A wheat scab prediction method, device, equipment and storage medium

Abstract

This application discloses a method, apparatus, device, and storage medium for predicting wheat scab. The method involves inputting meteorological and remote sensing factors corresponding to the wheat plant to be predicted into a target wheat scab prediction model for prediction processing to determine the disease incidence rate of the wheat plant. The target wheat scab prediction model is a model trained based on wheat plant samples. These wheat plant samples include wheat plants in several specific states, and each sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors, and disease incidence rate. Furthermore, there are transmission relationships for wheat scab among the various specific states. Based on this, the prediction process not only considers the influence of meteorological and remote sensing factors on the disease but also takes into account the disease's own epidemic mechanism, thus improving the accuracy of predicting wheat plants suffering from wheat scab.

Description

Technical Field

[0001] This application relates to the field of agricultural pest prediction, and more specifically, to a method, apparatus, equipment, and storage medium for predicting wheat scab. Background Technology

[0002] Fusarium head blight (FHB) is one of the most destructive wheat diseases. FHB occurs directly on the wheat head and can cause a sharp decline in wheat yield. Furthermore, because the pathogen in infected FHB wheat produces mycotoxins, the accidental consumption of infected wheat heads by humans could lead to food safety issues for humans or animal feed. Therefore, accurate prediction of FHB is currently of paramount importance. Summary of the Invention

[0003] In view of this, this application provides a method, apparatus, equipment and storage medium for predicting wheat scab, in order to solve the problem of low accuracy in predicting wheat scab.

[0004] To achieve the above objectives, the following solution is proposed:

[0005] A method for predicting wheat scab includes:

[0006] Obtain the meteorological and remote sensing factors corresponding to the wheat plants to be predicted;

[0007] The meteorological factors and the remote sensing factors are input into the target wheat Fusarium head blight prediction model for prediction processing to obtain the diseased ear rate of the wheat plant to be predicted.

[0008] The target wheat scab prediction model is a model trained based on wheat plant samples. The wheat plant samples include wheat plant samples in several specific states, and each wheat plant sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors, and disease prevalence. Furthermore, there is a transmission relationship between the various specific states for wheat scab.

[0009] Optionally, the method may further include:

[0010] Wheat plant samples matching each specific state, along with corresponding meteorological factors, remote sensing factors, and disease rate, are obtained to create a sample dataset. The specific states include susceptible state, latent state, infected state, and non-infectious state.

[0011] The sample dataset is divided according to a preset division ratio to obtain a training dataset, a validation dataset, and a test dataset.

[0012] Obtain the preset initial wheat scab prediction model;

[0013] The initial wheat scab prediction model is trained based on the training dataset and the validation dataset to obtain the target wheat scab prediction model. The training process uses the meteorological factors and the remote sensing factors in the training dataset as feature values ​​and the diseased ear rate corresponding to the meteorological factors and the remote sensing factors as the target value for training.

[0014] The meteorological factors and remote sensing factors in the test dataset are input into the target wheat Fusarium head blight prediction model to obtain the diseased ear rate to be tested;

[0015] Based on the diseased ear rate and the diseased ear rate to be tested in the test dataset, a diseased ear rate level correspondence table for the target wheat Fusarium head blight prediction model is determined, and the diseased ear rate level correspondence table characterizes the disease severity corresponding to the diseased ear rate.

[0016] Optionally, obtaining the preset initial wheat scab prediction model may include:

[0017] Based on the transmission process of wheat scab in the wheat plant samples, the infection rate corresponding to each specific state is preset;

[0018] Based on each specific state and the infection rate, a differential equation corresponding to each specific state is determined, and the differential equation characterizes the propagation relationship between the specific states;

[0019] Based on the differential equation and convolutional neural network corresponding to each specific state, disease simulation sub-models are created respectively corresponding to each specific state. The disease simulation sub-models are used to simulate the wheat plant growth process in the specific state based on meteorological factors and remote sensing meteorological factors, and the model structure of each disease simulation sub-model is different.

[0020] Based on the propagation relationship between the four specific states, the disease simulation sub-models corresponding to each specific state are spliced ​​together to obtain a single-day disease simulation sub-model.

[0021] The single-day disease simulation sub-model is concatenated with the long short-term memory network sub-model to obtain a multi-day disease simulation sub-model. The long short-term memory network sub-model is used to input the feature information output by the single-day disease simulation sub-model into the single-day disease simulation sub-model.

[0022] The multi-day disease simulation sub-model is concatenated with a fully connected layer and an activation function layer to obtain an initial wheat scab model. The fully connected layer and the activation function layer are used to process the feature information output by the multi-day disease simulation sub-model to obtain the disease spike rate corresponding to the feature information.

[0023] Optionally, training the initial wheat scab prediction model based on the training dataset and the validation dataset to obtain the target wheat scab prediction model may include:

[0024] Based on the validation dataset, the hyperparameters of the initial wheat scab prediction model are initialized, including the learning rate;

[0025] The meteorological factors and remote sensing factors of the wheat plant samples in the training dataset are input into the initial wheat Fusarium head blight prediction model for prediction processing to obtain the predicted disease rate;

[0026] Based on the predicted diseased ear rate and the diseased ear rate in the training dataset, a loss function is determined;

[0027] Based on the loss function, determine the cost function;

[0028] New model parameters are determined based on the cost function, and the new model parameters include weight values ​​and bias values.

[0029] Based on the new model parameters, the step of determining the loss function based on the predicted diseased ear rate and the diseased ear rate in the training dataset is performed to obtain the new loss function;

[0030] If the new loss function satisfies the convergence state, the new model parameters corresponding to the new loss function are determined as the target model parameters of the initial wheat scab prediction model, and the target wheat scab prediction model is determined based on the target model parameters;

[0031] If the new loss function does not satisfy the convergence state, the initial wheat scab prediction model is updated with the new model parameters, and the meteorological factors and remote sensing factors of the wheat plant samples in the training dataset are input into the initial wheat scab dynamic prediction model for prediction processing to obtain the predicted diseased ear rate, until the new loss function satisfies the convergence state.

[0032] Optionally, the step of creating a disease simulation sub-model corresponding to each specific state based on the differential equation and convolutional neural network for each specific state may include:

[0033] Based on the differential equation corresponding to the infectious state and the preset infectious state disease simulation sub-model structure, the disease simulation sub-model corresponding to the infectious state is determined. The preset infectious state disease simulation sub-model structure includes a convolutional kernel, a convolutional layer, a pooling layer, a batch normalization processing layer, and an activation function layer generated based on the first preset parameters.

[0034] Based on the differential equation corresponding to the susceptible state and the disease simulation sub-model structure of the susceptible state, the disease simulation sub-model corresponding to the susceptible state is determined. The disease simulation sub-model structure of the susceptible state includes a convolutional kernel, a convolutional layer, a batch normalization processing layer and an activation function layer generated based on the second preset parameters.

[0035] Based on the differential equation corresponding to the latent state and the latent state disease simulation sub-model structure, the disease simulation sub-model corresponding to the latent state is determined. The latent state disease simulation sub-model structure includes a convolutional kernel, a convolutional layer, a batch normalization layer, a pooling layer, and an activation function layer generated based on a third preset parameter.

[0036] Based on the differential equation corresponding to the non-infectious state and the structure of the disease simulation sub-model for the non-infectious state, the disease simulation sub-model corresponding to the non-infectious state is determined. The structure of the disease simulation sub-model for the non-infectious state includes a convolutional kernel, a convolutional layer, a batch normalization layer, and an activation function layer generated based on a fourth preset parameter.

[0037] Optionally, the target wheat scab prediction model includes a daily disease simulation sub-model and a long short-term memory network sub-model. The daily disease simulation sub-model includes disease simulation sub-models corresponding to four specific states, namely, susceptible state, latent state, infectious state, and non-infectious state.

[0038] The step of inputting the meteorological factors and the remote sensing factors into the target wheat Fusarium head blight prediction model for prediction processing to obtain the diseased ear rate corresponding to the wheat plant to be predicted may include:

[0039] Based on the meteorological factors and the remote sensing factors, predictive features are generated;

[0040] The predicted features are input into the single-day disease simulation sub-model;

[0041] Based on preset processing conditions and the four disease simulation sub-models, the predicted features are processed to obtain feature information corresponding to the four disease simulation sub-models respectively.

[0042] The four feature information pieces are concatenated to obtain combined feature information, and the number of concatenation processes is recorded.

[0043] If the number of splicing processes is less than the number threshold, the combined feature information is input as a prediction feature into the daily disease simulation model. The step of processing the prediction feature based on preset processing conditions and four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models is executed until the number of splicing processes is not less than the number threshold.

[0044] If the number of splicing processes is not less than the threshold number, the combined features are mapped to obtain the diseased ear rate corresponding to the wheat plant to be predicted.

[0045] Optionally, the step of processing the predicted features based on preset processing conditions and the four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models may include:

[0046] Based on the disease simulation sub-model corresponding to the infectious state, the predicted features are processed to obtain the initial infectious state feature information;

[0047] The initial infection state feature information is concatenated with the predicted feature to obtain the first input feature;

[0048] The first input feature is input into the disease simulation sub-model corresponding to the susceptible state for processing to obtain susceptible state feature information;

[0049] The easily infectable state feature information and the predicted feature are concatenated to obtain the second input feature;

[0050] The second input feature is input into the disease simulation sub-model corresponding to the latent state for processing to obtain latent state feature information;

[0051] The latent state feature information and the predicted feature are concatenated to obtain the third input feature;

[0052] The third input feature is input into the disease simulation sub-model corresponding to the infectious state for processing to obtain infectious state feature information;

[0053] The infection status feature information and the prediction feature are concatenated to obtain the fourth input feature;

[0054] The fourth input feature is input into the disease simulation sub-model corresponding to the non-infectious state for processing to obtain the non-infectious state feature information;

[0055] The susceptible state feature information, the latent state feature information, the infectious state feature information, and the non-infectious state feature information are determined as feature information corresponding to the four disease simulation sub-models, respectively.

[0056] A wheat scab prediction device, comprising:

[0057] The information acquisition unit acquires the meteorological factors and remote sensing factors corresponding to the wheat plants to be predicted.

[0058] The information processing unit is used to input the meteorological factors and the remote sensing factors into the target wheat Fusarium head blight prediction model for prediction processing, and obtain the diseased ear rate corresponding to the wheat plant to be predicted.

[0059] The target wheat scab prediction model is a model trained based on wheat plant samples. The wheat plant samples include wheat plant samples in several specific states, and each wheat plant sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors, and disease prevalence. Furthermore, there are transmission relationships between different specific states for wheat scab.

[0060] A wheat scab prediction device, including a memory and a processor;

[0061] The memory is used to store programs;

[0062] The processor is used to execute the program to implement any of the steps of the above-described wheat scab prediction method.

[0063] A storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described wheat scab prediction methods.

[0064] As can be seen from the above technical solutions, the wheat scab prediction method, device, equipment and storage medium provided in this application embodiment can predict wheat scab using meteorological factors and remote sensing factors. Moreover, during the prediction, the growth process of wheat plants under specific conditions during the spread of wheat scab can be simulated, and the transmission relationship between various specific conditions of wheat plants can be simulated. Based on this, not only can the influence of meteorological factors and remote sensing factors on the disease be taken into account during the prediction process, but the epidemic mechanism of the disease itself can also be considered. By taking multiple factors into account, the accuracy of predicting wheat plants suffering from wheat scab can be improved. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0066] Figure 1 This is a schematic flowchart illustrating a method for predicting wheat scab provided in an embodiment of this application.

[0067] Figure 2 A schematic diagram of an optional process for creating a target wheat scab prediction model provided in an embodiment of this application;

[0068] Figure 3 This is a schematic diagram illustrating the propagation relationship between specific states provided in an embodiment of this application.

[0069] Figure 4 This is a schematic diagram of the structure of a disease simulation sub-model corresponding to an infectious state, provided in an embodiment of this application.

[0070] Figure 5 This is a schematic diagram of the structure of a disease simulation sub-model corresponding to a susceptible state provided in an embodiment of this application;

[0071] Figure 6 This is a schematic diagram of the structure of a disease simulation sub-model corresponding to a latent state, provided in an embodiment of this application.

[0072] Figure 7 This is a schematic diagram of the structure of a disease simulation sub-model corresponding to a non-infectious state provided in an embodiment of this application;

[0073] Figure 8 This is a schematic diagram of the structure of an initial wheat scab prediction model provided in an embodiment of this application;

[0074] Figure 9 A schematic diagram of a wheat scab prediction device provided in this application embodiment;

[0075] Figure 10 This is a schematic diagram of a wheat scab prediction device provided in an embodiment of this application. Detailed Implementation

[0076] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0077] This application can be applied to the prediction of wheat scab in agronomic disease prediction. Observers can input meteorological factors and remote sensing factors of the current environment of the wheat plant to be tested into the target wheat scab prediction model for simulation and prediction. The simulation will show the growth of the wheat plant under the influence of the meteorological factors and remote sensing factors, as well as the wheat plant during the spread of wheat scab, and further predict the degree of wheat scab disease in the wheat plant to be predicted. The degree of disease can be expressed as the diseased ear rate of the wheat plant.

[0078] Based on this, we can simulate the growth of wheat plants using meteorological and remote sensing factors, taking into account parameters such as the growth environment and growth trend of wheat plants. We can also simulate the spread of wheat scab among wheat plants based on the transmission relationship between different specific states of wheat scab, and finally obtain the diseased ear rate corresponding to the wheat plants. This comprehensive approach improves the accuracy and efficiency of wheat scab prediction.

[0079] Reference Figure 1 The illustration shows a flowchart of a method for predicting wheat scab according to an embodiment of this application. The process may include:

[0080] Step S110: Obtain the meteorological factors and remote sensing factors corresponding to the wheat plants to be predicted.

[0081] The meteorological factors characterize the atmospheric state or weather phenomena of the environment in which the wheat plants are located, and in this embodiment, they may include temperature, humidity, and rainfall. The remote sensing factors may include vegetation indices characterizing the growth momentum and status of wheat plants. In this embodiment, the remote sensing factors may include Fiber Optic Vegetation Cover (FVC), Optimized Soil-Regulated Vegetation Index (OSAVI), and Normalized Difference Vegetation Index (NDVI).

[0082] In this embodiment, wheat plants do not refer to individual wheat plants, but can also refer to wheat covering a certain area. When acquiring meteorological and remote sensing factors for prediction, meteorological factors of the current wheat plants can be collected through weather forecasts, meteorological stations, etc., and remote sensing data of the planting area of ​​the wheat plants can be detected through remote sensing technology to collect remote sensing factors for prediction.

[0083] Step S120: Input meteorological factors and remote sensing factors into the target wheat Fusarium head blight prediction model for prediction processing to obtain the disease rate of the wheat plants to be predicted.

[0084] The input meteorological factors and remote sensing factors can generate a feature vector or feature information to facilitate computation and processing in the target wheat scab prediction model. In this embodiment, the meteorological factors and remote sensing factors can be combined into a prediction feature of size 1*1*6.

[0085] The target wheat scab prediction model is a model trained based on wheat plant samples. The wheat plant samples include wheat plant samples in several specific states, and each wheat plant sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors, and disease spike rate (DI value). Furthermore, there is a transmission relationship between the various specific states for wheat scab.

[0086] The specific states represent different disease stages of wheat plants. In this embodiment, specific states may include susceptible loading, latent state, infectious state, and non-infectious loading. Wheat plants have different infectivity rates at different disease stages, and there is a transmission sequence between different specific states. Based on this, the constructed target wheat scab prediction model can simulate the growth of wheat plants under different specific states, as well as the transmission between wheat plants under different specific states. Based on this, it can simulate the development process of wheat plants contracting wheat scab. In this embodiment, a complete transmission process can be recorded as a single day of transmission.

[0087] The target wheat scab prediction model may include disease simulation sub-models constructed based on convolutional neural networks, each corresponding to a specific state. These sub-models simulate the growth status of wheat plant samples or individual wheat plants under specific states based on meteorological and remote sensing factors (i.e., prediction features), and simulate the spread of wheat scab among wheat plants in each sub-model based on the transmission relationships between these specific states. Finally, by combining the growth environment data from meteorological and remote sensing factors with the transmission factors of wheat scab, the prediction results for a single transmission cycle of wheat plants are obtained. These prediction results may include information on the occurrence of wheat scab, and further processing based on these results yields the diseased ear rate corresponding to each wheat plant.

[0088] In addition, the target wheat scab prediction model can also realize the temporal simulation of the wheat scab epidemic process. First, there is a corresponding transmission relationship between each specific state in a complete transmission process, which reflects a temporal sequence. In addition, the above simulation and prediction completes the daily transmission process of wheat plants and obtains the corresponding prediction results. If the prediction period set by the observer is not limited to daily prediction, the long short-term memory network model can be used to execute the simulation and prediction process of daily transmission multiple times. The prediction results obtained from the previous daily transmission can be used as the input features of the next daily transmission, and as the prediction features of the next daily transmission cycle.

[0089] After multiple single-day transmissions, a multi-day prediction result is obtained. This result is then processed to obtain the disease spike rate corresponding to each single-day transmission in the multiple single-day transmissions. The disease spike rates corresponding to each single-day transmission are then integrated to obtain a disease spike rate index map. This allows observers to monitor the infection trend of wheat plants infected with wheat scab and to take timely treatment measures for the wheat plants.

[0090] Based on this, the embodiments of this application can consider parameters of the wheat plant's growth environment, namely meteorological factors and remote sensing factors, and can also refer to the transmission mechanism of wheat scab. By simulating the single-day transmission cycle, the effect of continuous multi-day transmission of wheat scab in wheat plants can be achieved. Dynamic prediction can be made for wheat plants to obtain the trend of wheat plants suffering from wheat scab and the corresponding diseased ear rate.

[0091] In summary, the embodiments of this application can make predictions using meteorological and remote sensing factors. Furthermore, during the prediction process, the growth process of wheat plants under specific conditions during the spread of wheat scab can be simulated, and the transmission relationship between various specific conditions of wheat plants can be simulated. Based on this, not only can the influence of meteorological and remote sensing factors on the disease be taken into account during the prediction process, but the epidemic mechanism of the disease itself can also be considered. By taking multiple factors into account, the accuracy of predicting wheat plants suffering from wheat scab can be improved.

[0092] Next, this application will further describe the method for predicting wheat scab.

[0093] Reference Figure 2 This document illustrates an optional flowchart for creating a target wheat scab prediction model according to an embodiment of this application. Based on the flowchart, the process for creating the target wheat scab prediction model is described, and the creation process may include:

[0094] Step S210: Obtain the sample data corresponding to the wheat plant samples and divide them to obtain the training dataset, validation dataset and test dataset.

[0095] Specifically, wheat plant samples matching each specific state, along with corresponding meteorological factors, remote sensing factors, and disease rate of the wheat plant samples, can be obtained to form a sample dataset. The specific state may include susceptible state, latent state, infected state, and non-infected state. The sample dataset is then divided based on a preset division ratio to obtain a training dataset, a validation dataset, and a test dataset.

[0096] In this embodiment, the sample dataset mainly consists of data collected from wheat plant samples at ground survey points. Specifically, it can collect meteorological factors, remote sensing factors, and DI (Discretionary Indicator) for each day at 158 ​​ground survey points, or it can collect meteorological factors, remote sensing factors, and disease rate of wheat plants under various specific conditions for each day as the sample dataset. The meteorological factors, remote sensing factors, and DI values ​​for the same day are stored accordingly to obtain the sample dataset. The data in the sample dataset can be collected over a time period of 15 days. The sample subset corresponding to each day within the time period is sorted chronologically to obtain a time series dataset.

[0097] Furthermore, the sample dataset is divided according to a preset ratio into a training dataset, a validation dataset, and a test dataset. The validation dataset is used to update the model parameters and hyperparameters of the initial wheat scab prediction model in conjunction with the training dataset, so that the prediction results of the final trained target wheat scab prediction model are more accurate.

[0098] Step S220: Obtain the preset initial wheat scab prediction model.

[0099] An initial wheat scab prediction model needs to be created in advance, and then trained or tested based on the initial wheat scab prediction model. In this embodiment, the initial wheat scab prediction model is a model created based on a compartmentalized model, a neural network, and a long short-term memory network. The specific process of creating the initial wheat scab model may include:

[0100] Based on the transmission process of wheat scab in the wheat plant samples, an infection rate corresponding to each specific state is preset; based on each specific state and the infection rate, a differential equation corresponding to each specific state is determined, and the differential equation represents the transmission relationship between the specific states; based on the differential equation corresponding to each specific state and a convolutional neural network, a disease simulation sub-model corresponding to each specific state is created, and the disease simulation sub-model is used to simulate the wheat plant growth process in the specific state based on meteorological factors and remote sensing meteorological factors, and the model structure of each disease simulation sub-model is different; based on the four specific states The propagation relationship between them is concatenated with the disease simulation sub-model corresponding to each specific state to obtain a single-day disease simulation sub-model; the single-day disease simulation sub-model is concatenated with a long short-term memory network sub-model to obtain a multi-day disease simulation sub-model, wherein the long short-term memory network sub-model is used to input the feature information output by the single-day disease simulation sub-model into the single-day disease simulation sub-model; the multi-day disease simulation sub-model is concatenated with a fully connected layer and an activation function layer to obtain an initial wheat scab model, wherein the fully connected layer and the activation function layer are used to process the feature information output by the multi-day disease simulation sub-model to obtain the disease spike rate corresponding to the feature information.

[0101] In this embodiment, the SEIR model (Susceptible-Exposed-Infected-Removed model, or susceptible-latent-infected-recovered model) is used to simulate the development process of wheat scab. Wheat plants are classified according to their infection status into susceptible (S), latent (E), infectious (I), and non-infected (R) wheat plants. Susceptible wheat plants are those that have never been infected with wheat scab; latent wheat plants are those currently in the incubation period of wheat scab; infectious wheat plants are those that have been infected and exhibit symptoms of wheat scab; and non-infected wheat plants are those that have recovered or died from the disease.

[0102] Wheat scab is a contagious plant disease. As wheat plants in different specific states change over time, with variations in their growing environment and individual conditions, the disease can spread between them. In the SEIR model, the transmission or spread relationships between different specific states can be referenced as follows: Figure 3 The diagram shown illustrates the propagation relationship between specific states provided in the embodiments of this application.

[0103] The transmission relationship between wheat plants in different specific states can be as follows: after a susceptible wheat plant (S) comes into contact with an infected wheat plant (I), it will become a latent wheat plant (E) with an infection rate of α. The latent wheat plant (E) will become an infected wheat plant (I) with an infection rate of β. As the external environment and the conditions of the wheat plant itself are no longer suitable for the development of the disease, the infected wheat plant (I) will become a non-infectious wheat plant (R) with a conversion rate of γ.

[0104] Based on the transmission relationships and infection rates among wheat plants in each specific state, a system of differential equations is generated to describe the evolution of wheat plants in the four specific states S, E, I, and R. Assuming the total number of wheat plants is always M, and the number of wheat plants removed from the non-infected state later is negligible, i.e., S(t) + I(t) + E(t) + R(t) = M, where S(t), I(t), E(t), and R(t) represent the number of wheat plants in states S, I, E, and R respectively on day t. Based on this, the system of differential equations corresponding to each specific state is as follows:

[0105]

[0106] As described above, α represents the transmission rate from wheat plants in state S to wheat plants in state E, β represents the transmission rate from wheat plants in state E to wheat plants in state I, and γ represents the transmission rate from wheat plants in state I to wheat plants in state R. Differential equation (1) can be used to represent the transmission relationship between susceptible and infectious states, differential equation (2) can represent the transmission relationship between latent and susceptible states, differential equation (3) can represent the transmission relationship between susceptible and infectious states, and differential equation (4) can represent the transmission relationship between infectious and non-infectious states.

[0107] Furthermore, based on the aforementioned system of differential equations and neural networks, disease simulation sub-models corresponding to each specific state are created, and each disease simulation sub-model is distinct from the others. Specifically, the process of creating a disease simulation sub-model corresponding to each specific state may include:

[0108] Based on the differential equation corresponding to the infectious state and the preset infectious state disease simulation sub-model structure, the disease simulation sub-model corresponding to the infectious state is determined. The preset infectious state disease simulation sub-model structure includes a convolutional kernel, a convolutional layer, a pooling layer, a batch normalization processing layer, and an activation function layer generated based on the first preset parameters.

[0109] Based on the differential equation corresponding to the susceptible state and the disease simulation sub-model structure of the susceptible state, the disease simulation sub-model corresponding to the susceptible state is determined. The disease simulation sub-model structure of the susceptible state includes a convolutional kernel, a convolutional layer, a batch normalization processing layer and an activation function layer generated based on the second preset parameters.

[0110] Based on the differential equation corresponding to the latent state and the latent state disease simulation sub-model structure, the disease simulation sub-model corresponding to the latent state is determined. The latent state disease simulation sub-model structure includes a convolutional kernel, a convolutional layer, a batch normalization layer, a pooling layer, and an activation function layer generated based on a third preset parameter.

[0111] Based on the differential equation corresponding to the non-infectious state and the structure of the disease simulation sub-model for the non-infectious state, the disease simulation sub-model corresponding to the non-infectious state is determined. The structure of the disease simulation sub-model for the non-infectious state includes a convolutional kernel, a convolutional layer, a batch normalization layer, and an activation function layer generated based on a fourth preset parameter.

[0112] Convolutional neural networks typically consist of convolutional layers, pooling layers, fully connected layers, and activation function layers. In this embodiment, each disease simulation sub-model employs a convolutional layer. This convolutional layer can effectively extract features strongly correlated with wheat scab from the input meteorological and remote sensing factors, while also effectively reducing the number of parameters and simplifying the computation process. Specifically, the extraction process is mainly completed by the convolutional kernels in the convolutional layer performing calculations on the meteorological and remote sensing factors. The calculation formula can be found in the following formula (5):

[0113] M(i,j)=∑ m ∑ n I(i+m,j+n)*K(i,j) (5)

[0114] Where I represents the matrix corresponding to the predicted features formed by the combination of meteorological factors and remote sensing factors, and K represents the convolution kernel. The convolutional layer includes a set of multiple learnable convolution kernels, and the size, stride, and padding size of the convolution kernels need to be preset. In this embodiment, the convolution kernel of each disease simulation sub-model can be set and created according to preset parameters, which may include convolution kernel setting parameters such as kernel size, stride, and padding size. When setting these preset parameters, the parameters can be determined by statistically analyzing past convolution operation cases, or custom parameters can be defined.

[0115] In this embodiment, the use of pooling layers in the disease simulation sub-model can reduce the complexity of features related to wheat scab detection in meteorological and remote sensing factors, reduce the computational load in the simulation and prediction process, improve the training efficiency of the disease model sub-model and the subsequent initial wheat scab prediction model, and prevent overfitting. Similarly, the pooling size, step size, and pooling type of each pooling layer used to build the model need to be preset. These can be set according to the corresponding pooling layer parameters in the preset parameters, and the pooling type can be preset to max pooling or average pooling.

[0116] In summary, based on the above set of differential equations (differential equations (1), (2), (3), (4)), the above convolutional layers and pooling layers are used to simulate wheat plants in a specific state.

[0117] In this embodiment, based on the differential equation (3) corresponding to the infectious state (I), a disease simulation sub-model corresponding to the infectious state is constructed using a convolutional layer, a BN layer (batch normalization layer), a ReLU layer (activation function layer), a max pooling layer, and a ReLU layer. For details, please refer to... Figure 4 This illustrates a schematic diagram of a disease simulation sub-model corresponding to the infectious state provided in an embodiment of this application.

[0118] Furthermore, based on the first preset parameters corresponding to the infectious state, the sizes or parameters of the aforementioned convolutional layers, BN layers, etc., are set to limit them, ultimately creating a disease simulation sub-model consisting of a convolutional layer (Conve) with a size of 1*3, padding of (0,1), stride of 1, and 8 channels, a BN layer, a ReLU layer, a max pooling layer with a size of 1*3, padding of (0,1), and stride of 1, and a ReLU layer. The BN layer can accelerate the model's training speed and generalization, while the ReLU layer can increase the model's non-linear expressive power.

[0119] Similarly, based on the differential equation (1) corresponding to the susceptible state (S) and the corresponding second preset parameters, a disease simulation sub-model can be generated, consisting of a convolutional layer (Conve), a BN layer, and a ReLU layer, with a kernel size of 1*3, padding of (0,1), a stride of 1, and 16 channels. For details, please refer to... Figure 5 The diagram shown is a schematic of a disease simulation sub-model corresponding to a susceptible state provided in an embodiment of this application.

[0120] For wheat plants in a latent state, a disease simulation sub-model can be generated based on differential equation (2) and the third preset parameters. This sub-model consists of a convolutional layer (Conve) with a kernel size of 1*3, padding of (0,1), stride of 1, and 16 channels, a BN layer, a ReLU layer, a 1*3 average pooling layer with padding of (0,1) and stride of 1, and a ReLU layer. For details, please refer to [reference needed]. Figure 6 The diagram shows a disease simulation sub-model corresponding to the latent state provided in an embodiment of this application. The disease simulation sub-model can simulate the growth status and disease development of wheat plants in the latent state based on the feature information model input to it. The feature information can be predictive features composed of meteorological factors and remote sensing factors, and / or feature information output by the disease simulation sub-model corresponding to the susceptible state.

[0121] Similarly, for wheat plants in a non-infected state, a disease simulation sub-model can be generated based on differential equation (4) and the fourth preset parameter. This sub-model consists of a convolutional layer with a kernel size of 1*7, padding of (0,3), stride of 1, and number of channels of 8, a BN layer, and a ReLU layer. For details, please refer to [reference needed]. Figure 7 The diagram shown is a sub-model of the disease simulation corresponding to the non-infectious state provided in an embodiment of this application. Figure 7 In the diagram, A represents the collection of convolutional layers, batch normalization (BN) layers, and ReLU layers. The arrows from left to right represent the input and output, respectively. The disease simulation model corresponding to the non-infectious state can simulate the growth status and disease development of wheat plants in the non-infectious state based on the feature information input to the disease simulation sub-model. This feature information can be predicted features composed of meteorological and remote sensing factors and / or feature information output by the disease simulation sub-model corresponding to the infectious state.

[0122] This application embodiment also requires splicing each disease simulation sub-model to obtain a single-day disease simulation sub-model capable of simulating the complete spread of wheat scab. In this application embodiment, "single day" refers to one wheat scab completing a single day's spread. Figure 3 The transmission process cycle is shown, and the single-day disease simulation sub-model can process the input prediction features to obtain the corresponding feature information. When predicting wheat scab in wheat plants, it is also necessary to consider the disease trend that wheat plants may eventually exhibit under the multi-day transmission mechanism of wheat scab, namely the disease spike rate and the trend of the disease spike rate curve.

[0123] Therefore, in this embodiment, the obtained single-day disease simulation sub-model is concatenated with the long short-term memory network sub-model. The long short-term memory network sub-model is used to re-input the feature information output by the single-day disease simulation sub-model into the single-day disease simulation sub-model. The single-day disease simulation sub-model uses the feature information as the prediction feature to start a new round of simulation and prediction processing. This process is repeated sequentially to achieve multi-day continuous simulation of wheat plants under wheat scab. In this embodiment, the long short-term memory network sub-model can use an LSTM cell module, which can be referred to... Figure 8 The structural diagram of the initial wheat scab prediction model shown is presented for your understanding.

[0124] Figure 8 The susceptible state, latent state, infectious state, and non-infectious state in the figure are disease simulation sub-models corresponding to specific states. All disease simulation sub-models are spliced ​​together to obtain the single-day disease simulation sub-model within the dashed box in the figure. The single-day disease simulation sub-model is then spliced ​​with D, where D is the LSTM cell module. In this embodiment of the application, multiple LSTM cell modules can be spliced ​​together to complete the transmission of feature information.

[0125] In addition, after satisfying the multi-day transmission of wheat scab, the LSTMcell module, i.e., the long short-term memory network sub-model, will output combined feature information. The combined feature information represents the disease status of wheat plants in the last single-day transmission process during the multi-day transmission process, as well as the features that are related to the occurrence of wheat scab in the last single-day transmission process. Processing the combined features can obtain the DI value corresponding to the wheat plants in the last single-day transmission process.

[0126] Similarly, if it is necessary to obtain the DI curve for the multi-day propagation process, the combined feature information obtained from each single-day propagation can be processed after each single-day propagation. Finally, the combined feature information corresponding to each single-day propagation in the multi-day propagation process can be processed to obtain the DI value curve, so that observers can check the development trend of wheat scab on wheat plants.

[0127] In this embodiment of the application, the DI value is obtained by processing the combined feature information. This can be achieved by processing the feature information through a fully connected layer (Linear layer) and an activation function layer (Sigmoid layer). Figure 8 As shown, a Linear layer (E) and a Sigmoid layer (F) are concatenated, and the data output from the LSTM cell module is transmitted to E and F for processing to obtain the DI value curve, thereby realizing the prediction of wheat scab using the wheat scab prediction model.

[0128] Among them, the Linear layer (E) is the fully connected layer. In this embodiment, the fully connected layer is mainly used to establish the mapping relationship between the features related to wheat scab and the disease rate in meteorological factors and remote sensing factors. Based on the mapping relationship, the characteristic information of wheat plants or the disease rate can be analyzed. The parameters of the fully connected layer are composed of weights and biases. Each neuron in the fully connected layer is connected to all neurons in the previous layer. The calculation formula of the fully connected layer can be as follows: formula (6):

[0129] Y h =W h *Y h-1 +b h (6)

[0130] Where h represents the number of layers, Y h-1 For the output of layer h-1, W h Let b be the weight of the h-th layer. h For the h-th layer bias, Y h This is the output of the h-th layer.

[0131] Furthermore, the purpose of using ReLU and Sigmoid layers in the above model creation process is to improve the nonlinearity of the wheat scab prediction model and to map the model output feature information to the same range as DI, so as to obtain data or feature information that can characterize the DI value of wheat plants. In the embodiments of this application, activation function layers are used to implement the mapping processing of output feature information. Specifically, Sigmoid activation function and ReLU activation function can be used, and the specific formula can be referred to in the following formula:

[0132]

[0133]

[0134] Where x represents the feature information or combined feature information input into the ReLU layer or Sigmoid layer.

[0135] Step S230: Train the initial wheat scab prediction model based on the training dataset and the validation dataset to obtain the target wheat scab prediction model.

[0136] The optimal hyperparameter values ​​of the initial wheat scab prediction model are obtained by performing calculations based on the validation set. Furthermore, meteorological and remote sensing factors from the training dataset are input as feature values ​​into the initial wheat scab prediction model, and the diseased ear rate corresponding to the meteorological and remote sensing factors is used as the target value. This target value is compared and calculated with the predicted diseased ear rate to update the model parameters of the initial wheat scab prediction model.

[0137] In this embodiment, the process of training and optimizing the initial wheat scab prediction model may include: initializing the hyperparameters of the initial wheat scab prediction model based on the validation dataset, the hyperparameters including the learning rate; inputting the meteorological factors and remote sensing factors of the wheat plant samples in the training dataset into the initial wheat scab prediction model for prediction processing to obtain the predicted diseased ear rate; determining a loss function based on the predicted diseased ear rate and the diseased ear rate in the training dataset; determining a cost function based on the loss function; determining new model parameters based on the cost function, the new model parameters including weight values ​​and bias values; and performing the training and optimization process based on the new model parameters. The steps of determining the loss function based on the disease rate in the training dataset are as follows: If the new loss function satisfies the convergence state, the new model parameters corresponding to the new loss function are determined as the target model parameters of the initial wheat scab prediction model, and a target wheat scab prediction model is determined based on the target model parameters; if the new loss function does not satisfy the convergence state, the initial wheat scab prediction model is updated with the new model parameters, and the steps of inputting the meteorological factors and remote sensing factors of the wheat plant samples in the training dataset into the initial wheat scab dynamic prediction model based on the data batch value are executed to obtain the predicted disease rate, until the new loss function satisfies the convergence state.

[0138] The hyperparameters of the initial wheat scab prediction model may include an initial learning rate η = 0.01, a batch size of 500, an epoch of 2000 iterations, a number of layers in the long short-term memory network sub-model (number_layers = 1), a hidden size of 4, and a step size of 15. In this embodiment, the hyperparameter that needs to be optimized for the initial wheat scab prediction model is mainly the learning rate, which needs to be further optimized based on the set number of iterations and the iterative algorithm.

[0139] In this embodiment, the Adam algorithm can be selected as the iterative algorithm. Meteorological factors and remote sensing factors from the validation dataset are input into the initial wheat scab prediction model, and the learning rate is adjusted using the aforementioned set number of iterations as the termination condition. Specifically, the preset learning rate decreases in a stepwise manner; for example, the learning rate is adjusted to η*0.6 every 250 iterations, and after 1000 iterations, it is adjusted to η*0.6 every 1000 iterations. Based on the above iterative algorithm for the learning rate, after reaching the preset number of iterations, the obtained learning rate is used as the optimal learning rate for the initial wheat scab prediction model, and the hyperparameters are updated.

[0140] Based on this, the initial wheat scab prediction model can be trained by first inputting meteorological factors and remote sensing factors from the training dataset into the initial wheat scab prediction model according to the data batch size set above. The model parameters are then updated through forward and backward propagation. The model parameters may include the weights and biases of the convolutional layers, the weights and biases of the fully connected layers, and the weights and biases of the LSTM cell layers.

[0141] Specifically, the initial wheat scab prediction model can predict the corresponding DI value based on the input meteorological factors and remote sensing factors through forward propagation of data. However, the current initial wheat scab prediction model has not yet updated its model parameters, and the predicted DI value obtained has an error compared with the actual DI value in the training dataset. In this embodiment, the mean squared error loss function (MSELOSS) is selected to calculate the error. The calculation formula (9) of MSELOSS is as follows:

[0142]

[0143] Where W represents the weight value of the initial wheat scab prediction model, b represents the bias value of the initial wheat scab prediction model, and y represents the actual DI value. The value represents the DI value obtained through forward propagation, i.e., the predicted DI value. N represents the dimension of y. In this embodiment, N = 1.

[0144] Furthermore, based on the loss function obtained in this calculation, the cost function is determined. The formula (10) for the cost function and the formulas (11) and (12) for calculating the new model parameters based on the cost function are as follows:

[0145]

[0146]

[0147]

[0148] Where W represents the weight values ​​of the initial wheat scab prediction model, W1 represents the new weight values ​​of the initial wheat scab prediction model to be updated, b represents the bias value of the initial wheat scab prediction model, b1 represents the new bias value of the initial wheat scab prediction model to be updated, M is the batch size value, and η is the learning rate. To determine the sign of the partial derivative.

[0149] Based on the above calculations, after this forward propagation and backpropagation, new model parameters, namely W1 and b1, are obtained. To determine whether the new model parameters can enable the initial Fusarium head blight prediction model to reach the optimal prediction state, in the embodiments of this application, it is determined by whether the loss function corresponding to the new model parameters converges. Specifically, substituting the new model parameters W1 and b1 into formula (10) to obtain a new loss function corresponding to the new model parameters. Further, it is determined whether the new loss function satisfies the convergence state. If it satisfies the convergence state, it is proved that the new model parameters are the best model parameters that can be obtained from the current training dataset. The new model parameters are determined as the target model parameters for the Fusarium head blight prediction model, and the initial Fusarium head blight prediction model is adjusted accordingly to obtain the target Fusarium head blight prediction model.

[0150] However, if the new loss function does not satisfy the convergence state, it is necessary to continue to input the data of the training dataset for training. Before re-inputting the data of the training dataset for prediction, it is also necessary to adjust the model parameters of the initial Fusarium head blight prediction model to the new model parameters obtained from this calculation, and use the new model parameters as the model parameters for the next calculation. Subsequently, perform the step of inputting the meteorological factors and the remote sensing factors of the wheat plant samples in the training dataset into the initial Fusarium head blight prediction model for prediction processing to obtain the predicted diseased spike rate and subsequent steps until the loss function corresponding to the new model parameters satisfies the convergence state, and further determine the target model parameters.

[0151] Step S240: Test the target Fusarium head blight prediction model based on the test dataset to obtain a corresponding table of diseased spike rate levels.

[0152] Specifically, the meteorological factors and the remote sensing factors in the test dataset can be input into the target Fusarium head blight prediction model to obtain the to-be-tested diseased spike rate; based on the diseased spike rate and the to-be-tested diseased spike rate in the test dataset, determine the corresponding table of diseased spike rate levels for the target Fusarium head blight prediction model. The corresponding table of diseased spike rate levels represents the degree of disease corresponding to the diseased spike rate.

[0153] The to-be-tested diseased spike rate within a preset time range is divided by levels. For example, the numerical range of the to-be-tested diseased spike rate is between 0 and 1, and within the range of 0 to 1, it is divided by levels. Such as the initial stage of disease: DI value ≤ 0.3, the middle stage of disease 2: 0.3 < DI value < 0.6, the late stage of disease 3: DI value > 0.6. Based on this, during actual prediction, the observer can determine the current degree of disease of the wheat plant and the possible degree of development of the Fusarium head blight of the current wheat plant according to the diseased spike rate of the wheat plant to be predicted and the above corresponding table of diseased spike rate levels.

[0154] In addition, the target wheat scab prediction model can be evaluated based on the test dataset. In this embodiment, after obtaining the test disease rate corresponding to the test dataset, the difference between the test disease rate and the actual disease rate is calculated using the coefficient of determination R2 (R-squared) and the root mean square error (RMSE) to evaluate the prediction accuracy of the target wheat scab prediction model.

[0155] Based on the above training, the target wheat scab disease is predicted for the wheat plants to be predicted. Specifically, the prediction process may include: generating prediction features based on the meteorological factors and the remote sensing factors; inputting the prediction features into the daily disease simulation sub-model; processing the prediction features based on preset processing conditions and four disease simulation sub-models to obtain feature information corresponding to each of the four sub-models; splicing the four feature information to obtain combined feature information and recording the number of splicing operations; if the number of splicing operations is less than a threshold, inputting the combined feature information as a prediction feature into the daily disease simulation model, and executing the step of processing the prediction features based on preset processing conditions and four sub-models to obtain feature information corresponding to each of the four sub-models, until the number of splicing operations is not less than the threshold; if the number of splicing operations is not less than the threshold, mapping the combined features to obtain the disease spike rate corresponding to the wheat plant to be predicted.

[0156] based on Figure 8 As shown in the schematic diagram of the initial wheat scab prediction model, meteorological and remote sensing factors can first generate a 1*1*6 prediction feature, which is then input into the daily disease simulation sub-model for simulation, completing one daily simulation. In the daily disease simulation sub-model, based on the disease simulation sub-models corresponding to four specific states, the occurrence of wheat scab in wheat plants between the four specific states needs to be simulated respectively, ultimately obtaining the feature information of this daily simulation. Specifically, the preset processing conditions and processes between each disease simulation sub-model may include:

[0157] Based on the disease simulation sub-model corresponding to the infectious state, the predicted features are processed to obtain initial infectious state feature information; the initial infectious state feature information is concatenated with the predicted features to obtain the first input feature;

[0158] The first input feature is processed by inputting it into the disease simulation sub-model corresponding to the susceptible state to obtain susceptible state feature information; the susceptible state feature information and the predicted feature are concatenated to obtain the second input feature; the second input feature is processed by inputting it into the disease simulation sub-model corresponding to the latent state to obtain latent state feature information.

[0159] The latent state feature information and the predicted feature are concatenated to obtain the third input feature; the third input feature is input into the disease simulation sub-model corresponding to the infectious state for processing to obtain the infectious state feature information;

[0160] The infectious status feature information and the predicted feature are concatenated to obtain the fourth input feature; the fourth input feature is input into the disease simulation sub-model corresponding to the non-infectious status for processing to obtain the non-infectious status feature information;

[0161] The susceptible state feature information, the latent state feature information, the infectious state feature information, and the non-infectious state feature information are determined as feature information corresponding to the four disease simulation sub-models, respectively.

[0162] In this embodiment of the application, the predicted features are input into a disease simulation sub-model corresponding to the infectious state for processing to obtain the feature information of wheat plants in the initial state of infectious state. According to Figure 4 , Figure 5 , Figure 6 , Figure 7 As shown in the disease simulation sub-model structures corresponding to the infectious state, susceptible state, latent state, and non-infectious state, after the disease simulation sub-model corresponding to the infectious state processes the 1*1*6 predicted features, the disease simulation sub-model can output an 8*1*6 feature information, which is the initial infectious state feature information. The feature information is the information related to the occurrence of wheat scab extracted by the convolutional layer in the disease simulation sub-model after the predicted features are processed.

[0163] Furthermore, based on the transmission relationship between various specific states of wheat scab, the feature information calculated from the previous transmission relationship is concatenated with the predicted features to obtain an input feature. This input feature is then input into the disease simulation sub-model corresponding to the next transmission relationship for processing to obtain the feature information of the corresponding specific state.

[0164] Based on this, it can be seen that the initial infectious state feature information and the predicted feature are concatenated to obtain a first input feature of size 9*1*6. The first input feature is then input into the disease simulation model corresponding to the susceptible state to simulate and obtain a feature information of size 16*1*6, which is the susceptible state feature information.

[0165] The susceptible state feature information is concatenated with the predicted feature to obtain a second input feature of size 17*1*6, which is then input into the disease simulation sub-model corresponding to the latent state to obtain a latent state feature information of size 16*1*6. Similarly, the latent state feature information is concatenated with the predicted feature to obtain a third input feature of size 17*1*6, which is then input into the disease simulation model corresponding to the infectious state to obtain an infectious state feature information of size 8*1*6. Finally, the infectious state feature information is concatenated with the predicted feature to obtain a fourth input feature of size 9*1*6, which is then input into the disease simulation sub-model corresponding to the non-infectious state to obtain a non-infectious state feature information of size 8*1*6.

[0166] Based on this, simulations of various specific states of wheat plants were completed, and characteristic information corresponding to each specific loading was obtained. The four characteristic information pieces were then spliced ​​together. In this embodiment, the number of splicing operations is recorded to determine which single-day simulation is being performed. Assuming that only the disease rate of wheat plants within 15 days needs to be predicted, a preset threshold of 15 can be set. When the number of splicing operations meets the threshold, it proves that 15 single-day simulations have been performed, and the prediction of wheat plants can be terminated.

[0167] After the splicing is completed, a combined feature information is obtained. The combined feature information may include information that can characterize the disease status of wheat plants during this single-day simulation, as well as features that are related to wheat scab during this single-day simulation. If the number of splicing is within a preset threshold, the combined feature information is transmitted to the single-day disease simulation sub-model through the long short-term memory network sub-model as a prediction feature for the next single-day simulation, and the next single-day simulation is started until the number of splicing is not less than the threshold.

[0168] When the number of splicing operations meets the threshold, the combined feature information can no longer be input into the daily simulation sub-model. Instead, the combined feature information can be mapped based on the fully connected layer and the activation function layer to obtain the corresponding DI value.

[0169] Next, the wheat scab prediction method described in the above-mentioned application embodiments will be introduced through the following application examples. The wheat scab prediction method can also simulate and predict the wheat growth process to obtain the DI value or DI curve output by the target wheat scab prediction model. Based on this, growers can adjust the wheat planting environment or, according to the DI curve, take timely treatment measures for wheat plants at corresponding time points to avoid wheat disease and yield reduction, thus achieving scientific planting and effective prevention.

[0170] When growers plant wheat in an experimental field and need to predict the incidence of Fusarium head blight, they can collect and detect the meteorological environment of the experimental field and remote sensing indices of the wheat to obtain meteorological and remote sensing factors corresponding to the wheat. In this method, meteorological factors can include humidity (R), temperature (T), and precipitation (P), while remote sensing factors can include vegetation cover, normalized difference vegetation index (NDI), and optimized soil-regulated vegetation index (SDI). These meteorological and remote sensing factors are then input into a target wheat Fusarium head blight prediction model for processing.

[0171] In the target wheat scab prediction model, meteorological factors and remote sensing factors are combined to generate a matrix-form prediction feature. Based on this prediction feature, wheat plant growth is simulated in disease simulation sub-models corresponding to various specific states to obtain corresponding feature information and combined feature information, thus completing the daily transmission of wheat scab in wheat plants. Processing the combined feature information based on the daily transmission yields the daily DI value of the wheat plant.

[0172] To obtain further simulation results of the multi-day spread of wheat scab on wheat plants, growers can set a preset cycle, i.e., the number of times mentioned above, to ensure that the preset number of single-day spread processes are completed, and to obtain the combined characteristic information, corresponding DI values, and DI curves after multi-day spread. Based on this, growers can conduct more scientific management and maintenance of the wheat in the experimental field.

[0173] The wheat scab prediction device provided in the embodiments of this application is described below. The wheat scab prediction device described below can be referred to in correspondence with the wheat scab prediction method described above.

[0174] First, combined Figure 9 The wheat scab prediction device will be introduced, such as... Figure 10 As shown, the wheat scab prediction device may include:

[0175] The information acquisition unit 100 acquires the meteorological factors and remote sensing factors corresponding to the wheat plants to be predicted.

[0176] The information processing unit 200 is used to input the meteorological factors and the remote sensing factors into the target wheat Fusarium head blight prediction model for prediction processing, and obtain the diseased ear rate of the wheat plant to be predicted.

[0177] The target wheat scab prediction model is a model trained based on wheat plant samples. The wheat plant samples include wheat plant samples in several specific states, and each wheat plant sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors, and disease prevalence. Furthermore, there are transmission relationships between different specific states for wheat scab.

[0178] The embodiments of this application can make predictions using meteorological factors and remote sensing factors. During the prediction process, the growth process of wheat plants under specific conditions during the spread of wheat scab can be simulated, and the transmission relationship between various specific conditions of wheat plants can be simulated. Based on this, not only can the influence of meteorological factors and remote sensing factors on the disease be taken into account during the prediction process, but the epidemic mechanism of the disease itself can also be considered. By taking multiple factors into account, the accuracy of predicting wheat plants suffering from wheat scab can be improved.

[0179] Optionally, the device may also include:

[0180] The sample data acquisition unit is used to acquire wheat plant samples that match each specific state, as well as the meteorological factors, remote sensing factors and disease rate of the wheat plant samples, to obtain a sample dataset. The specific state includes susceptible state, latent state, infected state and non-infectious state.

[0181] The sample data partitioning unit is used to partition the sample dataset according to a preset partitioning ratio to obtain a training dataset, a validation dataset, and a test dataset.

[0182] The model acquisition unit is used to acquire a preset initial wheat scab prediction model;

[0183] The model training unit is used to train the initial wheat scab prediction model based on the training dataset and the validation dataset to obtain the target wheat scab prediction model. The training process uses the meteorological factors and the remote sensing factors in the training dataset as feature values ​​and the diseased ear rate corresponding to the meteorological factors and the remote sensing factors as the target value for training.

[0184] The model testing unit is used to input the meteorological factors and remote sensing factors in the test dataset into the target wheat scab prediction model to obtain the diseased ear rate to be tested.

[0185] The model level determination unit is used to determine the disease rate level correspondence table of the target wheat Fusarium head blight prediction model based on the disease rate and the disease rate to be tested in the test dataset. The disease rate level correspondence table characterizes the disease severity corresponding to the disease rate.

[0186] Optionally, the model acquisition unit may include:

[0187] The infection rate preset subunit is used to preset the infection rate corresponding to each specific state based on the transmission process of wheat scab in the wheat plant sample.

[0188] The differential equation determination subunit is used to determine the differential equation corresponding to each of the specific states based on each specific state and the infection rate, wherein the differential equation characterizes the propagation relationship between the specific states;

[0189] The disease simulation sub-model creation sub-unit is used to create a disease simulation sub-model corresponding to each specific state based on the differential equation and convolutional neural network corresponding to each specific state. The disease simulation sub-model is used to simulate the wheat plant growth process of the specific state based on meteorological factors and remote sensing meteorological factors, and the model structure of each disease simulation sub-model is different.

[0190] The first sub-unit of the sub-model splicing is used to splice the disease simulation sub-model corresponding to each of the four specific states based on the propagation relationship between them, to obtain a single-day disease simulation sub-model.

[0191] The sub-model is spliced ​​into a second sub-unit, which is used to splice the single-day disease simulation sub-model with the long short-term memory network sub-model to obtain a multi-day disease simulation sub-model. The long short-term memory network sub-model is used to input the feature information output by the single-day disease simulation sub-model into the single-day disease simulation sub-model.

[0192] The third sub-unit of the sub-model splicing is used to splice the multi-day disease simulation sub-model with the fully connected layer and the activation function layer to obtain the initial wheat scab model. The fully connected layer and the activation function layer are used to process the feature information output by the multi-day disease simulation sub-model to obtain the disease spike rate corresponding to the feature information.

[0193] Optionally, the model training unit includes:

[0194] The hyperparameter initialization subunit is used to initialize the hyperparameters of the initial wheat scab prediction model based on the validation dataset, wherein the hyperparameters include the learning rate;

[0195] The training data prediction subunit is used to input the meteorological factors and remote sensing factors of the wheat plant samples in the training dataset into the initial wheat Fusarium head blight prediction model for prediction processing to obtain the predicted disease rate;

[0196] The loss function operation subunit is used to determine the loss function based on the predicted diseased ear rate and the diseased ear rate in the training dataset;

[0197] The cost function determination sub-unit is used to determine the cost function based on the loss function;

[0198] The model parameter determination subunit is used to determine new model parameters based on the cost function, wherein the new model parameters include weight values ​​and bias values;

[0199] The second operational subunit of the loss function is used to perform the step of determining the loss function based on the predicted diseased ear rate and the diseased ear rate in the training dataset, based on the new model parameters, to obtain the new loss function;

[0200] The first judgment subunit of the loss function is used to determine the new model parameters corresponding to the new loss function as the target model parameters of the initial wheat scab prediction model if the new loss function satisfies the convergence state, and to determine the target wheat scab prediction model based on the target model parameters.

[0201] The second judgment subunit of the loss function is used to update the initial wheat scab prediction model with the new model parameters if the new loss function does not meet the convergence state, and to perform the step of inputting the meteorological factors and remote sensing factors of the wheat plant samples in the training dataset into the initial wheat scab dynamic prediction model for prediction processing to obtain the predicted diseased ear rate, until the new loss function meets the convergence state.

[0202] Optionally, the disease simulation sub-model creation sub-unit includes:

[0203] The sub-unit for creating the infectious state sub-model is used to determine the disease simulation sub-model corresponding to the infectious state based on the differential equation corresponding to the infectious state and the preset infectious state disease simulation sub-model structure. The preset infectious state disease simulation sub-model structure includes a convolution kernel, a convolution layer, a pooling layer, a batch normalization processing layer, and an activation function layer generated based on a first preset parameter.

[0204] A sub-unit for creating a susceptible state sub-model is used to determine the disease simulation sub-model corresponding to the susceptible state based on the differential equation corresponding to the susceptible state and the disease simulation sub-model structure of the susceptible state. The disease simulation sub-model structure of the susceptible state includes a convolutional kernel, a convolutional layer, a batch normalization layer and an activation function layer generated based on a second preset parameter.

[0205] The latent state sub-model creation sub-unit is used to determine the disease simulation sub-model corresponding to the latent state based on the differential equation corresponding to the latent state and the latent state disease simulation sub-model structure. The latent state disease simulation sub-model structure includes a convolution kernel, a convolution layer, a batch normalization layer, a pooling layer, and an activation function layer generated based on a third preset parameter.

[0206] A sub-model for creating a non-infectious state sub-model is used to determine the disease simulation sub-model corresponding to the non-infectious state based on the differential equation corresponding to the non-infectious state and the structure of the disease simulation sub-model for the non-infectious state. The structure of the disease simulation sub-model for the non-infectious state includes a convolutional kernel, a convolutional layer, a batch normalization layer, and an activation function layer generated based on a fourth preset parameter.

[0207] Optionally, the target wheat scab prediction model includes a daily disease simulation sub-model and a long short-term memory network sub-model. The daily disease simulation sub-model includes disease simulation sub-models corresponding to four specific states, namely, susceptible state, latent state, infectious state, and non-infectious state.

[0208] The information processing unit 200 includes:

[0209] A prediction feature generation subunit is used to generate prediction features based on the meteorological factors and the remote sensing factors;

[0210] A feature input subunit is used to input the predicted features into the single-day disease simulation sub-model;

[0211] The feature processing subunit is used to process the predicted features based on preset processing conditions and the four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models respectively.

[0212] The feature splicing subunit is used to splice the four feature information to obtain combined feature information and record the number of splicing processes.

[0213] The first judgment subunit is used to input the combined feature information as a prediction feature into the single-day disease simulation model if the number of splicing processes is less than the number threshold, and to execute the step of processing the prediction feature based on preset processing conditions and four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models respectively, until the number of splicing processes is not less than the number threshold.

[0214] The second judgment subunit is used to perform mapping processing on the combined features if the number of splicing processes is not less than the number threshold, so as to obtain the diseased ear rate corresponding to the wheat plant to be predicted.

[0215] Optionally, the feature processing subunit may include:

[0216] The predictive feature processing subunit is used to process the predictive features based on the disease simulation sub-model corresponding to the infectious state to obtain the initial infectious state feature information.

[0217] The first information splicing subunit is used to splice the initial infection state feature information with the predicted feature to obtain the first input feature;

[0218] The first feature processing subunit is used to input the first input feature into the disease simulation sub-model corresponding to the susceptible state for processing, so as to obtain susceptible state feature information;

[0219] The second information splicing subunit is used to splice the easily infectable state feature information and the predicted feature to obtain the second input feature;

[0220] The second feature processing subunit is used to input the second input feature into the disease simulation sub-model corresponding to the latent state for processing to obtain latent state feature information;

[0221] The third information splicing subunit is used to splice the latent state feature information and the predicted feature to obtain the third input feature;

[0222] The third feature processing subunit is used to input the third input feature into the disease simulation sub-model corresponding to the infectious state for processing, so as to obtain infectious state feature information;

[0223] The fourth information splicing subunit is used to splice the infection status feature information and the prediction feature to obtain the fourth input feature;

[0224] The fourth feature processing subunit is used to input the fourth input feature into the disease simulation sub-model corresponding to the non-infectious state for processing, and obtain the non-infectious state feature information.

[0225] The feature information determination subunit is used to determine the susceptible state feature information, the latent state feature information, the infectious state feature information, and the non-infectious state feature information as feature information corresponding to the four disease simulation sub-models, respectively.

[0226] The wheat scab prediction device provided in this application embodiment can be applied to wheat scab prediction equipment.

[0227] Figure 10 A schematic diagram of a wheat scab prediction device is shown. (Refer to...) Figure 10 The structure of the wheat scab prediction device may include: at least one processor 10, at least one memory 20, and at least one communication bus 30 and at least one communication interface 40;

[0228] In this embodiment, the number of processor 10, memory 20, communication bus 30, and communication interface 40 is at least one, and processor 10, memory 20, and communication interface 40 communicate with each other through communication bus 30.

[0229] The processor 10 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0230] The memory 20 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0231] The memory stores a program, which the processor can call to implement the processing flow of each of the aforementioned wheat scab prediction methods.

[0232] This application embodiment also provides a storage medium that can store a program suitable for execution by a processor, the program being used to implement the processing flow of each of the aforementioned wheat scab prediction methods.

[0233] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0234] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0235] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method of predicting scab in wheat, characterized by, include: Obtain the meteorological and remote sensing factors corresponding to the wheat plants to be predicted; The meteorological factors and the remote sensing factors are input into the target wheat Fusarium head blight prediction model for prediction processing to obtain the diseased ear rate of the wheat plant to be predicted. The target wheat scab prediction model is a model trained based on wheat plant samples. These wheat plant samples include samples in several specific states, and each sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors, and disease incidence rate. Furthermore, there are transmission relationships for wheat scab among these specific states. The target wheat scab prediction model includes a daily disease simulation sub-model and a long short-term memory network sub-model. The daily disease simulation sub-model includes disease simulation sub-models corresponding to four specific states, namely, susceptible state, latent state, infectious state, and non-infectious state. The step of inputting the meteorological factors and the remote sensing factors into the target wheat Fusarium head blight prediction model for prediction processing to obtain the disease incidence rate of the wheat plant to be predicted includes: Based on the meteorological factors and the remote sensing factors, predictive features are generated; The predicted features are input into the single-day disease simulation sub-model; Based on preset processing conditions and the four disease simulation sub-models, the predicted features are processed to obtain feature information corresponding to the four disease simulation sub-models respectively. The four feature information pieces are concatenated to obtain combined feature information, and the number of concatenation processes is recorded. If the number of splicing processes is less than the number threshold, the combined feature information is used as a prediction feature and input into the single-day disease simulation sub-model. The step of processing the prediction feature based on preset processing conditions and four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models is performed until the number of splicing processes is not less than the number threshold. If the number of splicing processes is not less than the threshold number, the combined features are mapped to obtain the diseased ear rate corresponding to the wheat plant to be predicted.

2. The method according to claim 1, characterized in that, Also includes: Wheat plant samples matching each specific state, along with corresponding meteorological factors, remote sensing factors, and disease rate, are obtained to create a sample dataset. The specific states include susceptible state, latent state, infected state, and non-infectious state. The sample dataset is divided according to a preset division ratio to obtain a training dataset, a validation dataset, and a test dataset. Obtain the preset initial wheat scab prediction model; The initial wheat scab prediction model is trained based on the training dataset and the validation dataset to obtain the target wheat scab prediction model. The training process uses the meteorological factors and the remote sensing factors in the training dataset as feature values ​​and the diseased ear rate corresponding to the meteorological factors and the remote sensing factors as the target value for training. The meteorological factors and remote sensing factors in the test dataset are input into the target wheat Fusarium head blight prediction model to obtain the diseased ear rate to be tested; Based on the diseased ear rate and the diseased ear rate to be tested in the test dataset, a diseased ear rate level correspondence table for the target wheat Fusarium head blight prediction model is determined, and the diseased ear rate level correspondence table characterizes the disease severity corresponding to the diseased ear rate.

3. The method according to claim 2, characterized in that, The process of obtaining the preset initial wheat scab prediction model includes: Based on the transmission process of wheat scab in the wheat plant samples, the infection rate corresponding to each specific state is preset; Based on each specific state and the infection rate, a differential equation corresponding to each specific state is determined, and the differential equation characterizes the propagation relationship between the specific states; Based on the differential equation and convolutional neural network corresponding to each specific state, disease simulation sub-models are created respectively corresponding to each specific state. The disease simulation sub-models are used to simulate the wheat plant growth process in the specific state based on meteorological factors and remote sensing meteorological factors, and the model structure of each disease simulation sub-model is different. Based on the propagation relationship between the four specific states, the disease simulation sub-models corresponding to each specific state are spliced ​​together to obtain a single-day disease simulation sub-model. The single-day disease simulation sub-model is concatenated with the long short-term memory network sub-model to obtain a multi-day disease simulation sub-model. The long short-term memory network sub-model is used to input the feature information output by the single-day disease simulation sub-model into the single-day disease simulation sub-model. The multi-day disease simulation sub-model is concatenated with a fully connected layer and an activation function layer to obtain an initial wheat scab model. The fully connected layer and the activation function layer are used to process the feature information output by the multi-day disease simulation sub-model to obtain the disease spike rate corresponding to the feature information.

4. The method according to claim 2, characterized in that, The process of training the initial wheat scab prediction model based on the training dataset and the validation dataset to obtain the target wheat scab prediction model includes: Based on the validation dataset, the hyperparameters of the initial wheat scab prediction model are initialized, including the learning rate; The meteorological factors and remote sensing factors of the wheat plant samples in the training dataset are input into the initial wheat Fusarium head blight prediction model for prediction processing to obtain the predicted disease rate; Based on the predicted diseased ear rate and the diseased ear rate in the training dataset, a loss function is determined; Based on the loss function, determine the cost function; New model parameters are determined based on the cost function, and the new model parameters include weight values ​​and bias values. Based on the new model parameters, the step of determining the loss function based on the predicted diseased ear rate and the diseased ear rate in the training dataset is performed to obtain the new loss function; If the new loss function satisfies the convergence state, the new model parameters corresponding to the new loss function are determined as the target model parameters of the initial wheat scab prediction model, and the target wheat scab prediction model is determined based on the target model parameters; If the new loss function does not satisfy the convergence state, the initial wheat scab prediction model is updated with the new model parameters, and the meteorological factors and remote sensing factors of the wheat plant samples in the training dataset are input into the initial wheat scab prediction model for prediction processing to obtain the predicted diseased ear rate, until the new loss function satisfies the convergence state.

5. The method according to claim 3, characterized in that, The process of creating disease simulation sub-models corresponding to each specific state based on the differential equation and convolutional neural network for each specific state includes: Based on the differential equation corresponding to the infectious state and the preset infectious state disease simulation sub-model structure, the disease simulation sub-model corresponding to the infectious state is determined. The preset infectious state disease simulation sub-model structure includes a convolutional kernel, a convolutional layer, a pooling layer, a batch normalization processing layer, and an activation function layer generated based on the first preset parameters. Based on the differential equation corresponding to the susceptible state and the disease simulation sub-model structure of the susceptible state, the disease simulation sub-model corresponding to the susceptible state is determined. The disease simulation sub-model structure of the susceptible state includes a convolutional kernel, a convolutional layer, a batch normalization processing layer and an activation function layer generated based on the second preset parameters. Based on the differential equation corresponding to the latent state and the latent state disease simulation sub-model structure, the disease simulation sub-model corresponding to the latent state is determined. The latent state disease simulation sub-model structure includes a convolutional kernel, a convolutional layer, a batch normalization layer, a pooling layer, and an activation function layer generated based on a third preset parameter. Based on the differential equation corresponding to the non-infectious state and the structure of the disease simulation sub-model for the non-infectious state, the disease simulation sub-model corresponding to the non-infectious state is determined. The structure of the disease simulation sub-model for the non-infectious state includes a convolutional kernel, a convolutional layer, a batch normalization layer, and an activation function layer generated based on a fourth preset parameter.

6. The method according to claim 1, characterized in that, Based on preset processing conditions and the four disease simulation sub-models, the predicted features are processed to obtain feature information corresponding to the four disease simulation sub-models, including: Based on the disease simulation sub-model corresponding to the infectious state, the predicted features are processed to obtain the initial infectious state feature information; The initial infection state feature information is concatenated with the predicted feature to obtain the first input feature; The first input feature is input into the disease simulation sub-model corresponding to the susceptible state for processing to obtain susceptible state feature information; The easily infectable state feature information and the predicted feature are concatenated to obtain the second input feature; The second input feature is input into the disease simulation sub-model corresponding to the latent state for processing to obtain latent state feature information; The latent state feature information and the predicted feature are concatenated to obtain the third input feature; The third input feature is input into the disease simulation sub-model corresponding to the infectious state for processing to obtain infectious state feature information; The infection status feature information and the prediction feature are concatenated to obtain the fourth input feature; The fourth input feature is input into the disease simulation sub-model corresponding to the non-infectious state for processing to obtain the non-infectious state feature information; The susceptible state feature information, the latent state feature information, the infectious state feature information, and the non-infectious state feature information are determined as feature information corresponding to the four disease simulation sub-models, respectively.

7. A wheat scab prediction device, characterized in that, include: The information acquisition unit acquires the meteorological factors and remote sensing factors corresponding to the wheat plants to be predicted. The information processing unit is used to input the meteorological factors and the remote sensing factors into the target wheat Fusarium head blight prediction model for prediction processing, and obtain the diseased ear rate corresponding to the wheat plant to be predicted. The target wheat scab prediction model is a model trained based on wheat plant samples. The wheat plant samples include wheat plant samples in several specific states, and each wheat plant sample in a specific state is labeled with corresponding meteorological factors, remote sensing factors and disease rate; and there is a transmission relationship between the various specific states for wheat scab. The target wheat scab prediction model includes a daily disease simulation sub-model and a long short-term memory network sub-model. The daily disease simulation sub-model includes disease simulation sub-models corresponding to four specific states, namely, susceptible state, latent state, infectious state, and non-infectious state. The information processing unit includes: A prediction feature generation subunit is used to generate prediction features based on the meteorological factors and the remote sensing factors; A feature input subunit is used to input the predicted features into the single-day disease simulation sub-model; The feature processing subunit is used to process the predicted features based on preset processing conditions and the four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models respectively. The feature splicing subunit is used to splice the four feature information to obtain combined feature information and record the number of splicing processes. The first judgment subunit is used to input the combined feature information as a prediction feature into the single-day disease simulation sub-model if the number of splicing processes is less than the number threshold, and to execute the step of processing the prediction feature based on preset processing conditions and four disease simulation sub-models to obtain feature information corresponding to the four disease simulation sub-models respectively, until the number of splicing processes is not less than the number threshold. The second judgment subunit is used to perform mapping processing on the combined features if the number of splicing processes is not less than the number threshold, so as to obtain the diseased ear rate corresponding to the wheat plant to be predicted.

8. A wheat scab prediction device, characterized in that, Including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the steps of the wheat scab prediction method as described in any one of claims 1-6.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the wheat scab prediction method as described in any one of claims 1-6.

Images