A wheat scab prediction method based on multi-source data and ES-GBM algorithm

Abstract

The application discloses a wheat scab prediction method based on multi-source data and an ES-GBM algorithm, and comprises the following steps: S1, collecting field wheat scab disease grade sample point data of a target area; S2, acquiring multi-source remote sensing data of the target area, wherein the multi-source remote sensing data at least include multiple characteristic variables in atmospheric parameters, weather parameters, vegetation spectrum parameters and atmospheric environment parameters; S3, using the sample point data and sample point attribute data extracted from the multi-source remote sensing data, constructing a prediction model based on an evolutionary strategy ES optimizing a gradient boosting machine GBM, wherein the ES is used for automatically searching an optimal hyperparameter combination of the GBM; and S4, inputting gridded multi-source remote sensing data of the target area into the prediction model to obtain a wheat scab disease degree spatial distribution map of the target area. The application can fuse multi-source data, and through the evolutionary strategy, an optimal hyperparameter combination is searched, so that the prediction precision of the model is improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural disease prediction technology, specifically to a method for predicting wheat scab based on multi-source data and the ES-GBM algorithm. Background Technology

[0002] Fusarium head blight is a major disease affecting wheat throughout its entire growth cycle in my country, primarily occurring in major wheat-producing areas such as the middle and lower reaches of the Yangtze River, the Jianghuai region, the Huanghuai region, and southern North China. It not only causes significant yield reduction and quality decline, but the toxins produced by infected wheat grains also threaten human and animal health. Driven by a combination of factors including global warming, changes in farming systems, and the southward shift of wheat-growing areas, the distribution range, affected area, and severity of this disease are continuously expanding.

[0003] Traditional methods for monitoring wheat scab mainly rely on field surveys conducted by agricultural researchers to obtain dynamic information on the disease's occurrence. These traditional methods are time-consuming, labor-intensive, lack timeliness, and are highly susceptible to human error. In recent years, the emergence and development of remote sensing and machine algorithms have gradually overcome the challenges of predicting wheat scab.

[0004] Previously, it was mainly believed that the occurrence and development of wheat scab were related to meteorological factors, such as temperature, humidity, precipitation, and sunshine, without considering the close relationship between atmospheric parameters and the occurrence of wheat scab. For example, Chinese patent CN111768051B discloses a dynamic early warning method and device for wheat scab, which predicts wheat scab incidence data by collecting meteorological characteristics corresponding to the wheat heading and flowering stages. However, predicting scab based solely on meteorological and spectral factors has certain limitations, mainly due to the unique and unstable meteorological conditions in different wheat fields, resulting in low prediction accuracy. Furthermore, a visualized spatial distribution map of wheat scab severity in raster image form has not been obtained.

[0005] Accurate prediction and control of wheat scab are crucial for ensuring my country's food security. Therefore, to address the shortcomings of existing technologies, there is an urgent need to develop a prediction method based on multi-source data and algorithms. Multi-source data fusion can integrate atmospheric parameters, meteorological parameters, spectral data, and field sampling information, and the model can automatically search for the optimal hyperparameter combination, improving the model's prediction accuracy and stability. Summary of the Invention

[0006] To overcome the problems in the existing technology, the purpose of this invention is to provide a wheat scab prediction method based on multi-source data and the ES-GBM algorithm, which can integrate multi-source data and search for the optimal hyperparameter combination through an evolutionary strategy to improve the prediction accuracy of the model.

[0007] To achieve the above objectives, this invention provides a method for predicting wheat scab based on multi-source data and the ES-GBM algorithm, comprising the following steps: S1: Collect field data on wheat scab disease severity at sampling points in the target area; S2: Acquire multi-source remote sensing data of the target area, wherein the multi-source remote sensing data includes at least multiple characteristic variables among atmospheric parameters, weather parameters, vegetation spectral parameters, and atmospheric environmental parameters; S3: Using the sample data and attribute values ​​from the multi-source remote sensing data, construct a prediction model based on the evolutionary strategy ES optimized gradient booster GBM, wherein the ES is used to automatically search for the optimal hyperparameter combination of the GBM; S4: Input the attribute values ​​of the rasterized multi-source remote sensing data of the target area into the prediction model to obtain a spatial distribution map of the severity of wheat scab in the target area.

[0008] Furthermore, the field wheat scab disease level sampling data mentioned in step S1 includes disease-free sampling points and sampling points with different disease severity levels; after preprocessing the sampling data to remove missing values ​​and outliers, the wheat scab disease severity is divided into four levels: disease-free, mild disease, moderate disease, and severe disease.

[0009] Furthermore, the multiple characteristic variables of the multi-source remote sensing data mentioned in step S2 include: air temperature, aerosol optical thickness, precipitation, enhanced vegetation index, red band reflectance, blue band reflectance, normalized difference vegetation index, near-infrared band reflectance, and PM2.5. 2.5 concentration.

[0010] Furthermore, the construction steps of the ES-optimized GBM prediction model in step S3 include: S31: Define the GBM algorithm function, set the severity of wheat scab disease as the dependent variable, set the multiple feature variables as independent variables, and set the distribution characteristics of the dependent variable, the number of decision trees, the depth of the decision trees, the learning rate, and the minimum number of samples per node in the function. S32: Define an ES function that receives hyperparameters of GBM as optimization objects. The hyperparameters include at least the decision tree depth, the number of decision trees, the learning rate, and the minimum number of samples per node. S33: Randomly generate the initial parent population, and randomly generate the initial Q set of parameters within the preset parameter value range; S34: Calculate the fitness of each set of parameters in the parent population to evaluate the model prediction error corresponding to each set of parameters; S35: Calculate variable asynchronous length based on the current algebra; S36: Introduce Gaussian noise, randomly select individuals from the parent generation and apply random perturbation to generate the offspring population; S37: Calculate the fitness of each parameter in the offspring population; S38: Sort the parent and offspring individuals according to fitness, and retain the top Q groups of parameters with the best fitness as the next generation of parent population; S39: Repeat steps S34 to S38 until the preset termination condition is met, and output the parameter combination with the best fitness as the optimal hyperparameter combination of the GBM. S310: Construct the final GBM prediction model using the optimal hyperparameter combination.

[0011] Furthermore, in step S33, the generation of the j-th parameter X of the i-th individual in the initial parent population... i,j The formula is expressed as: ; Where: X i,j This represents the specific value of the j-th parameter of the i-th individual, where i ranges from 1 to Q and j ranges from 1 to 4; rand(0,1) generates a random decimal number that is uniformly distributed between 0 and 1; B max,j and B min,j These are the maximum and minimum values ​​allowed for the j-th parameter, respectively, and Bmax,j–Bmin,j is the allowable range of variation for the j-th parameter. In step S34, the function used to calculate the fitness is the root mean square error (RMSE), which is expressed by the formula: ; Among them: E i This represents the root mean square error value corresponding to the i-th group of parameters; N represents the actual number of samples measured; This represents the w-th wheat scab disease level as measured in reality; This represents the w-th wheat scab disease level predicted by the model.

[0012] Furthermore, the formula for the variable asynchronous length coefficient M in step S35 is expressed as follows:

[0013] Where: M is the variable asynchronous length coefficient of the t-th generation; This is the initial step size scaling factor; is the decay rate factor; exp represents an exponential function with the natural base e; t is the current generation; T is the preset maximum total number of generations.

[0014] Furthermore, the offspring parameters mentioned in step S36 From parent parameters After adding Gaussian noise and subjecting it to boundary constraints, the following is the process: ; ; in: This represents the j-th parameter of the generated offspring; This represents the j-th parameter of the selected parent generation; M is the variable asynchronous length coefficient of the current generation; Z represents random noise that follows a standard normal distribution; and These are the maximum and minimum values ​​allowed for the j-th parameter, respectively; `max` and `min` are the maximum and minimum value operations, respectively, used to limit parameters that exceed the boundaries to within the preset boundaries.

[0015] Furthermore, step S38 specifically involves: sorting the parent and offspring individuals according to their root mean square error from smallest to largest, and retaining a preset number of individuals at the top of the sort as the next generation of parent population; the preset termination condition is that the number of iterations reaches a preset maximum number of generations.

[0016] Furthermore, after the termination condition is met, the parameters corresponding to the individual with the smallest root mean square error in the final population are taken as the optimal hyperparameter combination, denoted as . ;based on A GBM model was constructed, and the accuracy of the model was evaluated using the 10-fold cross-validation method. The coefficient of determination was calculated. The The calculation formula is: ; Where: N represents the actual number of samples measured; This represents the kth measured level of wheat scab disease. This represents the kth wheat scab disease level predicted by the model. m is the average value of all measured wheat scab disease severity data.

[0017] Further, step S4 includes: inputting the attribute values ​​of multi-source remote sensing data of the target area into the prediction model, and the prediction model outputting the prediction result of the severity of wheat scab disease; Obtain land classification data for the target area, extract rasters representing crop distribution areas, and extract rasters with a normalized vegetation index greater than a preset threshold; the intersection of the rasters representing crop distribution areas and the rasters with a normalized vegetation index greater than the preset threshold is taken as the wheat planting area of ​​the target area. The predicted results of wheat scab disease severity output by the prediction model are processed with the wheat planting area grid. The predicted values ​​for non-wheat planting areas are set to zero, and only the disease severity distribution in the wheat planting area is retained. Finally, a spatial distribution map of wheat scab disease severity in the target area is generated.

[0018] This invention integrates atmospheric parameters to comprehensively characterize the environmental features affecting the occurrence and development of wheat scab from multiple dimensions, improving the comprehensiveness of prediction information. At the same time, it combines evolutionary strategies with gradient boosting machines to construct an automatic hyperparameter optimization model, which searches for the optimal hyperparameter combination in a complex parameter space. After constructing the prediction model, it uses a screening method to extract the true wheat planting area range, and finally generates a spatial distribution map of the severity of scab in the target area, which intuitively displays the disease risk level in different areas, providing a visual basis for relevant agricultural departments to make corresponding control measures. Attached Figure Description

[0019] Figure 1 This is a flowchart of the wheat scab prediction based on multi-source data and the ES-GBM algorithm of the present invention; Figure 2 This refers to the severity of wheat scab disease at the sampling points in this invention and the attribute values ​​of the corresponding remote sensing data; Figure 3 The target area is remote sensing data in the form of a raster image, as described in this embodiment of the invention. Figure 4 This invention uses the ES hyperparameter optimization trajectory based on minimum error and optimal fit. Figure 5 The prediction results of the wheat scab prediction model based on the ES-GBM algorithm in this invention are shown. Figure 6 The importance ranking of predictors in the ES-GBM prediction model of this invention; Figure 7 This is a spatial distribution map of the severity of wheat scab disease according to the present invention. Detailed Implementation

[0020] This embodiment provides a method for predicting wheat scab based on multi-source data and the ES-GBM algorithm, such as... Figure 1 As shown, it includes the following steps: S1: Collect field data on wheat scab disease severity at sampling points in the target area.

[0021] The field wheat scab disease level sampling data includes disease-free sampling points and sampling points with different disease severity levels. After preprocessing the sampling data to remove missing and outlier values, the wheat scab disease severity is divided into four levels: disease-free, mild disease, moderate disease, and severe disease.

[0022] Specifically, representative areas were selected in each sampling field, and multiple 1m×1m quadrats were established. The number of diseased ears was recorded in each quadrat, and the disease rate (number of diseased ears / total number of ears × 100%) was calculated. The precise geographic coordinates of each sampling point were recorded using a handheld GPS device, and field photos were taken for archiving.

[0023] The collected raw data is preprocessed to remove sample data with abnormal GPS signals, incomplete records, or obvious human error, and missing values ​​are marked and excluded.

[0024] Based on the national standard "Technical Specification for Monitoring and Forecasting of Wheat Fusarium Head Blight" (GB / T15796-2011) and combined with actual field performance, the following grading standards are used to classify the severity of the disease into four levels: Grade 0 (Disease-free ears): All effective ears per plant are healthy, with full grains and normal thousand-grain weight; Level 1: Disease incidence rate is less than 10%, and the disease is characterized by scattered diseased ears in the field, without forming disease centers; Grade 2: The incidence of diseased ears is between 10% and 20%, the lesions extend to the middle of the spikelets, and a pink mold layer is visible. In the field, diseased ears are distributed in patches. Level 3: Disease incidence rate is between 20% and 30%, lesions cover more than one-third of the ear length, the mold layer is obvious and accompanied by yellowish-brown gum, the grains are deformed, and contiguous disease areas are formed in the field. Grade 4: Disease incidence rate is greater than 30%, lesions extend to more than two-thirds of the entire ear, the mold layer is thick and dense, and the kernels are moldy and rotten.

[0025] Among them, level 0 is classified as no disease, level 1 and level 2 are classified as mild disease, level 3 is classified as moderate disease, and level 4 is classified as severe disease.

[0026] S2: Obtain multi-source remote sensing data of the target area, wherein the multi-source remote sensing data includes at least multiple characteristic variables among atmospheric parameters, weather parameters, vegetation spectral parameters, and atmospheric environmental parameters.

[0027] In this embodiment, remote sensing data of the target area in various raster image formats were downloaded from the NASA Earth Data website. Specifically, the data included the following nine characteristic variables: air temperature (degrees Celsius), aerosol optical depth (AOD), precipitation (mm / day), enhanced vegetation index (EVI), red reflectance, blue reflectance, normalized difference vegetation index (NDVI), near-infrared reflectance (NIR), and PM2.5. 2.5 Concentration (unit: kg / m²). Wheat disease severity data for each sampling point and the corresponding remote sensing data attribute values ​​are as follows: Figure 2 As shown, the remote sensing data of the target area in the form of a raster image is as follows: Figure 3 As shown.

[0028] S3: Using the attribute values ​​of the sample data and the multi-source remote sensing data, construct a prediction model based on an evolutionary strategy (ES) optimized gradient booster (GBM), wherein the ES is used to automatically search for the optimal hyperparameter combination of the GBM.

[0029] The steps in step S3 for constructing the prediction model based on the evolutionary strategy (ES) optimized gradient booster (GBM) include: S31: Define the GBM algorithm function, setting the severity of wheat scab as the dependent variable. Include the following parameters for each sample point: air temperature, aerosol optical thickness, precipitation, EVI, red band reflectance, blue band reflectance, NDVI, near-infrared band reflectance, and PM2.5. 2.5 The attribute values ​​of the nine feature variables are set as independent variables. The gbm() function is called inside the function, and the distribution feature of the dependent variable is set to multinomial. All valid samples are used as the original dataset and randomly divided into 10 subsets for 10-fold cross-validation. In each round, 9 subsets are used as the training set and the remaining 1 subset is used as the validation set.

[0030] S32: Define an ES function that receives the hyperparameters of GBM as the optimization target. The hyperparameters include at least the decision tree depth, the number of decision trees, the learning rate, and the minimum number of samples per node.

[0031] S33: Randomly generate the initial parent population. Randomly generate the initial Q set of parameters within the preset parameter value range. In this embodiment, Q is preferably 5.

[0032] The j-th parameter X is generated using the i-th individual in the initial parent population. i,j The formula is expressed as: ; Among them, X i,j This represents the specific value of the j-th parameter of the i-th individual, where i ranges from 1 to Q and j ranges from 1 to 4; rand(0,1) generates a random decimal number that is uniformly distributed between 0 and 1; B max,j and B min,j These are the maximum and minimum allowed values ​​for the j-th parameter, respectively, and Bmax,j–Bmin,j is the allowed range of variation for the j-th parameter.

[0033] S34: Calculate the fitness of each set of parameters in the parent population to evaluate the model prediction error corresponding to each set of parameters.

[0034] Substitute the five sets of parameters into the GBM function, and calculate the fitness using the root mean square error (RMSE). The calculation formula is as follows: ; Among them: E i This represents the root mean square error value corresponding to the i-th group of parameters; N represents the actual number of samples measured. This represents the w-th wheat scab disease level as measured in reality; This represents the w-th wheat scab disease level predicted by the model.

[0035] E i The smaller the value, the smaller the model prediction error and the higher the fitness of the corresponding set of parameters.

[0036] In order to achieve a smooth transition from the initial large-scale coarse search to the later small-scale fine-tuning, while taking into account both global optimization capability and local convergence accuracy, this application further introduces a variable asynchronous long step S35.

[0037] S35: Calculate variable asynchronous length based on the current algebra; In this embodiment, the maximum number of generations T is set to 10. For the t-th generation (t=0,1,2,…,9), the variable generation length coefficient M is calculated according to the following formula:

[0038] in: M is the variable-asynchronous length coefficient of the t-th generation, used to control the magnitude of parameter variation. This is the initial step size scaling factor; This is the decay rate factor; exp represents an exponential function with base e; t is the current generation; T is the preset maximum total number of generations, 10.

[0039] When t=0, M=0.3, the initial step size is limited by the scaling factor to avoid the initial search being too divergent; as t gradually increases, M decays rapidly according to an exponential law to achieve the transition from the initial stage to the later stage.

[0040] S36: Introduce Gaussian noise, randomly select individuals from the parent generation and apply random perturbation to generate the offspring population.

[0041] One parent individual is randomly selected from the five parent individuals, and the parent is used as the parent. Based on the parameters, Gaussian noise perturbation is added to generate offspring individuals. The specific formula is expressed as follows: ; ; in: This represents the j-th parameter of the generated offspring; This represents the j-th parameter of the offspring after the boundary clipping operation, ensuring that the offspring parameters do not go out of a reasonable range; represents the j-th parameter of the selected parent generation; M is the variable asynchronous length coefficient of the current generation; Z is random noise that conforms to a standard normal distribution; and These are the maximum and minimum values ​​allowed for the j-th parameter, respectively; max and min are the maximum and minimum value operations, used to limit parameters that exceed the boundaries to within the preset boundaries.

[0042] Repeat the above process 10 times to generate 10 offspring individuals.

[0043] S37: Calculate the fitness of each parameter in the offspring population; Substitute the 10 child parameters into the GBM function and calculate the RMSE value for each child in step S34.

[0044] S38: Sort the parent and offspring individuals according to fitness, and retain the top Q groups of parameters with the best fitness as the next generation of parent population.

[0045] In this embodiment, 5 parent individuals and 10 offspring individuals are mixed, resulting in a total of 15 individuals, which are then sorted in ascending order of their RMSE values. The top 5 individuals are retained as the next generation of parent individuals, while the bottom 10 individuals are eliminated and deleted.

[0046] S39: Repeat steps S34 to S38 until the preset termination condition is met, and output the parameter combination with the best fitness as the optimal hyperparameter combination of the GBM.

[0047] Repeat steps S34 to S38, evolving generation by generation. In this embodiment, the evolutionary process is terminated after a preset T=10 generations of iterations. The parameter combination corresponding to the individual with the smallest RMSE value in the final population is taken as the hyperparameter combination X for obtaining the optimal number of decision trees, decision tree depth, learning rate, and minimum number of samples per node in the ES search. best ,like Figure 4 As shown, the number of decision trees is n.trees=1124, the depth of the decision trees is interaction.depth=13, the learning rate is shrinkage=0.0459, and the minimum number of samples per node is n.minobsinnode=6.

[0048] S310: Construct the final GBM prediction model using the optimal hyperparameter combination.

[0049] based on A GBM model was constructed, and the model's predicted values ​​were compared with the actual observed values ​​using the 10-fold cross-validation method to calculate the coefficient of determination. To assess prediction accuracy, the The calculation formula is: ; Where N represents the actual number of samples measured; This represents the kth measured level of wheat scab disease. This represents the kth wheat scab disease level predicted by the model. This represents the average value of all measured wheat scab disease severity data.

[0050] This indicates the degree of explanation for data variation; the closer it is to 1, the higher the model's prediction accuracy.

[0051] like Figure 5 As shown, the average RMSE of cross-validation in this embodiment is 0.67, and the R² is 0.84, indicating good model prediction accuracy.

[0052] In addition, this embodiment uses the summary() function to extract the relative importance of the nine feature variables in the GBM model, such as... Figure 6 As shown in the feature importance ranking results, among all nine feature variables involved in the modeling, PM... 2.5 The relative importance of PM2.5 concentration was as high as 49.04%, ranking first, far exceeding the second-ranked enhanced vegetation index and the third-ranked air temperature. This result also confirms the importance of the atmospheric environmental parameter PM2.5. 2.5 The impact on the occurrence of wheat scab.

[0053] S4: After unifying the spatial resolution of the rasterized multi-source remote sensing data of the target area, input it into the prediction model to obtain a spatial distribution map of the severity of wheat scab in the target area. The remote sensing data in raster image format is as follows: Figure 3 As shown.

[0054] The attribute values ​​of multi-source remote sensing data of the target area are input into the prediction model, and the prediction model outputs the prediction result of the severity of wheat scab disease. In order to ensure that the prediction model displays prediction results only for areas where wheat is planted, it is further necessary to select fields that are actually planted with wheat.

[0055] Obtain land classification data for the target area, extract rasters representing crop distribution areas, and extract rasters with a normalized vegetation index greater than a preset threshold; the intersection of the rasters representing crop distribution areas and the rasters with a normalized vegetation index greater than the preset threshold is taken as the wheat planting area of ​​the target area. In this embodiment, land classification data for the target area is downloaded from the NASA Earth Data website, and grids with an attribute value of 12 are identified. These grids represent the "crop" category. However, farmland contains a variety of crops, and land classification data alone cannot distinguish wheat from other crops. Furthermore, the Normalized Difference Vegetation Index (NDVI) is an important indicator reflecting vegetation growth status and coverage. For example, in May, winter wheat in the target area is in the grain-filling to milk-ripe stage. Healthy wheat fields typically have an NDVI value higher than 0.7, and their condition differs significantly from other crops at the same time. Therefore, in this embodiment, grids with an NDVI greater than 0.708 are selected. This separates the grids that simultaneously meet the crop distribution and NDVI requirements as wheat-grown fields, thus obtaining the true wheat-growing area.

[0056] The predicted severity of wheat scab disease output from the prediction model is multiplied by the wheat-growing area using a raster multiplication operation. Non-wheat areas are multiplied by 0, and wheat areas are multiplied by 1, remaining unchanged. This resets the predicted values ​​for non-wheat-growing areas to zero, retaining only the disease severity distribution within wheat-growing areas, ultimately generating a spatial distribution map of wheat scab disease severity in Henan Province. Figure 7 As shown in the figure, light green indicates no disease, light yellow indicates mild disease, brick red indicates moderate disease, and bright purple indicates severe disease.

Claims

1. A method for predicting wheat scab based on multi-source data and the ES-GBM algorithm, characterized in that, Includes the following steps: S1: Collect field data on wheat scab disease severity at sampling points in the target area; S2: Acquire multi-source remote sensing data of the target area, wherein the multi-source remote sensing data includes at least multiple characteristic variables among atmospheric parameters, weather parameters, vegetation spectral parameters, and atmospheric environmental parameters; S3: Using the attribute values ​​of the sample data and the multi-source remote sensing data, construct a prediction model based on the evolutionary strategy ES to optimize the gradient booster machine GBM, wherein the ES is used to automatically search for the optimal hyperparameter combination of the GBM; S4: Input the attribute values ​​of the rasterized multi-source remote sensing data of the target area into the prediction model to obtain a spatial distribution map of the severity of wheat scab in the target area.

2. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 1, characterized in that, The field wheat scab disease level sampling data mentioned in step S1 includes disease-free sampling points and sampling points with different disease severity levels; after preprocessing the sampling data to remove missing values ​​and outliers, the wheat scab disease severity level is divided into four levels: disease-free, mild disease, moderate disease, and severe disease.

3. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 1, characterized in that, The multiple characteristic variables of the multi-source remote sensing data mentioned in step S2 include: air temperature, aerosol optical thickness, precipitation, enhanced vegetation index, red band reflectance, blue band reflectance, normalized difference vegetation index, near-infrared band reflectance, and PM2.

5. 2.5 concentration.

4. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 1, characterized in that, Step S3, which describes the steps for constructing a prediction model based on the evolutionary strategy ES-optimized gradient boosting machine GBM, includes: S31: Define the GBM algorithm function, set the severity of wheat scab disease as the dependent variable, set the multiple feature variables as independent variables, and set the distribution characteristics of the dependent variable, the number of decision trees, the depth of the decision trees, the learning rate, and the minimum number of samples per node in the function. S32: Define an ES function that receives hyperparameters of GBM as optimization objects. The hyperparameters include at least the decision tree depth, the number of decision trees, the learning rate, and the minimum number of samples per node. S33: Randomly generate the initial parent population, and randomly generate the initial Q set of parameters within the preset parameter value range; S34: Calculate the fitness of each set of parameters in the parent population to evaluate the model prediction error corresponding to each set of parameters; S35: Calculate variable asynchronous length based on the current algebra; S36: Introduce Gaussian noise, randomly select individuals from the parent generation and apply random perturbation to generate the offspring population; S37: Calculate the fitness of each parameter in the offspring population; S38: Sort the parent and offspring individuals according to fitness, and retain the top Q groups of parameters with the best fitness as the next generation of parent population; S39: Repeat steps S34 to S38 until the preset termination condition is met, and output the parameter combination with the best fitness as the optimal hyperparameter combination of the GBM. S310: Construct the final GBM prediction model using the optimal hyperparameter combination.

5. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 4, characterized in that, In step S33, the generation of the j-th parameter X of the i-th individual in the initial parent population... i,j The formula is expressed as: ； Where: X i,j This represents the specific value of the j-th parameter of the i-th individual, where i ranges from 1 to Q and j ranges from 1 to 4; rand(0,1) is a uniformly distributed random decimal between 0 and 1; B max,j and B min,j These are the maximum and minimum values ​​allowed for the j-th parameter, respectively, and Bmax,j–Bmin,j is the allowable range of variation for the j-th parameter. In step S34, the function used to calculate the fitness is the root mean square error (RMSE), which is expressed by the formula: ； Among them: E i This represents the root mean square error value corresponding to the i-th group of parameters; N represents the actual number of samples measured. This represents the w-th wheat scab disease level as measured in reality; This represents the w-th wheat scab disease level predicted by the model.

6. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 4, characterized in that, The formula for the variable asynchronous length coefficient M in step S35 is expressed as follows: ； Where: M is the variable asynchronous length coefficient of the t-th generation; This is the initial step size scaling factor; is the decay rate factor; exp represents an exponential function with the natural base e; t is the current generation; T is the preset maximum total number of generations.

7. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 4, characterized in that, The offspring parameters mentioned in step S36 From parent parameters After adding Gaussian noise and subjecting it to boundary constraints, the following is the process: ； ； in:: This represents the j-th parameter of the generated offspring; This represents the j-th parameter of the child generation after the boundary clipping operation; represents the j-th parameter of the selected parent generation; M is the variable asynchronous length coefficient of the current generation; Z is random noise that conforms to a standard normal distribution; and These are the maximum and minimum values ​​allowed for the j-th parameter, respectively; max and min are the maximum and minimum value operations, used to limit parameters that exceed the boundaries to within the preset boundaries.

8. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 4, characterized in that, Step S38 specifically involves: sorting the parent and offspring individuals in ascending order of root mean square error, and retaining a preset number of individuals at the top of the sort as the next generation of parent population; the preset termination condition is that the number of iterations reaches a preset maximum number of generations.

9. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 8, characterized in that, After the termination condition is met, the parameters corresponding to the individual with the smallest root mean square error in the final population are taken as the optimal hyperparameter combination, denoted as . ;based on A GBM model was constructed, and the accuracy of the model was evaluated using the 10-fold cross-validation method. The coefficient of determination was calculated. The The calculation formula is: ； Where: N represents the actual number of samples measured; This represents the kth measured level of wheat scab disease. denoted as the kth wheat scab disease level predicted by the model; m is the average value of all measured wheat scab disease severity sample data.

10. The wheat scab prediction method based on multi-source data and the ES-GBM algorithm according to claim 1, characterized in that, Step S4 includes: The attribute values ​​of multi-source remote sensing data of the target area are input into the prediction model, and the prediction model outputs the prediction result of the severity of wheat scab disease. Obtain land classification data for the target area, extract rasters representing crop distribution areas, and extract rasters with a normalized vegetation index greater than a preset threshold; the intersection of the rasters representing crop distribution areas and the rasters with a normalized vegetation index greater than the preset threshold is taken as the wheat planting area of ​​the target area. The predicted results of wheat scab disease severity output by the prediction model are multiplied with the wheat planting area using a raster multiplication operation. The predicted values ​​for non-wheat planting areas are set to zero, and only the disease severity distribution within the wheat planting area is retained. Finally, a spatial distribution map of wheat scab disease severity in the target area is generated.

Images