Construction and application of regional scale wheat continuous rain spike sprouting mildew identification model

Abstract

The invention discloses construction and application of a regional scale wheat continuous cloudy and rainy ear germination mildew identification model, and aims to solve the technical problem that a method for accurately monitoring and quickly identifying wheat ear germination mildew conditions is lacked at present. According to the method, meteorological and multi-source satellite remote sensing data are comprehensively applied, model factors are constructed, a ReliefF feature selection method is utilized to screen key factors, and finally, an LDA classifier is adopted to construct a pre-harvest germination mildew recognition model, so that accurate recognition of pre-harvest germination mildew of wheat during continuous rainy days on a regional scale is realized, and the accuracy of wheat pre-harvest germination mildew recognition is improved. And powerful technical support is provided for wheat continuous rain loss reduction and post-disaster evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of crop disaster estimation, and specifically to the construction and application of a regional scale wheat ear sprouting and mildew identification model. Background Art

[0002] As one of the most important food crops in the world, the stability of wheat yield and quality is of great significance to ensuring global food security. The Huanghuaihai region is the main wheat producing area in my country, and the stability of its wheat production is directly related to the overall national food security. In recent years, climate change has led to frequent extreme weather events, which has brought unprecedented challenges to wheat production. Especially during the wheat maturity period, continuous rainy weather not only affects the normal maturity and harvest of wheat, but also easily leads to mildew and ear germination, which seriously affects the yield and quality of wheat, causing wheat to lose its value as food or feed; in addition, the mycotoxins that may be produced during the germination and mildew of wheat ears will also pose a threat to human health. Therefore, accurate monitoring and rapid identification of wheat ear germination and mildew are crucial for wheat yield and quality regulation and management.

[0003] The information disclosed in this background technology section is only used to deepen the understanding of the background technology of the present disclosure, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention

[0004] In view of at least one of the above technical problems, the present disclosure provides a regional-scale wheat ear sprouting and mildew identification model construction and application, aiming to solve the current technical problem of lack of methods for accurately monitoring and quickly identifying wheat ear sprouting and mildew.

[0005] According to a first aspect of the present disclosure, a method for constructing a regional scale wheat ear sprouting and mildew identification model under continuous rainy weather is provided, comprising the following steps: (1) Collecting meteorological data during the continuous rainy period in the area to be identified and preprocessing the meteorological data, wherein the meteorological data includes daily rainfall, sunshine hours, daily average relative humidity, and average wind speed; the preprocessing includes accumulating the daily rainfall and sunshine hours to obtain the corresponding cumulative rainfall during the continuous rainy period. P total Total sunshine hours S total , average the daily average relative humidity and daily average wind speed respectively to obtain the corresponding average humidity of the continuous rainy process U avg and process average wind speed F avg ; (2) Determine the continuous rainy weather index using the following formula C To express the intensity of continuous rain: C = P total / S total , In the formula, P total The accumulated rainfall for continuous rainy process; S total Cumulative sunshine hours for continuous rainy periods; C The larger the value, the greater the risk of disasters caused by continuous rain, and vice versa. (3) Performing Kriging spatial interpolation processing on the preprocessed meteorological data to obtain the corresponding meteorological factor raster data of the area to be identified; (4) Collect satellite optical remote sensing data during the continuous rainy period, 5 days before the continuous rainy period, the day after the continuous rainy period ends, and the same day or adjacent days in the past 5 years in the area to be identified, and perform preprocessing; the preprocessing includes: accumulating the SSI products of the geostationary satellite every 15 minutes during the continuous rainy process to obtain the SSI of the continuous rainy process; calculating the normalized vegetation index for the visible light-near infrared satellite remote sensing data of the 5 days before the continuous rainy period, and performing maximum value synthesis to characterize the maturity of wheat NDVI1; calculating the vegetation index for the visible light-shortwave infrared satellite remote sensing data of the day after the continuous rainy period ends, and obtaining NDVI2, NDVI3, normalized difference infrared index NDII, ratio index RI, atmospheric impedance vegetation index ARVI, structure-insensitive pigment index SIPI, and anthocyanin reflectance index 2 ARI2 to characterize the health status of wheat; calculating the Euclidean distance of the similarity of the visible light bands of the day after the continuous rainy period ends and the same day or adjacent days in the past 5 years. ED , used to characterize the change in remote sensing images after continuous rain compared with the normal year in the same period; (5) The continuous rain remote sensing index VCRI is determined by the following formula: V VCRI = M NDVI1 × I SSI × W NDII , In the formula, M NDVI1 It is NDVI1 before continuous rain; I SSI It is the total solar radiation received by vegetation during continuous rainy periods; W NDII is the NDII after continuous rain; VCRI indicates the response of vegetation to continuous rain; (6) Use remote sensing crop classification technology to extract the wheat planting distribution raster data of the area to be identified, and use it as the raster base map; (7) Randomly select plots in the area to be graded, take samples and record the number of sprouted and moldy stems, and after threshing, record the number of grains per ear, the number of moldy grains, and the number of sprouted grains per ear to obtain sample data on the sprouting and moldy disaster of wheat ears; (8) Construction of wheat ear germination and mildew index EGMI: E EGMI =( R Gstem · R Gear + R Mstem · R Mear )×50, In the formula, R Gstem is the ratio of stems to spikes; R Gear is the germination ratio of grains on the ear; R Mstem is the ratio of stems with ear mold; R Mear is the ratio of mildewed grains on the ear; 50 is the weight of germination and mildewed ratio; E EGMI The value range of is 0 to 100, which indicates the germination and mildew rate of wheat ears in the smallest space unit monitored; (9) using the wheat ear sprouting and mildew disaster sample data, calculating the ratio of the number of stems sprouting and mildewing and the ratio of the number of grains sprouting and mildewing on the ear, and calculating the wheat ear sprouting and mildew index EGMI; Set EGMII≤5.0 as no disaster, EGMI>5.0 as ear sprout mildew disaster, and code them respectively; (10) Use the local outlier factor algorithm to handle sample outliers; (11) Based on the two types of sample data obtained above, the ReliefF feature selection method was used to predetermine the weight threshold conditions for selection, and the key meteorological and remote sensing factors for identifying wheat ear sprout mildew were screened; the expression formula of the ReliefF feature selection method is: , Where: d ( X i , NearHit) is a sample X i and its similar neighboring samples in the feature A j The distance on d ( X i ,NearHit( c ))sampleX i and its heterogeneous neighbor samples in the feature A j The distance on m The number of iterations, P ( c ) is the category c The prior probability of (12) Normalize the selected key meteorological and remote sensing factors; (13) Based on the key factors screened out by the ReliefF feature selection method, a wheat ear sprout and mildew recognition model was constructed using the LDA classifier model. The wheat disaster sample data was used to train the recognition model and determine its hyperparameters.

[0006] In some embodiments of the present disclosure, in step (12), the normalization process is performed using the following formula: , Where: f i are the values ​​of each factor, f max and f min They are the maximum and minimum values ​​of each factor in the regional wheat pixels.

[0007] In some embodiments of the present disclosure, in step (11), the weight threshold is set to 0.01.

[0008] In some embodiments of the present disclosure, in step (4), the similarity Euclidean distance is calculated using the following formula: ED : , In the formula, k is the reflectivity band number, p is the number of bands, i This is a remote sensing image after the continuous rainy process ended. j It is a reference image of the same period in a normal year. x is the reflectivity of the wheat pixel.

[0009] In some embodiments of the present disclosure, in step (4), the spectral range of visible light-near infrared in the first five days of continuous rain is 0.4-0.9 μm; the spectral range of visible light-short-wave infrared in the day after the continuous rain ends is 0.4-2.5 μm.

[0010] According to the second aspect of the present disclosure, a method for remote sensing identification of wheat ear sprouting and mildew during continuous rainy weather at a regional scale is provided. Based on the above-mentioned model, the meteorological factors and remote sensing factors of the minimum spatial pixels in the region extracted from the wheat distribution grid base map of the area to be identified are used to realize the identification of wheat ear sprouting and mildew at a regional scale.

[0011] One or more technical solutions provided in the embodiments of the present application have at least any of the following technical effects or advantages: 1. The technical solution of this application comprehensively applies meteorological and multi-source satellite remote sensing data to achieve accurate monitoring and identification of wheat ear sprouting and mildew. By accurately identifying the disaster situation, farmers and relevant departments can quickly take measures to rush to harvest and minimize the yield losses caused by ear sprouting and mildew. It is of great significance for the accurate implementation of rush harvesting, disaster relief and disaster loss reduction. Based on the technical solution of this application, the sprouting and mildew of wheat ears can be discovered at the first time, so that farmers and agricultural departments can respond quickly, and resources such as harvesting machinery and drying equipment can be deployed more scientifically and reasonably, improving disaster relief efficiency and avoiding waste of resources. It is also helpful to scientifically assess disaster losses, provide a basis for post-disaster compensation and insurance claims, and protect the legitimate rights and interests of farmers.

[0012] 2. Based on the results of disaster identification by the technical solution of this application, the causes and patterns of disaster occurrence can be analyzed, the agricultural climate resource conditions of wheat production risks can be further explored, farmers can be guided to adjust the planting structure, optimize cultivation management measures, and improve the disaster resistance of wheat. It can promote technological innovation in disaster monitoring, early warning and response in the agricultural field, enhance the scientific and technological content and intelligent level of agricultural production, and further promote the sustainable development of agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The ranking and weights of factors screened by the ReliefF method in one embodiment of the present application.

[0014] Figure 2 This is the result of identifying sprouting and mildew of wheat ears due to continuous rain in the western Huanghuai region from May 24 to 31, 2023 in one embodiment of the present application. DETAILED DESCRIPTION

[0015] In order to better understand the technical solution of the present application, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0016] Example 1: Acquisition and preprocessing of meteorological data and remote sensing data during continuous rainy season for wheat crops 1. Data Acquisition Meteorological data comes from the meteorological big data cloud platform "Tianqing" of the National Meteorological Information Center (http: / / idata.cma / cmadaas / ), including daily rainfall, sunshine hours, average relative humidity, and average wind speed during continuous rainy periods.

[0017] Satellite remote sensing data include the surface solar irradiance (SSI) of the geostationary meteorological satellite Fengyun-4A every 15 minutes during the continuous rainy period, with a spatial resolution of 4 km, from the National Satellite Meteorological Center. Polar-orbiting satellite data include the Moderate Resolution Imaging Spectroradiometer (MODIS) data with a resolution of 250 m before the continuous rainy period, and the Visible Infrared Imaging Radiometer Suite (VIIRS) data of the Suomi NPP satellite with a resolution of 375 m and 750 m on the day after the continuous rainy period ended and on the same day or adjacent dates in the past five years, from the Henan Provincial Meteorological Bureau.

[0018] 2. Data Preprocessing (1) Meteorological data processing In order to characterize the continuous rainy process, the daily rainfall and sunshine hours in the area to be identified are accumulated respectively to obtain the cumulative rainfall during the continuous rainy period. P total Cumulative sunshine hours during continuous rainy periods S total , average the daily average relative humidity and average wind speed to get the average humidity during continuous rainy periods U avg Average wind speed during continuous rainy periods F avg , and introduce continuous rainy weather index C , the calculation formula is: C= P total / S total , In the formula, P total is the rainfall during continuous rainy periods (mm); S total is the sunshine duration during continuous rainy periods (h), C The larger it is, the greater the risk of disasters caused by continuous rain, and vice versa.

[0019] All meteorological elements were interpolated to 1 km × 1 km grids using the Bayesian Kriging method. After processing, the meteorological factor variables shown in Table 1 were obtained for further factor analysis and modeling.

[0020] Table 1 Meteorological factor variables for wheat ear germination and mildew identification .

[0021] (2) Remote sensing data processing The surface solar radiation (SSI) of the continuous rain process in the identified area is accumulated to obtain the continuous rain process SSI, which represents the total solar radiation that the surface can receive during the continuous rain. Since the wheat disaster caused by continuous rain is closely related to the maturity, and the Normalized Difference Vegetation Index (NDVI) of wheat gradually decreases from the filling to the maturity stage, the MODIS data (spectral range of 0.4~0.9μm) of the five days before the continuous rain are used to calculate the Normalized Vegetation Index 1 (NDVI1), and the maximum value is synthesized to characterize the maturity of wheat.

[0022] In order to explore the spectral and remote sensing parameter characteristics of wheat after continuous rain, the vegetation index was calculated based on the VIIRS data (spectral range of 0.4-2.5 μm) on the day after the continuous rain ended (Table 2), and NDVI2, NDVI3, Normalize Difference Infrared Index (NDII), Ratio Index (RI), Atmospherically Resistant Vegetation Index (ARVI), Structure Insensitive Pigment Index (SIPI), Anthocyanin Reflectance Index 2 (ARI2) were obtained; In normal years, the characteristics of remote sensing images of wheat after maturity are not much different from year to year. In order to characterize the difference in wheat images caused by continuous rain, the Euclidean distance of the similarity of the visible light band of VIIRS data on the day after the continuous rain ended and the same day or adjacent days in the past five years was calculated. ED ,Right now: , In the formula, k is the reflectivity band number, p is the number of bands, i This is a remote sensing image after the continuous rainy process ended. j It is a reference image of the same period in a normal year. x is the reflectance of the wheat pixel. In this example, ED Calculated from the reflectivity band of VIIRS 375m resolution, where: iThis is the data from June 1st. j It is the average of the same or adjacent dates over the five years from 2018 to 2022.

[0023] The Vegetation Cloudy-Rainy Index (VCRI) is introduced, and the calculation formula is: V VCRI = M NDVI1 × I SSI × W NDII In the formula, M NDVI1 It is NDVI1 before continuous rain; I SSI is the total solar radiation received by vegetation during continuous rainy periods, W / m 2 ; W NDII It is the NDII after continuous rain; V VCRI It can indicate the response degree of vegetation to continuous rain. The smaller the value, the greater the response to continuous rain, and vice versa.

[0024] After processing multi-source remote sensing data before, during and after the continuous rainy period, the remote sensing factors in Table 2 were obtained to express the maturity, low light stress and disaster status of wheat.

[0025] Table 2 Remote sensing factors for wheat ear germination and mildew .

[0026] (3) Sample data on wheat ear sprouting and mildew disaster Disaster sample collection method: Randomly select wheat plots and record the longitude and latitude coordinates of the plots. Select 2 samples from each plot. Take 10 consecutive stems of wheat in a row for each sample, record the number of sprouted and moldy stems, repeat twice to calculate the average; take 5 ears of wheat from each sample, record the number of grains per ear, the number of moldy grains, and the number of sprouted grains after threshing, and calculate the average. The number and distribution of disaster samples and non-disaster samples collected must meet the basic requirements of statistics.

[0027] (4) Regional wheat spatial distribution raster data Using remote sensing crop classification technology, the wheat planting distribution raster data of the area to be identified is extracted and used as the basic base map.

[0028] Example 2: Screening of key factors for identifying wheat ear germination and mildew (1) Wheat ear germination and mildew index and indicators The data of the disaster survey of the plots were used to calculate the ratio of stems with germination and mildew and the ratio of grains on the ears with germination and mildew. Since it was found in the survey that wheat ear germination and mildew often occurred at the same time, and there was rarely a single disaster, the two were not distinguished, and the Ear Germination and Moldiness Index (EGMI) was uniformly constructed. The calculation method is: E EGMI =( R Gstem · R Gear + R Mstem · R Mear )×50, In the formula, R Gstem is the ratio of stems to spikes; R Gear is the germination ratio of grains on the ear; R Mstem is the ratio of stems with ear mold; R Mear is the ratio of mildewed grains on the ear; 50 is the weight of the germination and mildew ratio. E EGMI The value range is 0 to 100, which represents the germination and mildew rate of wheat ears in the smallest monitored spatial unit.

[0029] In order to avoid the situation where remote sensing is difficult to monitor when the proportion of wheat ear germination and mildew is low, the index of no disaster is set as EGMII ≤5.0, the index of ear sprouting and mildew is EGMI >5.0 and are coded as 0 and 1, respectively.

[0030] (2) Sample outlier processing method EGMI is calculated for wheat samples obtained from the survey, and local outlier factor (LOF) algorithm is used for detection according to spatial distribution to remove significantly abnormal samples. The number and distribution of disaster samples should meet the general requirements of statistics.

[0031] (3) Key factor screening Based on the sample data of the above two types of disasters and those without disasters, the weight threshold conditions for selection were predetermined (the weight threshold was set to 0.01, and factors greater than the weight threshold were selected as the selected factors). The ReliefF feature selection method was used to construct the key factors for wheat ear sprout and mildew identification from the perspectives of meteorology and remote sensing. Figure 1). ReliefF assumes that the importance of the predictor variable to the target variable is determined by the sum of the relevant statistical components corresponding to each predictor in the predictor set. By randomly selecting samples in the high-dimensional space of the predictor variable X i , and calculate the ability of each predictor of the sample to distinguish between similar samples to determine its importance. Similar samples come from neighbors, including neighbors of the same type and neighbors of different types (called Nearhit and NearMiss), and their feature distance weights are calculated respectively, as follows: , Where: d ( X i , NearHit) is a sample X i and its similar neighboring samples in the feature A j The distance on d ( X i ,NearHit( c )) is a sample X i and its heterogeneous neighbor samples in the feature A j The distance on m is the number of iterations, P ( c ) is the category c The prior probability of the classification is finally determined through iteration. The essence of the iteration process is to increase or decrease the weight. When a certain feature factor can well distinguish samples of different categories, the feature weight increases, otherwise it decreases. This method is suitable for multi-classification problems, can process high-dimensional feature factors, and has good robustness to data noise and incompleteness.

[0032] In order to eliminate the differences in magnitude and dimension of different factors, the selected key factors for modeling are normalized as follows: , In the formula, f i are the values ​​of each factor, f max and f min They are the maximum and minimum values ​​of each factor in the regional wheat pixels.

[0033] Taking no disaster and disaster (0 and 1) as the target variables for prediction classification, the five-fold cross validation of the ReliefF method was used to screen the factor variables. The weight threshold was set to 0.01, and the selected factors greater than 0.01 were as follows: Stotal , U avg 、NDII、 C 、M11、SIPI、 P total , and this group of factors is represented as RFF (ReliefF Feature Factors).

[0034] Example 3: Construction of wheat ear sprout and mildew identification model and performance evaluation during continuous rainy period Linear Discriminant Analysis (LDA) is a classic classifier based on Bayesian decision theory and has a linear decision plane. LDA assumes that data follows a Gaussian distribution and that data of different categories have the same covariance matrix but different means. It achieves classification by finding a projection direction that makes the covariance of data points of the same category as small as possible and the mean of data points of different categories as large as possible after projection. The principle and technical details of LDA are as follows: (1) The linear discriminant function is: y = g (x) =w T x That is, the linear weighting of each component of the sample vector x; the vector dot product of the sample vector x and the weight vector w; if || w || = 1, it is regarded as the projection of vector x on vector w.

[0035] The basic principle of LDA criterion is to find the most suitable projection axis so that the distance between the projections of the two types of samples on the axis is as far as possible; the projection of each type of sample is as compact as possible. The function of the mean and dispersion of the projected data is used as the judgment criterion.

[0036] (2) Calculate the representation of the sample distribution in d-dimensional space The mean vector of two types of samples (0 and 1, i.e., no disaster and disaster) m i : , Sample class dispersion matrix Si and the total intra-class dispersion matrix S w : , , , Sample class dispersion matrix S b : , .

[0037] (3) Calculate the one-dimensional projection space ( Y space) to characterize the sample distribution The mean vector of two types of samples (0 and 1, i.e., no disaster and disaster) m i : , Sample intra-class dispersion matrix and total intra-class dispersion matrix S w : , , Sample class dispersion matrix : .

[0038] (4) Establish the relationship between sample X and projection space y: .

[0039] (5) Establish the criterion function of LDA: .

[0040] (6) Use the Lagrange multiplier algorithm to solve the constrained optimization problem w * : .

[0041] Based on the key meteorological and remote sensing factors screened in Example 2, an LDA classifier was used to construct a wheat ear sprouting and mildew recognition model, which was denoted as an RFF-LDA model.

[0042] Two-thirds of the survey samples were used to construct the above model, and one-third of the samples were used to evaluate the model performance. The accuracy indicators are the overall accuracy (AC) of the recognition model, the Kappa coefficient, and the precision (Precision), recall (Recall) and F1 score of the disaster-free and disaster-affected areas. The model performance was evaluated, and the results are shown in Table 3.

[0043] Table 3 Performance evaluation of RFF-LDA model .

[0044] Example 4: Remote sensing identification of ear sprouting and mildew during continuous rainy periods in the western Huanghuai wheat region This example takes a continuous rain disaster in the western Huanghuai wheat area from May 24 to 31, 2023 as an example, and identifies the sprouting and mildew of wheat ears based on the data and preprocessing method obtained in Example 1, the key factor screening method in Example 2, and the RFF-LDA model constructed in Example 3.

[0045] 1. Data collection and preprocessing. The continuous rainy process of the meteorological stations in the application area, that is, the daily rainfall, sunshine hours, average relative humidity, and average wind speed daily meteorological data from May 24 to 31, 2023, are used to calculate the factor variables in Table 1, and Bayesian kriging spatial interpolation is performed to obtain the raster data of each meteorological factor in the application area. The satellite remote sensing data include the surface solar radiation product (SSI) every 15 minutes during the continuous rainy period (May 24-31) of the geostationary meteorological satellite Fengyun-4A, with a spatial resolution of 4 km, and the cumulative SSI is calculated. The high-quality polar-orbiting satellite data collected include the Moderate Resolution Imaging Spectroradiometer (MODIS) data with a resolution of 250 m before the continuous rainy season (May 19-24) and the Visible Near-Infrared Imaging Radiometer (VIIRS) data of the Suomi NPP satellite with a resolution of 375 m and 750 m on the day after the continuous rainy season ends (June 1) and on the same day or adjacent dates from 2018 to 2022. The remote sensing factors in Table 2 are calculated.

[0046] 2. Identification results of regional wheat ear sprouting and mildew The meteorological factors and remote sensing factors of the minimum spatial pixel of wheat in the region were extracted using the raster base map extracted by remote sensing of wheat planting distribution in the Huanghuaihai region. The RFF-LDA model constructed in Example 3 was used to identify the sprouting and mildew of wheat ears in the region. The identification results are as follows: Figure 2 And as shown in Table 4.

[0047] Table 4 Statistics of the area of ​​wheat ears sprouting and mildewing during continuous rainy weather .

[0048] Although some preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0049] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the inventive concept. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for constructing a regional scale wheat ear sprouting and mildew identification model during continuous rainy weather, characterized in that: The steps include: (1) Collecting and preprocessing meteorological data during the continuous rainy period in the area to be identified, wherein the meteorological data includes daily rainfall, sunshine hours, daily average relative humidity, and average wind speed; The preprocessing includes accumulating the daily rainfall and sunshine hours to obtain the corresponding continuous rainy process accumulated rainfall P total Total sunshine hours S total The daily average relative humidity and daily average wind speed are averaged to obtain the corresponding average humidity of the continuous rainy process. U avg and process average wind speed F avg ; (2) Determine the continuous rainy weather index using the following formula C To express the intensity of continuous rain: C = P total / S total ， Where: P total The accumulated rainfall for continuous rainy process; S total Cumulative sunshine hours for continuous rainy periods; (3) Performing Kriging spatial interpolation processing on the preprocessed meteorological data to obtain the corresponding meteorological factor raster data of the area to be identified; (4) Collect satellite optical remote sensing data during the continuous rainy period, 5 days before the continuous rainy period, the day after the continuous rainy period ends, and the same day or adjacent days in the past 5 years in the area to be identified, and perform preprocessing; the preprocessing includes: accumulating the SSI products of the geostationary satellite every 15 minutes during the continuous rainy process to obtain the SSI of the continuous rainy process; calculating the normalized vegetation index for the visible light-near infrared satellite remote sensing data of the 5 days before the continuous rainy period, and performing maximum value synthesis to characterize the maturity of wheat NDVI1; calculating the vegetation index for the visible light-shortwave infrared satellite remote sensing data of the day after the continuous rainy period ends, and obtaining NDVI2, NDVI3, normalized difference infrared index NDII, ratio index RI, atmospheric impedance vegetation index ARVI, structure-insensitive pigment index SIPI, and anthocyanin reflectance index 2 ARI2 to characterize the health status of wheat; calculating the Euclidean distance of the similarity of the visible light bands of the day after the continuous rainy period ends and the same day or adjacent days in the past 5 years. ED , used to characterize the change in remote sensing images after continuous rain compared with the normal year in the same period; (5) The continuous rain remote sensing index VCRI is determined by the following formula: V VCRI = M NDVI1 × I SSI × W NDII ， Where: M NDVI1 It is NDVI1 before continuous rain; I SSI It is the total solar radiation received by vegetation during continuous rainy periods; W NDII is the NDII after continuous rain; VCRI indicates the response of vegetation to continuous rain; (6) Use remote sensing crop classification technology to extract the wheat planting distribution raster data of the area to be identified, and use it as the raster base map; (7) Randomly select plots in the area to be identified, take samples and record the number of sprouted and moldy stems, and after threshing, record the number of grains per ear, the number of moldy grains, and the number of sprouted grains per ear to obtain sample data on the sprouting and moldy disaster of wheat ears; (8) Construction of wheat ear germination and mildew index EGMI: E EGMI =（ R Gstem · R Gear + R Mstem · R Mear ）×50 ， Where: R Gstem is the ratio of stems to spikes; R Gear is the germination ratio of grains on the ear; R Mstem is the ratio of stems with ear mold; R Mear is the ratio of mildewed grains on the ear; 50 is the weight of germination and mildewed ratio; E EGMI The value range of is 0 to 100, which indicates the germination and mildew rate of wheat ears in the smallest space unit monitored; (9) using the wheat ear sprouting and mildew disaster sample data, calculating the ratio of the number of stems sprouting and mildewing and the ratio of the number of grains sprouting and mildewing on the ear, and calculating the wheat ear sprouting and mildew index EGMI; Set EGMII≤5.0 as no disaster, EGMI>5.0 as ear sprout mildew disaster, and code them respectively; (10) Use the local outlier factor algorithm to handle sample outliers; (11) Based on the two types of sample data obtained in step (9), the ReliefF feature selection method is used to predetermine the weight threshold conditions for selection, and the key meteorological and remote sensing factors for identifying wheat ear sprout mildew are screened; the expression formula of the ReliefF feature selection method is: ， Where: d ( X i , NearHit) is a sample X i and its similar neighboring samples in the feature A j The distance on d ( X i , NearHit( c ))sample X i and its heterogeneous neighbor samples in the feature A j The distance on m The number of iterations, P ( c ) is the category c The prior probability of (12) Normalize the selected key meteorological and remote sensing factors; (13) Based on the key factors screened out by the ReliefF feature selection method, a wheat ear sprout and mildew recognition model was constructed using the LDA classifier model. The wheat disaster sample data was used to train the recognition model and determine its hyperparameters.

2. The construction method according to claim 1, characterized in that: In step (12), the normalization process is performed using the following formula: ， Where: f i are the values ​​of each factor, f max and f min They are the maximum and minimum values ​​of each factor in the regional wheat pixels.

3. The construction method according to claim 1, characterized in that: In step (11), the weight threshold is set to 0.

01.

4. The construction method according to claim 1, characterized in that: In step (4), the similarity Euclidean distance is calculated using the following formula: ED : ， In the formula, k is the reflectivity band number, p is the number of bands, i This is a remote sensing image after the continuous rainy process ended. j It is a reference image of the same period in a normal year. x is the reflectivity of the wheat pixel.

5. The construction method according to claim 1, characterized in that: In step (4), the visible light-near infrared spectrum range is 0.4-0.9 μm in the five days before the continuous rain, and the visible light-short-wave infrared spectrum range is 0.4-2.5 μm on the day after the continuous rain ends.

6. A regional scale wheat ear sprouting and mildew remote sensing identification method during continuous rainy periods, characterized in that: Based on the model obtained in claim 1, the meteorological factors and remote sensing factors of the minimum spatial pixels in the region extracted from the wheat distribution grid base map of the area to be identified are used to realize the identification of wheat ear sprouting and mildew on a regional scale.

Images