A method for monitoring a lodging crop, an electronic device and a readable storage medium

Abstract

The application provides a lodging crop monitoring method, comprising: obtaining pre-processed field investigation sample data, optical remote sensing data and full polarization SAR satellite image data; establishing lodging sample set data and non-lodging sample set data; extracting all polarization parameters of the farmland plot in the operation area, and counting the lodging proportion of crops under each grid; establishing probability density distribution information under different polarization parameters, screening out optimal polarization parameters by using a multiple linear regression model and a backtracking method; constructing a parameter weight evaluation index by using the probability density distribution information, and determining the weight of the optimal polarization parameters; and obtaining probability spatial distribution information of the lodging crops in the operation area according to all optimal polarization parameters and the weight of all optimal polarization parameters by using a Gaussian mixture model. By constructing the parameter weight evaluation index as the weight of the estimated pixel lodging proportion, the weight of all optimal parameters is determined, and the precision of the monitoring result of the lodging crops is improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a method for monitoring lodged crops, an electronic device, and a readable storage medium. Background Technology

[0002] Against the backdrop of global climate change, extreme weather events are becoming more frequent, leading to agricultural disasters and a decline in crop yield and quality. For example, crop lodging caused by windstorms is a common type of agricultural disaster.

[0003] In recent years, the rapid development of satellite remote sensing and computer technologies has provided support for large-scale crop disaster monitoring. Currently, research on crop lodging monitoring based on satellite remote sensing technology is limited, and mainly focuses on the application of optical remote sensing data. Researchers have extracted features such as spectral reflectance, vegetation index, and texture, and combined them with machine learning algorithms to map the spatial distribution of lodging in different crops. However, for crop varieties with similar stem and leaf reflectance, optical remote sensing data cannot accurately distinguish them. Secondly, due to the spatial heterogeneity of crop growth and farmland environment, optical data commonly exhibits "different spectra for the same crop," causing confusion between lodged and non-lodged categories. Furthermore, because the quality and availability of optical satellite remote sensing data are highly dependent on weather conditions in the work area, it is difficult to apply it to real-time monitoring of lodged crops in large post-disaster areas.

[0004] Synthetic Aperture Radar (SAR) technology is virtually unaffected by weather conditions and possesses strong penetration capabilities, thus providing reliable observational data for crop lodging monitoring. However, due to the highly random scattering of crop scattering in farmland environments and the complex depolarization mechanisms of different crop types, SAR images contain significant speckle noise. While traditional qualitative lodging monitoring methods can provide clear pixel category information for plots, they lack discrimination rules for mixed pixels, resulting in poor accuracy in identifying mixed pixels and consequently, inaccurate monitoring results. Summary of the Invention

[0005] This invention provides a method, electronic device, and readable storage medium for monitoring lodged crops, in order to solve the problem that the monitoring results of lodged crops are inaccurate due to the poor recognition accuracy of mixed pixels.

[0006] Firstly, a method for monitoring lodged crops includes the following steps:

[0007] Acquire pre-processed field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data within the operational area after crop lodging;

[0008] Based on field survey sample data and optical remote sensing data, establish lodged sample set data and non-lodged sample set data;

[0009] Using fully polarimetric SAR satellite imagery data and this year's crop type spatial distribution data for masking, all polarization parameters of farmland plots in the work area were extracted, and the lodging ratio of crops under each grid was calculated.

[0010] Based on the lodging ratio of crops under each grid, probability density distribution information of lodged and non-lodged crops under different polarization parameters is established based on the lodged sample set data and the non-lodged sample set data. All preferred polarization parameters are screened out using a multiple linear regression model and the backtracking method.

[0011] Using the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters, a parameter weight evaluation index is constructed, and the weights of all preferred polarization parameters are determined.

[0012] A Gaussian mixture model was used to obtain the probability spatial distribution information of lodged crops in the work area based on all preferred polarization parameters and their weights.

[0013] Furthermore, the steps of establishing the lodged sample set data and the non-lodged sample set data include:

[0014] The collected optical remote sensing data is subjected to orthorectification and control point correction, and then fused with panchromatic and multispectral bands to obtain preprocessed optical remote sensing data.

[0015] The collected field survey sample data is organized into point vector format data to obtain preprocessed field survey sample data;

[0016] The preprocessed field survey sample data is overlaid with the preprocessed optical remote sensing data to establish an interpretation marker library for sample expansion. Using the interpretation marker library, the lodged sample set data and the non-lodged sample set data are established.

[0017] Furthermore, the step of extracting all polarization parameters of farmland plots in the work area includes:

[0018] The acquired fully polarimetric SAR satellite image data is preprocessed by amplitude conversion, multi-view processing, noise filtering and geocoding to obtain polarization parameters including polarimetric backscattering coefficient, vegetation index and polarization matrix scattering components.

[0019] All polarization parameters are combined to construct a polarization parameter set. The polarization parameters of farmland plots in the work area are obtained by masking the data of crop distribution types in the current year.

[0020] Furthermore, the step of calculating the lodging ratio of crops under each grid includes: rasterizing the lodged sample set data and the non-lodged sample set data respectively, superimposing them onto the independent grid after the polarization parameters are vectorized, traversing each independent grid through spatial statistical methods, and calculating the raster area of ​​the lodged sample set data and the raster area of ​​the non-lodged sample set data covered by each independent grid to calculate the lodging ratio of crops under each independent grid.

[0021] If the sum of the raster area of ​​the collapsed sample set data and the raster area of ​​the non-collapsed sample set data covered by a single independent grid is less than the area of ​​the independent grid, then the independent grid shall be removed.

[0022] Furthermore, the step of establishing probability density distribution information of lodged and non-lodged crops under different polarization parameters based on the lodged sample set data and the non-lodged sample set data, and using a multiple linear regression model and backtracking method to screen out the preferred polarization parameters includes:

[0023] Extract the feature values ​​of all polarization parameters from the lodged and unlodged sample data, calculate the mean and standard deviation of the feature value of each polarization parameter, and generate multiple simulated values ​​that conform to the standard normal distribution for lodged and unlodged crops respectively, and establish the probability density distribution information of lodged and unlodged crops respectively.

[0024] Using the KL divergence and JM distance methods, the KL divergence and JM distance values ​​of lodged and unlodged crops under different polarization parameters are calculated based on the simulated values.

[0025] All KL divergence values ​​and JM distance values ​​are normalized. The KL divergence values ​​and JM distance values ​​calculated for the same polarization parameter are summarized. Based on the summarized results, all polarization parameters are sorted in ascending order.

[0026] Based on all polarization parameters after sorting, a quantitative relationship between the polarization parameters and the lodging ratio of crops under each grid is established using a multiple linear regression model. The polarization parameters with the lowest ranking are eliminated one by one until the performance of the multiple linear regression model no longer declines, thus selecting all the preferred polarization parameters.

[0027] Furthermore, the step of constructing a parameter weight evaluation index and determining the weights of all preferred polarization parameters by utilizing the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters includes:

[0028] Using the probability density distribution information of lodged and non-lodged crops under different polarization parameters, the probability density distribution of lodged and non-lodged crops with all preferred polarization parameters is extracted sequentially to form a probability density distribution map. Feature values ​​including intersection points, peak values, low values, width, and half-width between lodged and non-lodged crops are obtained. Based on the feature values, a parameter weight evaluation index is constructed.

[0029] Input all preferred polarization parameters into the parameter weight evaluation index in sequence to obtain the weights of all preferred polarization parameters.

[0030] Furthermore, the step of obtaining the probability spatial distribution information of lodged crops in the work area using a Gaussian mixture model, based on all preferred polarization parameters and their weights, includes:

[0031] All the preferred polarization parameters are encoded into a matrix. Using a Gaussian mixture model and the expectation-maximization algorithm, the collapse probability at all pixel positions is obtained. The collapse probability at each pixel of each polarization parameter is multiplied by its weight to obtain the result of multiplying the collapse probability of all preferred polarization parameters by their corresponding weights.

[0032] The probability predictions of all the preferred polarization parameters are multiplied by the weights and then superimposed. Based on the superimposed result, the lodging probability of all pixels in the working area is traversed to obtain the probability spatial distribution information of lodged crops in the working area.

[0033] Furthermore, the Gaussian mixture model is an unsupervised method, and the parameters of the Gaussian mixture model mainly include the number of iterations and the number of classes;

[0034] The optimal number of iterations is searched by setting the value range and step size. For each iteration, the prediction result of the Gaussian mixture model is evaluated by the AIC value. Within the same value range, the number of iterations corresponding to the minimum AIC value is optimal. The number of categories is set to two.

[0035] In a second aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the above-described methods for monitoring lodged crops.

[0036] Thirdly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the above-described methods for monitoring lodged crops.

[0037] This invention provides a method for monitoring lodged crops, a non-transitory computer-readable storage medium, and an electronic device. By collecting field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data, all preferred polarization parameters are selected to ensure a true mapping relationship between the preferred polarization parameters and the proportion of lodged crop area. Simultaneously, a parameter weight evaluation index is constructed to determine the weights of all preferred polarization parameters. After inputting the preferred polarization parameters into a Gaussian mixture model, the lodged crop monitoring information is transformed into quantitative lodged crop information, effectively suppressing the impact of mixed pixels on the performance of lodged crop monitoring.

[0038] Meanwhile, based on the Gaussian mixture model, different polarization parameters are fused to make weighted decisions on the lodging state of crops within a pixel, predicting the lodging probability of crops in the work area. This overcomes the uncertainty caused by random scattering and speckle noise in single polarization features, thereby improving the accuracy of lodging crop monitoring results. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the lodging monitoring method for crops provided by this invention;

[0041] Figure 2 This is a schematic diagram of the weighted evaluation index constructed in the lodging crop monitoring method provided by the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0043] This invention uses a Gaussian mixture model to fuse different polarization parameters to predict the lodging probability of crops in the work area, overcoming the problem of poor accuracy of traditional classification methods in identifying mixed pixels and improving the accuracy of crop lodging monitoring.

[0044] The following is combined with Figure 1 and Figure 2 The method for monitoring lodged crops according to the present invention includes the following steps:

[0045] S1: Acquire pre-processed field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data within the work area after crop lodging.

[0046] The preprocessing steps include:

[0047] The acquired optical remote sensing data undergoes orthorectification and control point correction, and is then fused with panchromatic and multispectral bands to obtain preprocessed optical remote sensing data. In this embodiment, the optical remote sensing data is obtained from satellite imagery within the operational area. After fusion of multispectral and panchromatic data, the spatial resolution can reach 2 meters. Using two high-resolution 1-meter panchromatic and 4-meter multispectral cameras mounted on the optical remote sensing satellite, the fused spatial resolution can reach 1 meter.

[0048] Meanwhile, the collected field survey sample data was organized into point vector format data to obtain preprocessed field survey sample data.

[0049] After preprocessing the acquired fully polarimetric SAR satellite image data, including amplitude conversion, multi-view processing, noise filtering, and geocoding, the preprocessed fully polarimetric SAR satellite image data is obtained.

[0050] S2. Based on field survey sample data and optical remote sensing data, establish lodged sample set data and non-lodged sample set data.

[0051] The specific steps for establishing the lodged and non-lodged sample data include:

[0052] The preprocessed field survey sample data is overlaid with the preprocessed optical remote sensing data to establish an interpretation marker library for sample expansion. Using the interpretation marker library, the lodged sample set data and the non-lodged sample set data are established.

[0053] S3. Using fully polarimetric SAR satellite imagery data and the spatial distribution data of crop types for this year, a mask is created to extract all polarization parameters of farmland plots in the work area, and the lodging ratio of crops under each grid is calculated.

[0054] Specifically, the steps for extracting all polarization parameters of farmland plots in the work area are as follows:

[0055] After preprocessing, the fully polarimetric SAR satellite image data yields polarization parameters including polarimetric backscattering coefficients, vegetation indices, and polarization matrix scattering components.

[0056] All polarization parameters are combined to construct a polarization parameter set. The polarization parameters of farmland plots in the work area are obtained by masking the data of crop distribution types in the current year.

[0057] In this embodiment, fully polarimetric SAR satellite imagery can be acquired through fully polarimetric SAR satellites. For example, using a C-band multipolar SAR remote sensing satellite with a resolution of 1 meter, which has 12 imaging modes and a wide imaging swath, can meet the needs of large-scale disaster monitoring.

[0058] Then, all polarization parameters are combined to construct a polarization parameter set. Using the crop distribution type data of this year as a mask, the polarization parameters of farmland plots in the operation area are obtained.

[0059] The extracted polarization backscattering coefficients include VV, VH, HH, and HV.

[0060] Based on the VV, VH, HH and HV polarization backscattering coefficients, vegetation index features are extracted, including radar vegetation index RVI, biomass index BMI, volume scattering index VSI and canopy structure index CSI.

[0061] Specifically, the radar vegetation index (RVI) is determined according to equation (1):

[0062]

[0063] The biomass index (BMI) is determined according to equation (2):

[0064]

[0065] Based on the biomass index BMI and HV polarization backscattering coefficient, the volume scattering index VSI is calculated according to equation (3):

[0066]

[0067] The canopy structure index (CSI) is determined according to equation (4):

[0068]

[0069] The specific steps for extracting the scattering components of the polarization matrix include:

[0070] Based on single-look complex SLC data from fully polarimetric SAR satellite imagery, coherence matrix T3 is extracted, and the coherence matrix T3 is filtered using the Refine Lee filtering method.

[0071] Input the coherence matrix T3 into the Freeman-Durden decomposition model to obtain the polarization matrix scattering components of the crop in the work area. The polarization matrix scattering components include the volume scattering component Fv, the surface scattering component Fs, and the secondary scattering component Fd.

[0072] Specifically, based on the assumption of reflection symmetry, each pixel in the fully polarimetric SAR satellite imagery data can be expressed as a 3×3 coherence matrix T3, as shown in equation (4):

[0073]

[0074] Where * denotes conjugate, and A0, B, B0, C and D are Huynen parameters.

[0075] Input the coherence matrix T3 into the Freeman-Durden polarization decomposition model to obtain the volume scattering component Fv, the surface scattering component Fs, and the secondary scattering component Fd, as shown in equation (5):

[0076] [T3] = P v F v +P s F s +P d F d (5);

[0077] Wherein, Ps, Pd, and Pv represent the scattering energy of the surface scattering component Fs, the secondary scattering component Fd, and the volume scattering component Fv, respectively.

[0078] Furthermore, based on the first-order Bragg scattering model, the surface scattering component Fs can be determined according to equation (6):

[0079]

[0080] The secondary scattering component Fd is established based on the dihedral reflection model and can be determined according to equation (7):

[0081]

[0082] The volume scattering component Fv is determined according to equation (8):

[0083]

[0084] Where β represents the normalized difference in scattering from the Bragg surface in the co-polarization channel; α represents the normalized difference in Fresnel reflection from a pair of orthogonal surfaces with different dielectric materials in the co-polarization channel.

[0085] Specifically, the steps for calculating the lodging rate of crops under each grid cell include:

[0086] The lodged sample set data and the non-lodged sample set data are rasterized and superimposed onto independent grids after any polarization parameter is vectorized. Each independent grid is traversed by spatial statistical methods, and the raster area of ​​the lodged sample set data and the raster area of ​​the non-lodged sample set data covered by each independent grid are statistically analyzed to determine the lodging ratio of crops under each independent grid.

[0087] Specifically, if the sum of the raster areas of the lodged sample data and the non-lodged sample data covered by a single independent grid is less than the area of ​​the independent grid, that independent grid is discarded. This is to ensure, to the greatest extent possible, the accurate mapping relationship between polarization parameters and the area ratio of lodged crops.

[0088] S4. Based on the lodging ratio of crops under each grid cell, and using the lodged and unlodged sample sets, establish the probability density distribution information of lodged and unlodged crops under different polarization parameters. Then, using a multiple linear regression model and a backtracking method, select all preferred polarization parameters. The specific steps are as follows:

[0089] The eigenvalues ​​of all polarization parameters are extracted from the lodged and unlodged sample sets. The mean and standard deviation of the eigenvalues ​​of each polarization parameter are calculated, and multiple simulated values ​​conforming to a standard normal distribution are generated for both lodged and unlodged crops. This establishes the probability density distribution information for each crop. For example, 1000 sample data conforming to a normal distribution are simulated and generated for both lodged and unlodged crops, and these 1000 sample data are used as the probability density distributions for lodged and unlodged crops.

[0090] Using the KL divergence and JM distance methods, the KL divergence and JM distance values ​​of lodged and unlodged crops under different parameters were calculated based on simulated values.

[0091] All KL divergence values ​​and JM distance values ​​are normalized. The KL divergence values ​​and JM distance values ​​calculated for the same polarization parameter are added together. Based on the summed results, all polarization parameters are sorted in ascending order.

[0092] Using polarization parameters as the dependent variable and grid collapse ratio as the independent variable, a multiple linear regression model is input. A backtracking method is used to sequentially eliminate polarization parameters ranked last in feature importance. The model is then refitted, and its accuracy is calculated and compared with the previous model's accuracy. This process is repeated until the model's accuracy no longer decreases, and the variables included in the model at this point are selected as the preferred polarization parameters.

[0093] Specifically, establishing a quantitative relationship between polarization parameters and corresponding collapse ratios includes:

[0094] The data was split according to the polarization parameters and their corresponding grid collapse ratios. 70% of the data was used for model training, and 30% was used for model accuracy evaluation.

[0095] Using 70% of the segmented polarization parameter pixel values ​​as independent variables and the corresponding grid collapse percentage as the dependent variable, a multiple linear regression model was fitted to establish a quantitative relationship between polarization parameters and collapse percentage. After training, the model accuracy was evaluated using the remaining 30% of the polarization parameter pixel values ​​and their collapse percentages. Evaluation metrics included the coefficient of determination (R²) and the root mean square error (RMSE).

[0096] In this embodiment, after inputting the simulated normally distributed lodged sample data and non-lodged sample data into the KL divergence and JM distance methods, the present invention normalizes and adds the KL divergence values ​​and JM distance values. The calculation process includes:

[0097] First, the simulated samples are input into the KL divergence method to calculate the symmetric KL divergence value.

[0098] Input the probability density distribution p(x) of the collapsed simulated sample and the probability density distribution q(x) of the non-collapsed simulated sample into the KL divergence method, and calculate the KL divergence value H1(p,q) of p(x) relative to q(x), as shown in formula (9):

[0099]

[0100] Calculate the KL divergence value H2(q,p) of q(x) with respect to p(x), as shown in formula (10):

[0101]

[0102] Based on the divergence values ​​H1(p,q) and H2(q,p), the symmetric KL divergence value KL is determined by equation (11):

[0103] KL=H1(p,q)+H2(q,p) (11);

[0104] Secondly, the simulated samples are input into the JM distance calculation method to obtain the JM distance value:

[0105] Based on the probability density distribution p(x) of the collapsed simulated sample and the probability density distribution q(x) of the non-collapsed simulated sample, the Bartholin's distance B between the two distributions is calculated as shown in formula (12):

[0106]

[0107] JM distance is a metric based on the Bahia distance B, determined according to equation (13):

[0108] JM=2*(1-e -B (13);

[0109] Where mp and mq are the average values ​​of the simulated samples of lodged and unlodged crops, respectively, and ∑X p and ∑X q These are the covariance matrices of the simulated samples of lodged and unlodged crops, respectively.

[0110] Finally, extract the maximum and minimum values ​​of the symmetric KL divergence values ​​KL of all polarization parameters, as well as the maximum and minimum values ​​of the JM distance values ​​JM, and normalize them according to equation (14):

[0111]

[0112] Among them, X i The normalized KL or JM value, x min and x max These are the minimum and maximum KL values ​​for all polarization parameters, respectively, or x. min and x max These are the minimum and maximum JM values ​​for all polarization parameters, respectively.

[0113] The normalized KL and JM values ​​are summed, and the polarization parameters are sorted from largest to smallest according to the summation. All polarization parameters are used as independent variables, and the grid collapse percentage is used as dependent variable. The data are input into a multiple linear regression model to establish a quantitative relationship between polarization parameters and corresponding collapse percentages. The back-out method is used to eliminate the variables ranked last in turn, as shown in formula (15).

[0114] Y i = b0 + b1x1 + b2x2 + b3x3 + ... + b k x k (15);

[0115] Where Y i This represents the percentage of grid collapse, where b0 is a constant term, and b1, b2, b3…, b… k Let x1, x2, x3, ..., x be the regression coefficients. k For different polarization parameters.

[0116] S5. Using the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters, construct parameter weight evaluation index and determine the weight of all preferred polarization parameters.

[0117] In this embodiment, the specific steps for constructing the parameter weight evaluation index are as follows:

[0118] Using the probability density distribution information of lodged and non-lodged crops under different polarization parameters, the probability density distribution of lodged and non-lodged crops with all preferred polarization parameters is extracted sequentially to form a probability density distribution map. Feature values ​​including intersection points, peak values, low values, width, and half-width between lodged and non-lodged crops are obtained. Based on the feature values, a parameter weight evaluation index is constructed.

[0119] Input all preferred polarization parameters into the parameter weight evaluation index in sequence to obtain the weights of all preferred polarization parameters.

[0120] Specifically, please refer to the following: Figure 2 Extract the intersection points and low peaks of the two distributions, and obtain the probability density h at the two feature points respectively. int and h lp The weighted evaluation index OLR is determined by equation (16):

[0121]

[0122] Where root is the x-axis value of the intersection of the two probability density functions, and μ1 and μ2 are the average values ​​of the collapsed and uncollapsed categories, respectively.

[0123] S6 uses a Gaussian mixture model to obtain the probability spatial distribution information of lodged crops in the work area based on all preferred polarization parameters and their weights.

[0124] Specifically, the Gaussian mixture model is an unsupervised method. The main parameters of the model include the number of iterations and the number of classes. The optimal number of iterations is searched by setting a range and step size. For each iteration, the model's prediction results are evaluated using the Akaike information criterion (AIC). Within the same range, the iteration number corresponding to the minimum AIC value is considered optimal. The number of classes is set to two in all iterations.

[0125] In this embodiment, the specific steps for obtaining the spatial distribution information of the probability of lodged crops in the work area are as follows:

[0126] All preferred polarization parameters are encoded into a matrix. The model parameters are estimated using a Gaussian mixture model and the expectation-maximization algorithm (EM). The collapse probability at all pixel locations is obtained. The collapse probability at each pixel of each polarization parameter is multiplied by its weight to obtain the result of multiplying the collapse probability of all preferred polarization parameters by their corresponding weights.

[0127] The probability predictions of all preferred polarization parameters are multiplied by the weights and then superimposed. Based on the superimposed result, the lodging probability of all pixels in the working area is traversed to obtain the probability spatial distribution information of lodged crops in the working area.

[0128] Specifically, a Gaussian mixture model is used to express the likelihood estimate of the collapse probability for the input polarization parameters. For any element xi in each preferred polarization parameter matrix, it is assumed to come from a superpopulation D, which is a finite number (M) of populations D1, D2, D3, ..., DM, mixed in proportions α1, α2, α3, ..., αM. For the mixing proportion α... i Then we have, as in formula (17):

[0129] And a i ≥0(i=1,2,3,…,M)(17);

[0130] Therefore, all elements D in each preferred polarization parameter matrix can be modeled as independent mixing densities, as shown in formula (18):

[0131]

[0132] Where, p i (x i |θ j ) represents the probability density distribution corresponding to mixture j, and is given by θ j Parameterization, θ=(α1,α2,α3,…,α M ,θ1,θ2,θ3,…,θ M ) represents all unknown parameters related to the mixing density of the M component, x i It represents any pixel value in a given preferred polarization parameter.

[0133] For a multivariate normal distribution θ j From the mean vector μ j The mixture is composed of different components of the covariance matrix ∑j. The log-likelihood function L(θ) of this mixture density can be defined by formula (19):

[0134]

[0135] Specifically, the parameter vector θ is updated using the Expectation-Maximization (EM) algorithm:

[0136] The first step is to introduce the auxiliary function Q(θ,θ) t-1 Estimate the expectation of the log-likelihood function L(θ), as shown in formula (20):

[0137]

[0138] Where, θ t-1 These are the parameters from the previous iteration.

[0139] The second step is to calculate the new estimates of the parameters. The EM algorithm iterates through these two steps until convergence is achieved. The new estimates of the parameters (in the k-th iteration) based on the old parameters from the first step are given by the following formulas (21), (22), and (23):

[0140]

[0141]

[0142]

[0143] in, and All are new estimates of the parameter vector θ of the log-likelihood function L(θ) of mixture j in the k-th iteration, and n is the total number of pixels in a certain preferred polarization parameter.

[0144] Specifically, the probability of collapse is estimated for all polarization parameters. For each parameter, the probability matrix estimated by the Gaussian mixture model is multiplied by the weights, and the summation is performed over all parameters to determine the probability of collapse for each pixel, as given by the following formula (24):

[0145]

[0146] Where, p li Let f be the probability of the i-th pixel falling over in the image, and let w be the polarization parameter of the f-th pixel. f x is the weight of the f-th polarization parameter. fi This is the value of the i-th pixel in the f-th polarization parameter.

[0147] This invention uses collected field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite image data to screen out all preferred polarization parameters, ensuring a true mapping relationship between the preferred polarization parameters and the area ratio of crop lodging.

[0148] Meanwhile, by constructing a parameter weight evaluation index to determine the weights of all preferred polarization parameters, and inputting the preferred polarization parameters into the Gaussian mixture model, the lodging crop monitoring information is transformed into quantitative lodging crop information, effectively suppressing the impact of mixed pixels on the performance of lodging crop monitoring.

[0149] Furthermore, by using a Gaussian mixture model to fuse different polarization parameters to make weighted decisions on the lodging state of crops within a pixel, the probability of lodging of crops in the work area is predicted. This overcomes the uncertainty caused by random scattering and speckle noise in single polarization features, thereby improving the accuracy of lodging crop monitoring results.

[0150] The technical solution of this invention will be further explained below, using Lishu County, Siping City, Jilin Province as the operational area for monitoring lodged crops.

[0151] Lishu County, Siping City, Jilin Province, was selected as the monitoring area for lodging crops. Located on the Songliao Plain, Lishu County has a temperate semi-humid continental monsoon climate with simultaneous rainfall and heat. The average annual sunshine duration is 2644.2 hours. The county has diverse soil types, primarily black soil, and its overall altitude ranges from 250 to 400 meters. The main crops grown in the area are corn, soybeans, and rice. On September 4th and 8th, 2020, crops in the area experienced varying degrees of lodging due to the impact of Typhoons Maysak and Bavi.

[0152] Field survey samples stored in the GPS system were exported and imported into QGIS software for editing, with appropriate calibration performed on sample points with insufficient positioning accuracy. Simultaneously, preprocessed optical remote sensing data and the current year's crop type spatial distribution map were loaded into the software. Combined with the calibrated sample points, an interpretation image library of different lodging degrees and non-lodging areas for different crops was established. Based on the sample points, lodged and non-lodged sample sets were established. Specifically, the software's vector surface construction function was used to mark plot boundaries and the spatial extent of lodged crops within plots. Multiple plots were selected and converted into raster data consisting of "0" and "1", where "0" represents non-lodged areas and "1" represents lodged areas. All preprocessing of the optical remote sensing data was performed in PIE-basic software.

[0153] In this case, since feature selection requires building a regression model, a grid is generated using any fully polarimetric parameter. Raster sample data is then overlaid, and the proportion of lodged crops within each grid is calculated using spatial statistical methods. It is important to note that after completing the statistics, non-true value grids need to be removed; that is, grids where the total area of ​​lodged and non-lodged crops within the same grid is less than the area of ​​that grid. This ensures, to the greatest extent possible, the accurate mapping between the polarimetric parameter and the proportion of lodged area.

[0154] The acquired fully polarimetric SAR satellite imagery data was input into PIE-SAR 6.3 software to extract polarization parameters for the operational area. For each of the four polarimetric channels, the backscattering coefficients VV, VH, HH, and HV were obtained. The preprocessing steps mainly included amplitude conversion, multi-look processing, noise filtering, and geocoding. Based on the backscattering coefficients, four vegetation index features were calculated: RVI, BMI, VSI, and CSI.

[0155] Before constructing the polarization scattering components, the image is first subjected to amplitude transformation, and then the coherence matrix T3 is extracted. The Refine Lee filter is selected to denoise the T3 matrix, and the window size is set to 3×3. The matrix is ​​then input into the Freeman-Durden decomposition method to obtain three scattering components of the crops in the work area, including: volume scattering component Fv, surface scattering component Fs, and secondary scattering component Fd.

[0156] All polarization parameters are combined and masked using the crop category distribution map of the current year. Polarization parameters within the farmland area of ​​the work area are extracted to suppress the influence of non-farmland features on model performance.

[0157] Furthermore, feature values ​​of all polarization parameters in both lodged and unlodged areas were extracted, and their means and standard deviations were calculated. 1000 simulated values ​​conforming to a standard normal distribution were generated for each region. Based on these simulated values, the KL divergence and JM distance of lodged and unlodged crops under different parameters were calculated. All KL divergences and JM distances were normalized, and the KL divergences and JM distances calculated for the same polarization parameter were summed. Based on the summation results, all polarization parameters were sorted in ascending order. All polarization parameters were used as dependent variables, and the corresponding grid lodging percentage was used as independent variables. These were input into a multiple linear regression model, and the lowest-ranked polarization parameters were successively eliminated until the model's performance no longer declined. The variable included in the current model was selected as the optimal variable. The feature selection process was implemented using a Python environment.

[0158] Furthermore, based on the probability density distribution of lodged and unlodged areas, low-peak and intersection point features are extracted to construct a parameter weight evaluation index and calculate the weights of all optimal polarization feature parameters.

[0159] All optimized polarization parameters are input into a Gaussian mixture model. The "Component" parameter of the model is set to 2, and the iteration rule is set as follows: the iteration range is 100-1000, and the step size is 100. The AIC is fused to optimize the number of iterations, and the iteration number corresponding to the minimum AIC is selected. Based on the optimal model parameters, the probability of corn lodging in the work area is estimated pixel by pixel, generating spatial distribution results.

[0160] The lodging crop monitoring method based on Gaussian mixture model and fully polarimetric SAR data adopted in this invention uses large-scale, high-time-sensitivity SAR remote sensing data for crop lodging monitoring. It constructs a lodging crop monitoring method based on Gaussian mixture model and fully considers the uncertainty of lodging crop information in mixed pixels, proposes a weighted evaluation index, and then integrates probability estimates of different polarization parameters to solve the problem that the current lodging crop classification method has low classification accuracy for mixed pixels and cannot meet the requirements of precision agriculture.

[0161] On the other hand, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. The processor can call logical instructions in the memory to execute a lodging crop monitoring method, including the following steps:

[0162] Acquire pre-processed field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data within the operational area after crop lodging.

[0163] Based on field survey sample data and optical remote sensing data, lodged sample data and non-lodged sample data were established.

[0164] Using fully polarimetric SAR satellite imagery data and this year's crop type spatial distribution data for masking, all polarization parameters of farmland plots in the work area were extracted, and the lodging ratio of crops under each grid was calculated.

[0165] Based on the lodging ratio of crops under each grid, probability density distribution information of lodged and non-lodged crops under different polarization parameters is established based on the lodged sample set data and the non-lodged sample set data. All preferred polarization parameters are selected by using a multiple linear regression model and the backtracking method.

[0166] By utilizing the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters, a parameter weight evaluation index is constructed, and the weights of all preferred polarization parameters are determined.

[0167] A Gaussian mixture model was used to obtain the probability spatial distribution information of lodged crops in the work area based on all preferred polarization parameters and their weights.

[0168] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0169] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the lodged crop monitoring method based on fully polarimetric synthetic aperture radar data provided by the above methods, including the following steps:

[0170] Acquire pre-processed field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data within the operational area after crop lodging.

[0171] Based on field survey sample data and optical remote sensing data, lodged sample data and non-lodged sample data were established.

[0172] Using fully polarimetric SAR satellite imagery data and this year's crop type spatial distribution data for masking, all polarization parameters of farmland plots in the work area were extracted, and the lodging ratio of crops under each grid was calculated.

[0173] Based on the lodging ratio of crops under each grid, probability density distribution information of lodged and non-lodged crops under different polarization parameters is established based on the lodged sample set data and the non-lodged sample set data. All preferred polarization parameters are selected by using a multiple linear regression model and the backtracking method.

[0174] By utilizing the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters, a parameter weight evaluation index is constructed, and the weights of all preferred polarization parameters are determined.

[0175] A Gaussian mixture model was used to obtain the probability spatial distribution information of lodged crops in the work area based on all preferred polarization parameters and their weights.

[0176] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the lodging crop monitoring methods provided by the above methods, including the following steps:

[0177] Acquire pre-processed field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data within the operational area after crop lodging.

[0178] Based on field survey sample data and optical remote sensing data, lodged sample data and non-lodged sample data were established.

[0179] Using fully polarimetric SAR satellite imagery data and this year's crop type spatial distribution data for masking, all polarization parameters of farmland plots in the work area were extracted, and the lodging ratio of crops under each grid was calculated.

[0180] Based on the lodging ratio of crops under each grid, probability density distribution information of lodged and non-lodged crops under different polarization parameters is established based on the lodged sample set data and the non-lodged sample set data. All preferred polarization parameters are selected by using a multiple linear regression model and the backtracking method.

[0181] By utilizing the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters, a parameter weight evaluation index is constructed, and the weights of all preferred polarization parameters are determined.

[0182] A Gaussian mixture model was used to obtain the probability spatial distribution information of lodged crops in the work area based on all preferred polarization parameters and their weights.

[0183] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0184] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring lodged crops, characterized in that, Includes the following steps: Acquire pre-processed field survey sample data, optical remote sensing data, and fully polarimetric SAR satellite imagery data within the operational area after crop lodging; Based on field survey sample data and optical remote sensing data, establish lodged sample set data and non-lodged sample set data; Using fully polarimetric SAR satellite imagery data and this year's crop type spatial distribution data for masking, all polarization parameters of farmland plots in the work area were extracted, and the lodging ratio of crops under each grid was calculated. Based on the lodging ratio of crops under each grid, probability density distribution information of lodged and non-lodged crops under different polarization parameters is established based on the lodged sample set data and the non-lodged sample set data. All preferred polarization parameters are screened out using a multiple linear regression model and the backtracking method. Using the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters, a parameter weight evaluation index is constructed, and the weights of all preferred polarization parameters are determined. A Gaussian mixture model was used to obtain the probability spatial distribution information of lodged crops in the work area based on all preferred polarization parameters and their weights.

2. The method for monitoring lodged crops according to claim 1, characterized in that, The steps for establishing the lodged sample set data and the non-lodged sample set data include: The collected optical remote sensing data is subjected to orthorectification and control point correction, and then fused with panchromatic and multispectral bands to obtain preprocessed optical remote sensing data. The collected field survey sample data is organized into point vector format data to obtain preprocessed field survey sample data; The preprocessed field survey sample data is overlaid with the preprocessed optical remote sensing data to establish an interpretation marker library for sample expansion. Using the interpretation marker library, the lodged sample set data and the non-lodged sample set data are established.

3. The method for monitoring lodged crops according to claim 1 or 2, characterized in that, The steps for extracting all polarization parameters of farmland plots in the work area include: The acquired fully polarimetric SAR satellite image data is preprocessed by amplitude conversion, multi-view processing, noise filtering and geocoding to obtain polarization parameters including polarimetric backscattering coefficient, vegetation index and polarization matrix scattering components. All polarization parameters are combined to construct a polarization parameter set. The polarization parameters of farmland plots in the work area are obtained by masking the data of the spatial distribution of crop types in the current year.

4. The method for monitoring lodged crops according to claim 3, characterized in that, The steps for calculating the lodging ratio of crops under each grid include: rasterizing the lodging sample set data and the non-lodging sample set data respectively, superimposing them onto independent grids after the polarization parameters have been vectorized, traversing each independent grid using spatial statistical methods, and calculating the raster area of ​​the lodging sample set data and the raster area of ​​the non-lodging sample set data covered by each independent grid to calculate the lodging ratio of crops under each independent grid. If the sum of the raster area of ​​the collapsed sample set data and the raster area of ​​the non-collapsed sample set data covered by a single independent grid is less than the area of ​​the independent grid, then the independent grid shall be removed.

5. The method for monitoring lodged crops according to claim 1, characterized in that, The step of establishing probability density distribution information of lodged and non-lodged crops under different polarization parameters based on the lodged sample set data and the non-lodged sample set data, and selecting the preferred polarization parameters using a multiple linear regression model and a backtracking method, includes: Extract the feature values ​​of all polarization parameters from the lodged and unlodged sample data, calculate the mean and standard deviation of the feature value of each polarization parameter, and generate multiple simulated values ​​that conform to the standard normal distribution for lodged and unlodged crops respectively, and establish the probability density distribution information of lodged and unlodged crops respectively. Using the KL divergence and JM distance methods, the KL divergence and JM distance values ​​of lodged and unlodged crops under different polarization parameters are calculated based on the simulated values. All KL divergence values ​​and JM distance values ​​are normalized. The KL divergence values ​​and JM distance values ​​calculated for the same polarization parameter are summarized. Based on the summarized results, all polarization parameters are sorted in ascending order. Based on all polarization parameters after sorting, a quantitative relationship between the polarization parameters and the lodging ratio of crops under each grid is established using a multiple linear regression model. The polarization parameters with the lowest ranking are eliminated one by one until the performance of the multiple linear regression model no longer declines, thus selecting all the preferred polarization parameters.

6. The method for monitoring lodged crops according to claim 1, characterized in that, The steps of constructing parameter weight evaluation indicators and determining the weights of all preferred polarization parameters by utilizing the probability density distribution information of lodged and non-lodged crops under all preferred polarization parameters include: Using the probability density distribution information of lodged and non-lodged crops under different polarization parameters, the probability density distribution of lodged and non-lodged crops with all preferred polarization parameters is extracted sequentially to form a probability density distribution map. Feature values ​​including intersection points, peak values, low values, width, and half-width between lodged and non-lodged crops are obtained. Based on the feature values, a parameter weight evaluation index is constructed. Input all preferred polarization parameters into the parameter weight evaluation index in sequence to obtain the weights of all preferred polarization parameters.

7. The method for monitoring lodged crops according to claim 1, characterized in that, The step of obtaining the probability spatial distribution information of lodged crops in the work area using a Gaussian mixture model, based on all preferred polarization parameters and their weights, includes: All the preferred polarization parameters are encoded into a matrix. Using a Gaussian mixture model and the expectation-maximization algorithm, the collapse probability at all pixel positions is obtained. The collapse probability at each pixel of each polarization parameter is multiplied by its weight to obtain the result of multiplying the collapse probability of all preferred polarization parameters by their corresponding weights. The lodging probabilities of all the preferred polarization parameters are multiplied by their corresponding weights and then superimposed. Based on the superimposed result, the lodging probabilities of all pixels in the working area are traversed to obtain the probability spatial distribution information of lodged crops in the working area.

8. The method for monitoring lodged crops according to claim 7, characterized in that, The Gaussian mixture model is an unsupervised method, and the parameters of the Gaussian mixture model mainly include the number of iterations and the number of classes; The optimal number of iterations is searched by setting the value range and step size. For each iteration, the prediction result of the Gaussian mixture model is evaluated by the AIC value. Within the same value range, the number of iterations corresponding to the minimum AIC value is optimal. The number of categories is set to two.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the lodging crop monitoring method as described in any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the lodging crop monitoring method as described in any one of claims 1 to 8.

Images