extract important features related to the wheat growth cycle, including meteorological conditions during the growth phase.

[0025] Furthermore, the image data annotation includes: classifying and annotating the collected disease images to ensure that each image corresponds to a specific disease type. Using image processing tools to pre-process the images, including: scaling, cropping and denoising.

[0026] Furthermore, the knowledge graph construction includes combining expert knowledge with historical disease data to form a comprehensive disease information library.

[0027] Furthermore, the BiLSTM layer includes:

[0028] Define the parameters of the BiLSTM layer, including the number of units, activation function, and time step.

[0029] Construct forward and backward LSTM layers to capture the forward and backward dependencies of time series data.

[0030] The Transformer attention mechanism adds an attention mechanism layer to the output of BiLSTM to calculate the attention weight of each feature.

[0031] Furthermore, the workflow of the BiLSTM layer and Transformer attention mechanism is as follows:

[0032] The time series data is input into the BiLSTM layer, and BiLSTM extracts the features of the sequence, namely the temporal pattern of wheat disease occurrence and the changes in environmental factors.

[0033] The output of BiLSTM processing is a sequence vector containing time step features, each of which represents the information of each time step in the input sequence.

[0034] The output of BiLSTM is used as the input of Transformer. At this stage, the self-attention mechanism of Transformer can calculate the attention weight according to the features of each time step.

[0035] In this way, the Transformer focuses on those features that are most important for the current task (i.e., wheat disease prediction) and suppresses irrelevant information.

[0036] The Transformer's multiple encoder layers process the input and use the self-attention mechanism to integrate feature information from different time steps.

[0037] The final output of Transformer can be used for classification, regression or other decision support tasks. In other words, in wheat disease prediction, the model predicts the probability or type of disease occurrence based on the integrated features.

[0038] Furthermore, the model training uses a cross entropy loss function and an Adam optimization algorithm to optimize model parameters.

[0039] Furthermore, the prediction and decision-making are based on the model output to obtain the predicted probability and type of disease occurrence and provide agricultural producers with recommendations on prevention and control measures.

[0040] The present invention also discloses a wheat disease efficient prediction system, which can be used to implement the above-mentioned wheat disease efficient prediction method, specifically comprising:

[0041] Data collection module, used to collect meteorological data and historical disease data related to wheat diseases, and collect meteorological data in real time through sensor networks;

[0042] The data preprocessing module is responsible for cleaning, normalizing and feature engineering the collected data, and labeling the wheat disease images;

[0043] The knowledge graph construction module builds a wheat disease knowledge graph based on traditional agricultural expert knowledge and literature to form a comprehensive disease information database;

[0044] Model modules, including BiLSTM layers for processing time series data, and Transformer attention mechanism layers for calculating attention weights based on the output of the BiLSTM layers;

[0045] The model training module uses the preprocessed data to train the constructed model and uses the loss function and optimization algorithm to adjust the model parameters;

[0046] The prediction and decision-making module inputs new data into the trained model to obtain the predicted probability or category of wheat disease occurrence and provide corresponding decision-making recommendations.

[0047] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned efficient wheat disease prediction method when executing the program.

[0048] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned efficient wheat disease prediction method is implemented.

[0049] Compared with the prior art, the advantages of the present invention are:

[0050] 1. Comprehensive information capture: This method can simultaneously consider past and future information, comprehensively capture multiple factors related to the occurrence of wheat diseases, and significantly improve the accuracy of prediction.

[0051] 2. Key information screening: Transformer’s attention mechanism effectively screens out the most critical information for disease prediction, reducing the processing of unnecessary data, thereby improving training efficiency and reducing the consumption of computing resources.

[0052] 3. Real-time prediction capability: This method can quickly predict the data collected in real time and provide timely decision-making suggestions, enabling agricultural producers to respond quickly to environmental changes and optimize management strategies.

[0053] 4. Visualization and user-friendliness: The model's prediction results and decision recommendations can be intuitively displayed through a visual interface or mobile application, making it easy for agricultural producers to view and operate at any time, thereby improving the efficiency and convenience of agricultural production management.

[0054] 5. Enhance model generalization ability: This method reduces the dependence on specific data patterns, reduces the risk of overfitting, improves the performance and adaptability of the model in different scenarios, and maintains good prediction results when sample data is insufficient.

[0055] 6. Scientific decision support: Combined with the background knowledge constructed by the knowledge graph, this method provides agricultural experts with more scientific decision support tools to help them make more reasonable disease management strategies based on data. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of constructing a knowledge graph according to an embodiment of the present invention;

[0057] Figure 2 1 is a structural diagram of a bidirectional long short-term memory network (BiLSTM) according to an embodiment of the present invention;

[0058] Figure 3 It is a diagram of the overall network architecture of the long short-term memory network (LSTM) of an embodiment of the present invention;

[0059] Figure 4 This is a diagram of a forget gate architecture according to an embodiment of the present invention;

[0060] Figure 5 is a diagram of the input gate architecture of an embodiment of the present invention;

[0061] Figure 6 is a diagram of an output gate architecture according to an embodiment of the present invention;

[0062] Figure 7 It is a basic structural diagram of the Transformer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

[0064] The present invention provides an efficient prediction method for wheat diseases based on deep learning, comprising:

[0065] 1. Data Collection

[0066] Responsible for collecting various data related to wheat diseases, including meteorological data (such as temperature, humidity, rainfall, light time, etc.), pictures of wheat diseases, and historical disease data of a certain place. By deploying a sensor network in the farmland to collect meteorological data in real time, historical disease data and related statistical information are obtained from the agricultural database.

[0067] 2. Data Preprocessing

[0068] The collected data is cleaned to remove abnormal values ​​and missing values. For example, for temperature data with values ​​that are obviously beyond the reasonable range or missing time period data, the average value of the adjacent time period data is used to fill in the missing values.

[0069] Normalize the data to unify the data of different dimensions into the same numerical range (normalization) to facilitate model processing. Perform feature engineering to extract features with high correlation with wheat diseases. Calculate the mean value, standard deviation and other statistics of meteorological data as new features.

[0070] The collected wheat disease images are labeled with corresponding disease types for later training and identification to achieve real-time monitoring.

[0071] 3. Construction of Knowledge Graph

[0072] like Figure 1 As shown in the figure, by building a wheat disease knowledge graph, traditional agricultural expert knowledge and literature information are combined to form a comprehensive and systematic disease information database. The innovation of this step is to apply the knowledge graph to the agricultural field, providing rich feature information and background knowledge for subsequent prediction models.

[0073] 4. Model building and training module

[0074] 1. BiLSTM layer: such as Figure 2 As shown in the figure, a bidirectional long short-term memory network layer is constructed to process time series data. In the forward propagation process, the characteristics and trends of the data are learned from the beginning to the end of the time series; in the backward propagation process, the reverse dependency of the data is learned from the end to the beginning. In this way, the information at different time points in the wheat growth process and the related information before and after can be fully captured, which can better adapt to the temporal dynamics of the occurrence and development of diseases.

[0075] Principle: The bidirectional LSTM network is constructed by two LSTM networks in opposite directions. The overall architecture of a single LSTM network is as follows: Figure 3 As shown, the internal structure of LSTM can be clearly and intuitively seen. The green part represents the input x of this round. t and output y t ; The blue part represents the state h of the previous round t-1 And the state quantity h of this round output t ; The red part represents the information carrier c of the previous round t-1 and the information carrier c output in this round t Cascading multiple LSTM units together forms an LSTM deep learning network.

[0076] The reason why the entire LSTM architecture can remember long-term information is mainly due to the state c^n. t-1 to c t There is only a small amount of information interaction in the middle, so the information of the entire network can be transmitted between LSTMs.

[0077] The reason why LSTM can remember long-term and short-term information is that it has a "gate" structure to remove and add information to neurons. The "gate" is a method of selectively allowing information to pass through.

[0078] Here are the three “gates” of LSTM:

[0079] (1) Forget gate: The first step in LSTM is to decide what information we need to forget from the neuron state. As shown in the figure below, two inputs pass through a sigmoid function, so the output value is between 0 and 1, 1 means that the information is completely retained, and 0 means that the information is completely forgotten. Through the forget gate, LSTM can selectively forget some meaningless information. Figure 4 As shown in the figure, this part is the forget gate in LSTM.

[0080] (2) Input gate: It is necessary to determine what new information is stored in the neuron state. This part has two inputs: a sigmoid layer that determines what value the LSTM needs to update, and a tanh layer that creates a new candidate value vector that will be added to the state. These two pieces of information are then used to generate an update to the state. This is called the input gate. The process is as follows: Figure 5 shown.

[0081] (3) Output gate: In this process, z f *c t-1 Indicates the previous status information c t-1Forget some of the information that needs to be discarded, and then add the new candidate value vector of the LSTM system, which is the new round of information c of the system. t , we also need to update the system's neuron state h t The whole process is like Figure 6 shown.

[0082] This is the output gate that controls the output of LSTM. The system needs to determine what value to output. This output will also be based on the current state of the neuron. First, we use a sigmoid to determine which parts of the neuron need to output. Then, we process the information of the LSTM system through a tanh function, and finally multiply them and output them, which is the new state of LSTM.

[0083] Reversely splice the two LSTM networks to form a bidirectional long short-term memory network, and then splice and sum or weightedly fuse the outputs of the forward and reverse order. With this result, we can accurately predict the probability of various wheat diseases. In addition, we will adjust the network structure and optimize the algorithm to improve its prediction accuracy and model robustness.

[0084] At the same time, using the partial least squares method:

[0085] Partial Least Squares (PLS) is a statistical method that is mainly used to process data sets with multicollinearity and extract information from them for prediction. This method can better solve problems that cannot be solved by ordinary multivariate regression in the past, and can perform regression modeling when the number of sample points is less than the number of predictor variables.

[0086] The basic idea of ​​partial least squares is to project the high-dimensional data space of the independent variable and the dependent variable into the corresponding low-dimensional space, obtain the mutually orthogonal eigenvectors of the independent variable and the dependent variable, and then establish a univariate linear regression relationship between the eigenvectors of the independent variable and the dependent variable. This not only overcomes the problem of collinearity, but also emphasizes the explanatory and predictive role of the independent variable on the dependent variable when selecting the eigenvector, removes the influence of useless noise on regression, and makes the model contain the least number of variables.

[0087] One advantage of partial least squares is that it can perform regression modeling when there are severe multiple correlations among the independent variables, and allows regression modeling when the number of sample points is less than the number of variables. In addition, the partial least squares regression model is easier to identify system information and noise (even some non-random noise), and in the partial least squares regression model, the regression coefficient of each independent variable will be easier to interpret.

[0088] For this embodiment, its advantages are as follows:

[0089] 1. Dimensionality reduction and feature extraction

[0090] ① Processing high-dimensional data:

[0091] In wheat disease prediction, a large amount of feature data may be collected, such as various meteorological parameters, soil characteristics, wheat growth indicators, etc. These data often have high dimensions, which may lead to problems such as increased computational complexity and increased risk of overfitting.

[0092] Partial least squares (PLS) can reduce the dimensionality of high-dimensional data and extract the most valuable comprehensive features for wheat disease prediction. For example, multiple related meteorological variables (such as temperature, humidity, rainfall, etc.) are combined into several new latent variables, which not only retain the main information of the original data but also reduce the dimensionality of the data.

[0093] A PLS model is established, forming an input matrix X and an output matrix Y and P quality variables. PLS projects X and Y into a low-dimensional space defined by L latent variables as follows:

[0094]

[0095] where T = [t1, ..., t L ] is the potential score vector, P = [p1, ..., p L ] and Q=[q1,...q L ] are the loadings of X and Y, respectively. E and F are the PLS residuals corresponding to X and Y. The number of latent factors L is usually determined by cross-validation, which gives the maximum predictive power based on data excluded from the training data.

[0096] ② Extract key features:

[0097] Although the Bidirectional Long Short-Term Memory (BiLSTM) network can process time series data and capture certain feature information, it may not be able to fully extract the most critical features for very complex multivariate data.

[0098] After ① processing the high-dimensional data, the potential vector t can be calculated from the data sequence. i , in order to minimize the covariance between the input data:

[0099]

[0100] X1=X

[0101] And output the data Y of each factor. Among them, the weight vector w is introduced i To calculate the score t i =X i wi , the score can be expressed using the original data X as:

[0102] T=XR

[0103] Where R = W(P T W) -1 , from which we can get the following relationship:

[0104] R T =R T P=I l

[0105] To compute model estimates and residuals on new data samples x, the PLS model introduces an oblique projection into the input data space:

[0106]

[0107] in and is the oblique projection of x.

[0108] In summary, partial least squares (PLS) can extract the most relevant feature combination for wheat disease prediction by finding the direction of maximum covariance between input variables and output variables (disease occurrence). This helps to highlight those factors that have an important impact on disease occurrence and improve the prediction accuracy of the model.

[0109] 2. Solve the problem of multicollinearity

[0110] ① Dealing with complex data relationships:

[0111] In wheat disease prediction data, there may be strong correlations between different characteristic variables, namely multicollinearity. For example, some meteorological variables may affect each other, and soil properties and wheat growth indicators may also have complex associations.

[0112] This multicollinearity will affect the stability and accuracy of the model. PLS can effectively deal with the problem of multicollinearity by linearly combining variables and finding new unrelated latent variables, thereby eliminating the impact of collinearity.

[0113] ② Improve model stability:

[0114] When BiLSTM processes data with multicollinearity, it may be disturbed, resulting in unstable training or inaccurate prediction results.

[0115] After the introduction of PLS, the data can be preprocessed to reduce the impact of collinearity, so that BiLSTM can better learn the effective patterns in the data and improve the stability and reliability of the model.

[0116] 2. Transformer attention mechanism layer: Add a Transformer attention mechanism layer to the output of the BiLSTM layer. This layer calculates the attention weight of each time step or feature to highlight the information that contributes more to disease prediction and suppress unimportant information. For example, higher attention weights are given to certain specific meteorological conditions or wheat growth stage characteristics because they may be more closely related to the occurrence of diseases.

[0117] like Figure 7 As shown, Transformer consists of four parts as a whole:

[0118] 1.Inputs:Inputs=WordEmbedding(Inputs)+PositionalEmbedding

[0119] 2.Outputs:Ouputs=WordEmbedding(Outputs)+PositionalEmbedding

[0120] 3.Encoders stack: It consists of six identical encoder layers. Except for the first encoder layer whose input is Inputs, the input of other encoder layers is the output of the previous encoder layer.

[0121] 4.Decoders stack: It consists of six identical decoder layers. Except for the first decoder layer, whose input is Outputs and the output of the last encoder layer, the input of other decoder layers is the output of the previous decoder layer and the output of the last encoder layer.

[0122] 3. Model training: Use the preprocessed data to train the constructed model. Define a loss function, such as the cross entropy loss function, to measure the difference between the model's prediction results and the true label. Use an optimization algorithm, such as the Adam optimizer, to continuously adjust the model's parameters to minimize the loss function. During the training process, divide the dataset into a training set, a validation set, and a test set. The training set is used to learn the model's parameters, the validation set is used to monitor the model's performance and adjust the hyperparameters, and the test set is used to finally evaluate the model's generalization ability.

[0123] 4. Bidirectional long short-term memory network and Transformer attention mechanism splicing (serial combination): First, let the data pass through the bidirectional long short-term memory network for preliminary sequence information processing and feature extraction, and then use the output of BiLSTM as the input of the Transformer attention mechanism, so that the Transformer can further perform more advanced semantic understanding and feature integration on these features. BiLSTM can first capture the contextual information and long-term dependencies in the text, and then the Transformer attention mechanism can more accurately focus on the most important part for classification based on this information.

[0124] 5. Forecasting and Decision-making

[0125] Input new data into the trained model to obtain the predicted probability and category of wheat disease occurrence. For example, the probability of wheat being infected with a certain disease (such as rust, powdery mildew, fusarium head blight, etc.) in the next week is predicted to be a value between [0-1].

[0126] Based on the prediction results, decision-making recommendations are provided to agricultural producers, such as whether to take preventive measures in advance, which preventive agents to choose, when to irrigate or fertilize, etc., to reduce the risk and losses of disease occurrence.

[0127] In another embodiment of the present invention, a system for efficiently predicting wheat diseases is provided. The system can be used to implement the above-mentioned method for efficiently predicting wheat diseases based on deep learning, and specifically includes:

[0128] Data collection module, used to collect meteorological data and historical disease data related to wheat diseases, and collect meteorological data in real time through sensor networks;

[0129] The data preprocessing module is responsible for cleaning, normalizing and feature engineering the collected data, and labeling the wheat disease images;

[0130] The knowledge graph construction module builds a wheat disease knowledge graph based on traditional agricultural expert knowledge and literature to form a comprehensive disease information database;

[0131] Model modules, including BiLSTM layers for processing time series data, and Transformer attention mechanism layers for calculating attention weights based on the output of the BiLSTM layers;

[0132] The model training module uses the preprocessed data to train the constructed model and uses the loss function and optimization algorithm to adjust the model parameters;

[0133] The prediction and decision-making module inputs new data into the trained model to obtain the predicted probability or category of wheat disease occurrence and provide corresponding decision-making recommendations.

[0134] In another embodiment of the present invention, a terminal device is provided, the terminal device includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions; the processor described in the embodiment of the present invention can be used for the operation of the efficient prediction method of wheat diseases.

[0135] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0136] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the efficient wheat disease prediction method in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by the processor.

[0137] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and should be understood that the protection scope of the present invention is not limited to such special statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. An efficient prediction method for wheat diseases based on deep learning, characterized in that: The following steps are involved: S1: Data collection, collecting meteorological data and historical disease data related to wheat diseases, and collecting meteorological data in real time by deploying a sensor network in the farmland; S2: Data preprocessing: cleaning, normalizing and feature engineering the collected data to extract features related to wheat diseases and label wheat disease images; S3: Knowledge graph construction: Based on the knowledge of traditional agricultural experts and literature, a knowledge graph of wheat diseases is established to form a comprehensive disease information database; S4: Model construction, including: S4.1: BiLSTM layer, used to process time series data and learn the characteristics and trends of wheat growth; S4.2: Transformer attention mechanism layer, based on the output of the BiLSTM layer, calculates the attention weight of each time step to highlight the information that contributes most to disease prediction; S5: Model training, using the preprocessed data to train the constructed model, and using loss function and optimization algorithm to adjust the model parameters; S6: Prediction and decision-making: input new data into the trained model to obtain the predicted probability or category of wheat disease occurrence and provide corresponding decision-making recommendations.

2. The efficient prediction method for wheat diseases according to claim 1, characterized in that: The meteorological data collection is specifically as follows: Arrange the sensor network, including temperature, humidity, light and wind speed sensors, set the data recording frequency, and ensure that the sensor network covers different plots in the study area; The historical disease data collection specifically includes: obtaining historical records of wheat diseases from agricultural databases or research institutions, including disease types, occurrence time and location; collecting relevant disease image data to ensure that different diseases and their development stages are covered.

3. The efficient wheat disease prediction method according to claim 1, characterized in that: The data cleaning includes: checking missing values ​​and outliers in the data set, filling missing values ​​using interpolation or the mean of adjacent time periods, and deleting obviously erroneous or illogical data records; The normalization process includes: normalizing all meteorological data and features; The feature engineering includes: calculating the statistical characteristics of each meteorological variable, including: mean, standard deviation, maximum value and minimum value; Extract important features related to the wheat growth cycle, including meteorological conditions during the growth phase.

4. The efficient wheat disease prediction method according to claim 1, characterized in that: The image data annotation includes: classifying and annotating the collected disease images to ensure that each image corresponds to a specific disease type; and preprocessing the images using image processing tools, including: scaling, cropping and denoising.

5. The efficient wheat disease prediction method according to claim 1, characterized in that: The knowledge graph construction includes combining expert knowledge with historical disease data to form a comprehensive disease information base.

6. The efficient wheat disease prediction method according to claim 1, characterized in that: The BiLSTM layer includes: Define the parameters of the BiLSTM layer, including the number of units, activation function, and time step; Construct forward and backward LSTM layers to capture the forward and backward dependencies of time series data; The Transformer attention mechanism adds an attention mechanism layer to the output of BiLSTM to calculate the attention weight of each feature.

7. The efficient wheat disease prediction method according to claim 1, characterized in that: The workflow of the BiLSTM layer and Transformer attention mechanism is as follows: The time series data is input into the BiLSTM layer, and BiLSTM extracts the features of the sequence, i.e., the temporal pattern of wheat disease occurrence and the changes in environmental factors; The output of BiLSTM processing is a sequence vector containing time step features, each of which represents the information of each time step in the input sequence; The output of BiLSTM is used as the input of Transformer. At this stage, the self-attention mechanism of Transformer can calculate the attention weight according to the features of each time step. In this way, Transformer focuses on those features that are most important for the current task (i.e. wheat disease prediction) and suppresses irrelevant information; Transformer's multiple encoder layers process the input and use the self-attention mechanism to integrate feature information from different time steps; The final output of Transformer can be used for classification, regression or other decision support tasks; that is, in wheat disease prediction, the model predicts the probability or type of disease occurrence based on the integrated features.

8. The efficient wheat disease prediction method according to claim 1, characterized in that: The model training uses the cross entropy loss function and the Adam optimization algorithm to optimize the model parameters.

9. The efficient wheat disease prediction method according to claim 1, characterized in that: The prediction and decision-making are based on the model output, obtaining the predicted probability and type of disease occurrence and providing suggestions for prevention and control measures for agricultural producers.

10. An efficient wheat disease prediction system, characterized by: The system can be used to implement the efficient wheat disease prediction method according to any one of claims 1 to 9, specifically comprising: Data collection module, used to collect meteorological data and historical disease data related to wheat diseases, and collect meteorological data in real time through sensor networks; The data preprocessing module is responsible for cleaning, normalizing and feature engineering the collected data, and labeling the wheat disease images; The knowledge graph construction module builds a wheat disease knowledge graph based on traditional agricultural expert knowledge and literature to form a comprehensive disease information database; Model modules, including BiLSTM layers for processing time series data, and Transformer attention mechanism layers for calculating attention weights based on the output of the BiLSTM layers; The model training module uses the preprocessed data to train the constructed model and uses the loss function and optimization algorithm to adjust the model parameters; The prediction and decision-making module inputs new data into the trained model to obtain the predicted probability or category of wheat disease occurrence and provide corresponding decision-making recommendations.

Images