Method and device for estimating potential yield of rice

Abstract

The application relates to the computer field and provides a rice potential yield estimation method and device. The method comprises the following steps: identifying a rice planting mode of a to-be-estimated area according to synthetic aperture radar data; dividing different rice planting areas according to different rice planting modes in the to-be-estimated area; performing rice plant density inversion on different rice planting areas to obtain rice plant densities corresponding to the different rice planting areas; and estimating the rice potential yield in the corresponding rice planting area according to the different rice plant densities. The rice potential yield estimation method and device provided in the application can obtain more accurate rice plant densities in the corresponding rice planting area according to continuous large-range data in the to-be-estimated area, so that the rice potential yield in the corresponding rice planting area can be accurately estimated according to the rice plant density.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a rice potential yield estimation method and device. BACKGROUND

[0002] Rice is one of the most important food crops in China. To meet the growing demand for food, we must seek a way of agricultural production that can both high yield and efficient use of resources, and continuously and steadily improve rice yield. Potential yield refers to the yield obtained by a crop variety under sufficient water and fertilizer conditions. Studying potential yield and its influencing factors in a specific area can fully tap the production potential of rice, reveal the space for improving rice yield, and help farmers take targeted measures to improve crop yield, which has important theoretical and practical significance for realizing high and stable rice yield.

[0003] Current methods for estimating rice potential yield mainly include four types: field experiment method, high yield competition method, farmer data survey method, and crop growth model method. However, most of the above methods use single site data or multiple discrete site data to estimate the potential yield of rice in a certain area. Due to the small amount of data and lack of continuity, it is difficult to accurately estimate the potential yield of rice at the regional scale. SUMMARY

[0004] The embodiments of the present application provide a rice potential yield estimation method and device to solve the technical problem that current rice potential yield estimation methods are difficult to accurately estimate the potential yield of rice at the regional scale due to the small amount of data and lack of continuity.

[0005] In a first aspect, the embodiments of the present application provide a rice potential yield estimation method, comprising:

[0006] Identifying the rice planting method of the region to be estimated according to the synthetic aperture radar data;

[0007] Dividing different rice planting areas according to different rice planting methods in the region to be estimated;

[0008] Reversing the rice plant density for different rice planting areas to obtain the rice plant density corresponding to different rice planting areas;

[0009] Estimating the potential yield of rice in the corresponding rice planting area according to different rice plant densities.

[0010] In one embodiment, the rice planting method of the region to be estimated is identified according to the synthetic aperture radar data, comprising:

[0011] Obtaining the polarization backscattering coefficient from the full polarization synthetic aperture radar data of the region to be estimated, and generating a plurality of decomposition matrices;

[0012] Multiple polarization decomposition methods are used to perform polarization decomposition on the multiple matrices to be decomposed, resulting in multiple polarization decomposition components; the polarization decomposition methods correspond to the matrices to be decomposed.

[0013] The radar feature data of the area to be estimated is input into the land cover type identification model to obtain the land cover type of the area to be estimated; the radar feature data includes the polarization backscattering coefficient, the coherence matrix data in the multiple matrices to be decomposed, and the multiple polarization decomposition components.

[0014] The rice planting method in the area to be estimated is obtained based on the land cover type of the area to be estimated.

[0015] The land cover type identification model is obtained by training a random forest model with land cover type data and radar feature data of the sampled area within the area to be estimated.

[0016] In one embodiment, the land cover type identification model is constructed based on the following steps:

[0017] The land cover type data and radar feature data of any sampling area within the area to be estimated are used as a sample to obtain multiple samples within the area to be estimated.

[0018] A first specific proportion of samples is extracted from the plurality of samples as a first training set, and the first training set is input into a random forest model to obtain multiple decision trees in the random forest model.

[0019] The multiple decision trees are fitted sequentially in order of height from low to high to obtain the trained random forest model;

[0020] A second specific proportion of samples is extracted from the plurality of samples as a first test set. The first test set is input into the trained random forest model. The mode of the classification results of multiple decision trees in the trained random forest model is determined as the land cover type to be processed. The sum of the second feature proportion and the first feature proportion is 1.

[0021] If the type of land cover to be processed is inconsistent with the actual land cover type, then return to the step of extracting a first specific proportion of samples from the multiple samples as the first training set, and inputting the first training set into the random forest model to obtain multiple decision trees in the random forest model.

[0022] If the type of land cover to be processed is consistent with the actual land cover type, then the trained random forest model at this time is determined to be a land cover type recognition model.

[0023] In one embodiment, the step of inverting rice plant density for different rice-growing areas to obtain the corresponding rice plant density for different rice-growing areas includes:

[0024] Calculate the Spearman correlation coefficient between the rice plant density and radar features in the sampled areas of different rice planting regions; the radar features are the features corresponding to the radar feature data.

[0025] If the absolute value of the Spearman correlation coefficient is greater than the coefficient threshold, then the radar feature corresponding to the Spearman correlation coefficient is determined as the plant density feature in the corresponding rice planting area.

[0026] The plant density characteristic data of different rice planting areas are input into the corresponding rice plant density inversion model to obtain the rice plant density corresponding to different rice planting areas.

[0027] The rice plant density inversion model is obtained by training an elastic network model based on rice plant density data and plant density characteristic data from sampled areas in different rice planting regions.

[0028] In one embodiment, the rice plant density inversion model is constructed based on the following steps:

[0029] The rice plant density data and plant density characteristic data of any sampling area within any rice planting area are used as a sample to obtain multiple samples within the same rice planting area.

[0030] A third specific proportion of samples is extracted from the plurality of samples as a second training set, and the second training set is input into the elastic network model;

[0031] Calculate the minimum value of the elastic network model, and determine the elastic network model corresponding to the minimum value as the trained elastic network model;

[0032] A fourth specific proportion of samples is extracted from the plurality of samples as a second test set. The second test set is input into the trained elastic network model, and the rice plant density output by the trained elastic network model is determined as the plant density to be treated. The sum of the fourth feature proportion and the third feature proportion is 1.

[0033] If the absolute value of the error between the plant density to be processed and the actual rice plant density is greater than or equal to the first error threshold, and the number of training times of the elastic network model is less than the number of training times threshold, then after allocating the error to the weight of each plant density feature, the process returns to the step of extracting a third specific proportion of samples from the multiple samples as the second training set and inputting the second training set into the elastic network model.

[0034] If the absolute value of the error between the plant density to be processed and the actual rice plant density is less than the first error threshold, or the number of training iterations of the elastic network model is greater than or equal to the number of iterations threshold, then the trained elastic network model at this time is determined to be the rice plant density inversion model corresponding to any rice planting area.

[0035] In one embodiment, estimating the potential rice yield within a corresponding rice-growing area based on different rice plant densities includes:

[0036] Inverse distance weighted interpolation was performed on soil data from different rice-growing areas to obtain soil raster data for different rice-growing areas.

[0037] Rice growth is simulated based on rice genetic characteristic data from different rice planting areas. The rice genetic characteristic data is then adjusted using the generalized likelihood uncertainty estimation method and the trial-and-error method. If the absolute value of the error between the simulated value and the measured value of rice growth is less than the second error threshold, the rice genetic characteristic data corresponding to the simulated value is determined as the genetic characteristic data to be processed.

[0038] Based on the soil raster data, the genetic characteristic data to be processed, the rice plant density, and the daily meteorological data within a preset time period, the potential yield of rice in different rice planting areas is estimated.

[0039] In one embodiment, before obtaining the polarization backscattering coefficients based on the fully polarimetric synthetic aperture radar data of the region to be estimated, the following steps are included:

[0040] Radiometric calibration, geometric correction, and image filtering are performed on the fully polarimetric synthetic aperture radar data.

[0041] Secondly, embodiments of this application provide a rice potential yield estimation device, comprising:

[0042] The rice planting method identification module is used to: identify the rice planting method in the area to be estimated based on synthetic aperture radar data;

[0043] The rice planting area division module is used to: divide different rice planting areas according to different rice planting methods in the area to be estimated;

[0044] The rice plant density inversion module is used to: invert rice plant density for different rice planting areas and obtain the rice plant density corresponding to different rice planting areas.

[0045] The rice potential yield estimation module is used to estimate the potential yield of rice in a corresponding rice planting area based on different rice plant densities.

[0046] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the rice potential yield estimation method described in the first aspect.

[0047] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium, including a computer program, which, when executed by a processor, implements the steps of the rice potential yield estimation method described in the first aspect.

[0048] The method and apparatus for estimating potential rice yield provided in this application first identify the rice planting methods in the area to be estimated based on synthetic aperture radar (SAR) data. Then, based on the different rice planting methods in the area to be estimated, different rice planting areas are divided. Next, rice plant density is inverted for each rice planting area to obtain the corresponding rice plant density. Finally, the potential rice yield in the corresponding rice planting area is estimated based on the different rice plant densities. Since SAR data is continuous, large-scale data within the area to be estimated, the rice plant density obtained from this inversion can accurately simulate the actual rice plant density in the corresponding rice planting area, thus enabling accurate estimation of the potential rice yield in the corresponding rice planting area based on this rice plant density. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is one of the flowcharts illustrating the method for estimating the potential yield of rice provided in the embodiments of this application;

[0051] Figure 2 This is the second flowchart illustrating the method for estimating the potential yield of rice provided in the embodiments of this application;

[0052] Figure 3 This is a regional distribution map of rice planting methods provided in the embodiments of this application;

[0053] Figure 4 This is the third flowchart illustrating the rice potential yield estimation method provided in the embodiments of this application;

[0054] Figure 5 This is the fourth flowchart illustrating the rice potential yield estimation method provided in the embodiments of this application;

[0055] Figure 6This is the fifth flowchart illustrating the rice potential yield estimation method provided in the embodiments of this application;

[0056] Figure 7 This is the sixth flowchart illustrating the rice potential yield estimation method provided in the embodiments of this application;

[0057] Figure 8 This is a schematic diagram of the structure of the rice potential yield estimation device provided in the embodiments of this application;

[0058] Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] Figure 1 This is one of the flowcharts illustrating the rice potential yield estimation method provided in this application. (Refer to...) Figure 1 This application provides a method for estimating the potential yield of rice, which may include:

[0061] 101. Identify the rice planting method in the area to be estimated based on synthetic aperture radar data;

[0062] 102. Divide the rice planting areas into different regions based on the different rice planting methods in the area to be estimated;

[0063] 103. Rice plant density inversion was performed for different rice planting areas to obtain the corresponding rice plant density for different rice planting areas;

[0064] 104. Estimate the potential yield of rice in the corresponding rice planting area based on different rice plant densities.

[0065] The rice potential yield estimation method provided in this embodiment first identifies the rice planting methods in the area to be estimated based on synthetic aperture radar (SAR) data. Then, it divides the area into different rice planting regions based on these different planting methods. Next, it performs rice plant density inversion for each of these regions to obtain the corresponding rice plant density. Finally, it estimates the potential rice yield within each region based on the different rice plant densities. Since SAR data represents continuous, large-scale data within the area to be estimated, the rice plant density obtained from this inversion can accurately simulate the actual rice plant density in the corresponding rice planting region, thus enabling accurate estimation of the potential rice yield within the corresponding region.

[0066] Furthermore, since most current rice plant density inversion methods do not consider the impact of different planting methods, but the differences in the field distribution characteristics of plants caused by different planting methods will have a significant impact on their radar response characteristics, and thus have a significant impact on the rice plant density inversion through radar feature data, this embodiment first identifies the rice planting method in the area to be estimated, distinguishes different rice planting areas for different rice planting methods, and performs rice plant density inversion separately, which can make the rice density inversion of rice planting areas with different rice planting methods more accurate and reliable.

[0067] Figure 2 This is the second flowchart illustrating the method for estimating the potential yield of rice provided in the embodiments of this application;

[0068] Figure 3 This is a regional distribution map of rice planting methods provided in the embodiments of this application;

[0069] Reference Figure 2 In one embodiment, identifying the rice planting method in the area to be estimated based on synthetic aperture radar data may include:

[0070] 201. Based on the fully polarimetric synthetic aperture radar data of the region to be estimated, obtain the polarimetric backscattering coefficients and generate various matrices to be decomposed;

[0071] 202. Multiple polarization decomposition methods are used to perform polarization decomposition on multiple matrices to be decomposed, resulting in multiple polarization decomposition components;

[0072] The polarization decomposition method corresponds to the matrix to be decomposed;

[0073] 203. Input the radar feature data of the area to be estimated into the land cover type identification model to obtain the land cover type of the area to be estimated;

[0074] Radar characteristic data includes polarization backscattering coefficients, coherent matrix data in various matrices to be decomposed, and multiple polarization decomposition components;

[0075] 204. Based on the land cover types of the area to be estimated, obtain the rice planting method of the area to be estimated.

[0076] The land cover type identification model is obtained by training a random forest model with land cover type data and radar feature data of the sampled area within the area to be estimated.

[0077] In step 201, the polarization backscattering coefficient can be obtained from the fully polarimetric synthetic aperture radar data of a specific date in the area to be estimated. In this embodiment, the RADARSAT-2 fully polarimetric single-view complex data on June 27, 2012, is obtained from the rice phenological calendar of the area to be estimated, and the polarization backscattering coefficient is obtained accordingly. The fully polarimetric single-view complex data corresponds to the rice seedling stage, and the data spatial resolution (azimuth * range) is 5.2m × 7.6m. RADARSAT-2 is a high-resolution commercial radar satellite equipped with a C-band sensor.

[0078] Various matrices to be decomposed, including covariance matrix, coherence matrix, scattering matrix, and Mueller matrix, are generated based on the fully polarimetric synthetic aperture radar data of the region to be estimated.

[0079] In step 202, different polarization decomposition methods are applicable to different matrices to be decomposed. Therefore, it is necessary to use different polarization decomposition methods to decompose the corresponding matrices. For example, the covariance matrix can be decomposed using the Yamaguchi 3 / 4 polarization decomposition method, and the coherence matrix can be decomposed using the Cloude-Pottier polarization decomposition method.

[0080] In this embodiment, the polarization decomposition methods employed include Yamaguchi 3 / 4 polarization decomposition, An&Yang 3 / 4 polarization decomposition, Touzi polarization decomposition, vanZyl polarization decomposition, Freeman-Durden polarization decomposition, and Cloude-Pottier polarization decomposition. Among them, Yamaguchi 3 / 4 polarization decomposition is divided into Yamaguchi three-component polarization decomposition and Yamaguchi four-component polarization decomposition, and An&Yang 3 / 4 polarization decomposition is divided into An&Yang three-component polarization decomposition and An&Yang four-component polarization decomposition. By performing polarization decomposition on the corresponding matrix to be decomposed using these polarization decomposition methods, multiple polarization decomposition components can be obtained.

[0081] In step 203, the radar characteristic data includes polarization backscattering coefficients, coherent matrix data from various matrices to be decomposed, and multiple polarization decomposition components. In this embodiment, the radar characteristic data is specifically shown in the table below:

[0082] Table 1 Radar Feature Data of the Area to be Evaluated

[0083]

[0084]

[0085] All radar feature data images in the table above have a resolution of 5m*5m.

[0086] After inputting the radar feature data of the area to be estimated into the land cover type identification model for classification, the mode filtering can be used to further process the classification results, remove small fragments in the classification results, and mask out all land cover type areas corresponding to non-rice categories to obtain the final land cover type distribution of the area to be estimated.

[0087] In step 204, refer to Figure 3 Since different land cover types are typically suited to different rice cultivation methods, the distribution of rice cultivation methods in the area to be estimated can be obtained from the distribution of land cover types in the area to be estimated, such as... Figure 3 Distribution of rice transplanting and broadcasting methods.

[0088] Currently, most methods mainly use radar polarization backscattering coefficient to invert rice plant density. However, the polarization backscattering coefficient is the result of the interaction between the radar signal and the area to be evaluated, and it is easily affected by factors such as rainfall, irrigation, and underlying surface. This is especially true when rice seedlings are short, which can lead to large errors in the rice plant density inversion results.

[0089] In this embodiment, when identifying the rice planting method in the area to be estimated, in addition to introducing the radar polarization backscattering coefficient, multiple polarization decomposition components with different decomposition methods are also introduced. That is, as many features as possible are extracted from the radar data of the area to be evaluated to compensate for the error caused by interference in the polarization backscattering coefficient. This allows for an accurate description of the rice planting situation in the area to be evaluated, making the identification of the rice planting method more accurate. Consequently, when performing plant density inversion on different rice planting areas in the future, more accurate rice plant density data can be obtained.

[0090] Figure 4 This is the third flowchart illustrating the rice potential yield estimation method provided in this application's embodiments. (Refer to...) Figure 4 In one embodiment, the land cover type identification model is constructed based on the following steps:

[0091] 401. Take the land cover type data and radar feature data of any sampling area within the area to be estimated as a sample, and obtain multiple samples within the area to be estimated;

[0092] 402. Select a first specific proportion of samples from multiple samples as the first training set, and input the first training set into the random forest model to obtain multiple decision trees in the random forest model;

[0093] 403. Fit multiple decision trees sequentially in order of height from low to high to obtain the trained random forest model;

[0094] 404. Select a second specific proportion of samples from multiple samples as the first test set, input the first test set into the trained random forest model, and determine the mode of the classification results of multiple decision trees in the trained random forest model as the land cover type to be processed.

[0095] The sum of the proportion of the second feature and the proportion of the first feature is 1;

[0096] 405. If the type of land cover to be processed is inconsistent with the actual land cover type, return to step 402;

[0097] 406. If the type of land cover to be processed is consistent with the actual land cover type, then the trained random forest model is determined to be the land cover type recognition model.

[0098] In step 401, each land cover type in the sampling area corresponds to 62 radar feature data as shown in Table 1. That is, the land cover type data and the corresponding 62 radar feature data of each sampling area constitute a sample.

[0099] In step 402, the first feature ratio can be 60%, and the extraction method can be extraction with replacement. The number of decision trees in the random forest model is preset to be 100. Then, each decision tree is generated in the same way, that is, 60% of the samples are extracted with replacement from multiple samples as the first training set. The radar feature data in the first training set is used as the independent variable, and the corresponding land cover type data is used as the dependent variable. The data is then input into the random forest model for training, thereby generating 100 decision trees.

[0100] In step 403, during the generation of each decision tree, the Gini coefficient can be used to select different radar features as the classification basis each time the decision tree splits downwards. Since not all decision trees follow the same radar feature selection order when splitting downwards, the heights of the generated decision trees are different. A lower decision tree height means that the selection of radar features is more reasonable, and the classification result can be obtained through fewer classifications, resulting in higher classification efficiency. A higher decision tree height means that the selection of radar features is less reasonable, and the classification result can be obtained through more classifications, resulting in lower classification efficiency.

[0101] Multiple decision trees are fitted sequentially in order of height from low to high, that is, 100 decision trees are fitted in order of classification efficiency from high to low to obtain a trained random forest model. The recognition efficiency of this trained random forest model is relatively high.

[0102] In step 404, the proportion of the second feature can be 40%, and the extraction method can be extraction with replacement. Each decision tree is tested in the same way, that is, 40% of the samples are extracted with replacement from multiple samples as the first test set. The radar feature data in the first test set is used as the independent variable, and the corresponding land cover type data is used as the dependent variable. The random forest model is then trained and tested.

[0103] In step 405, when the type of land cover to be processed is inconsistent with the actual land cover type, the process returns to retraining the random forest model.

[0104] In step 406, the trained random forest model is relatively stable and has a relatively accurate recognition effect, so it can be identified as a land cover type recognition model.

[0105] This embodiment constructs a land cover type identification model using a random forest model. It can utilize sample radar feature data and land cover type data from the sampled areas in the area to be evaluated for multiple training and testing sessions, ultimately obtaining a land cover type identification model with high identification efficiency and accuracy. This helps to accurately and efficiently identify the land cover types of the entire area to be evaluated in the future.

[0106] Figure 5 This is the fourth flowchart illustrating the rice potential yield estimation method provided in this application's embodiments. (Refer to...) Figure 5 In one embodiment, rice plant density inversion is performed for different rice planting areas to obtain the rice plant density corresponding to different rice planting areas, which may include:

[0107] 501. Calculate the Spearman correlation coefficient between the rice plant density and radar characteristics in the sampling areas of different rice planting regions;

[0108] Radar features are the features corresponding to radar feature data;

[0109] 502. If the absolute value of the Spearman correlation coefficient is greater than the coefficient threshold, then the radar feature corresponding to the Spearman correlation coefficient is determined as the plant density feature in the corresponding rice planting area.

[0110] 503. Input the plant density characteristic data of different rice planting areas into the corresponding rice plant density inversion model to obtain the rice plant density corresponding to different rice planting areas.

[0111] The rice plant density inversion model is obtained by training an elastic network model based on rice plant density data and plant density characteristic data from sampled areas in different rice planting regions.

[0112] In step 501, the Spearman correlation coefficient uses a monotonic function to describe the correlation between two variables. It is a non-parametric indicator that measures the interdependence between variables. Therefore, by calculating the Spearman correlation coefficient between the rice plant density in the sampling area and the radar features in the corresponding area, radar features that are highly correlated with the rice plant density can be screened out.

[0113] In step 502, the threshold value of the coefficient can be 0.5. That is, if the absolute value of the Spearman correlation coefficient between a certain radar feature and the corresponding rice plant density is greater than 0.5, it is considered that the radar feature is highly correlated with the corresponding rice plant density, and the radar feature can be used as the plant density feature in the corresponding rice planting area.

[0114] Since each rice-growing region has its own corresponding rice cultivation method, the plant density characteristics corresponding to different rice-growing regions, obtained by screening based on the Spearman correlation coefficient, are the plant density characteristics corresponding to different rice cultivation methods, as shown in the table below:

[0115] Table 2. Plant density characteristics under different rice transplanting methods.

[0116]

[0117]

[0118] As shown in the table above, there are 12 plant density characteristics under the rice transplanting method.

[0119] Table 2. Plant density characteristics under broadcast seeding method

[0120]

[0121] As shown in the table above, there are 10 characteristics of plant density under the broadcast seeding method.

[0122] In step 503, the transplanting area and the broadcasting area each have their own rice plant density inversion model. For the transplanting area, the 12 plant density feature data corresponding to this area, as shown in Table 2, are input into the corresponding rice plant density inversion model to obtain the rice plant density of the transplanting area. For the broadcasting area, the 10 plant density feature data corresponding to this area, as shown in Table 3, are input into the corresponding rice plant density inversion model to obtain the rice plant density of the broadcasting area.

[0123] This embodiment uses Spearman correlation analysis to screen out radar features that are highly correlated with rice plant density under different rice planting methods (i.e., different rice planting areas). The rice plant density obtained by inverting these radar features will have higher accuracy.

[0124] Figure 6 This is the fifth flowchart illustrating the rice potential yield estimation method provided in this application's embodiments. (Refer to...) Figure 6 In one embodiment, the rice plant density inversion model is constructed based on the following steps:

[0125] 601. Take the rice plant density data and plant density characteristic data of any sampling area within any rice planting area as a sample, and obtain multiple samples within that rice planting area.

[0126] 602. Select a third specific proportion of samples from multiple samples as the second training set, and input the second training set into the elastic network model;

[0127] 603. Calculate the minimum value of the elastic network model, and determine the elastic network model corresponding to the minimum value as the trained elastic network model;

[0128] 604. Select a fourth specific proportion of samples from multiple samples as the second test set, input the second test set into the trained elastic network model, and determine the rice plant density output by the trained elastic network model as the plant density to be treated.

[0129] The sum of the proportions of the fourth feature and the third feature is 1;

[0130] 605. If the absolute value of the error between the plant density to be processed and the actual rice plant density is greater than or equal to the first error threshold, and the number of training times of the elastic network model is less than the number of training times threshold, then after allocating the error to the weight of each plant density feature, return to step 602.

[0131] 606. If the absolute value of the error between the plant density to be processed and the actual rice plant density is less than the first error threshold, or the number of training iterations of the elastic network model is greater than or equal to the number of iterations threshold, then the trained elastic network model at this time is determined to be the rice plant density inversion model corresponding to the rice planting area.

[0132] In step 601, taking the rice transplanting area as an example, the rice plant density of each sampling area in the rice transplanting area corresponds to 12 plant density feature data as shown in Table 2. That is, the rice plant density data of each sampling area in the rice transplanting area and the corresponding 12 plant density feature data are used as a sample of the rice transplanting area.

[0133] In step 602, the third specific proportion can be 60%, that is, 60% of the samples are drawn from multiple samples as the second training set. The elastic network model can be represented in the form of a cost function as follows:

[0134]

[0135] in:

[0136] Y1=β0+β1X1+β2X2+…+β 12 X 12 (6-2)

[0137]

[0138] Y1 represents the rice plant density within the transplanting area, and X1 to X 12 These represent the 12 plant density characteristics in Table 2, from β1 to β... 12 These represent the weights corresponding to the plant density feature, β0 is the model bias, C is the cost function value of the elastic network model, and Y1 is the weight of the plant density feature. i Let be the rice plant density corresponding to the i-th sample in the second training set. to For the i-th sample in the second training set, h corresponds to one of the 12 plant density features. i Y1 represents the rice plant density obtained from the 12 plant density feature data corresponding to the i-th sample in the second training set. i -h i Let n be the rice plant density error corresponding to the i-th sample in the second training set, and n be the number of samples in the second training set.

[0139] λ and α are regularization parameters, where 0 ≤ α ≤ 1. When α = 1, formula (6-1) retains the L1 regularization term, which is a Lasso regression model. When α = 0, formula (6-1) retains the L2 regularization term, which is a ridge regression model. This is the sum of the absolute values ​​of the weights of the 12 plant density characteristics in formula (6-2) or formula (6-3). It is the sum of squares of the weights of the 12 plant density features in formula (6-2) or formula (6-3).

[0140] The main function of regularization is to prevent overfitting. The regularization term in formula (6-1) can limit the complexity of the model, so that the model can achieve a balance between complexity and performance. At the same time, adding L1 regularization term and L2 regularization term allows the elastic network model to both use Lasso regression to remove invalid features and inherit the stability of ridge regression.

[0141] In step 603, that is, calculating the minimum value of formula (6-1), the minimum value is divided into β0 and β. 12Substituting into formula (6-2), we obtain the trained elastic network model.

[0142] In step 604, the fourth specific proportion can be 40%, that is, 40% of the samples are drawn from multiple samples as the second test set, and the plant density feature data in the second test set is used as the independent variable, and the corresponding rice plant density data is used as the dependent variable. The data is then input into the trained elastic network model for testing.

[0143] In step 605, when the error between the plant density to be processed and the actual rice plant density exceeds the acceptable range, and the number of training times of the elastic network model has not reached the predetermined number of times, the error is assigned to the weight of each plant density feature, and the training of the elastic network model is restarted.

[0144] In step 606, when the error between the density of the plants to be processed and the actual density of rice plants is within an acceptable range, or when the number of training times of the elastic network model reaches a predetermined number, the trained elastic network model at this time is determined to be the rice plant density inversion model of the transplanting area.

[0145] The rice plant density inversion model for the transplanting area obtained according to the method in this embodiment is as follows:

[0146] Y1 = -0.091T 11 -0.821VanZyl3_Odd+57.1HH-21.6VV; (6-4)

[0147] From formula (6-4), we can see that, except for T 11 Besides the four plant density features, VanZyl3_Odd, HH, and VV, the weights of the other eight plant density features were all ultimately assigned to 0, with HH and VV having relatively large weights, much greater than T. 11 The weight values ​​of VanZyl3_Odd indicate that the polarization backscattering coefficient dominates the inversion of rice plant density in the transplanting area. 11 The two plant density features, VanZyl3_Odd and VanZyl3_Odd, play a modifying and auxiliary role.

[0148] The inversion model for rice plant density in the broadcast planting area obtained according to the method in this embodiment is as follows:

[0149] Y2=-0.114gamma+2.38TSVM_alpha_s1+1.12TSVM_alpha_s2; (6-5)

[0150] As can be seen from formula (6-5), except for the three plant density features gamma, TSVM_alpha_s1 and TSVM_alpha_s2, the weights of the other seven plant density features are all assigned to 0. Furthermore, the weights of TSVM_alpha_s1 and TSVM_alpha_s2 are larger than the weight of gamma, indicating that the Touzi polarization decomposition component dominates in the inversion of rice plant density in broadcast-sown areas.

[0151] This embodiment constructs a rice plant density inversion model using an elastic network model. It can utilize Lasso regression to remove invalid features while inheriting the stability of ridge regression, preventing model overfitting. At the same time, it can use sample planting density feature data and rice planting density data from the sampling areas corresponding to specific planting methods for multiple training and testing sessions. Finally, it obtains a rice plant density inversion model with high accuracy corresponding to different rice planting areas, which helps to accurately identify the rice plant density of the corresponding rice planting areas in the future.

[0152] Figure 7 This is the sixth flowchart illustrating the rice potential yield estimation method provided in this application. (Refer to...) Figure 7 In one embodiment, estimating the potential rice yield within a corresponding rice-growing area based on different rice plant densities may include:

[0153] 701. Perform inverse distance weighted interpolation on soil data from different rice-growing areas to obtain soil raster data for different rice-growing areas;

[0154] 702. Simulate rice growth status based on rice genetic characteristic data from different rice planting areas, and adjust the rice genetic characteristic data using the generalized likelihood uncertainty estimation method and the trial-and-error method. If the absolute value of the error between the simulated value and the measured value of rice growth status is less than the second error threshold, then determine the rice genetic characteristic data corresponding to the simulated value as the genetic characteristic data to be processed.

[0155] 703. Estimate the potential yield of rice in different rice-growing areas based on soil raster data, genetic characteristic data to be processed, rice plant density, and daily meteorological data within a preset time period.

[0156] In step 701, soil data describes the physical and chemical properties of the soil, which helps to simulate soil moisture and nutrient transport characteristics as well as crop root growth processes. Soil data can be obtained by sampling and testing in a gradient manner in rice-growing areas. It can include cation exchange capacity, pH value, organic carbon content, total nitrogen content, clay content, silt content, sand content, etc. of each soil layer. Other soil profile data, such as field capacity, wilting coefficient, and saturated water content, can also be obtained from the Chinese Soil Science Database and the soil average profile measurement data and typical profile description data in the "Chinese Soil Species Records".

[0157] Inverse distance weighted interpolation was performed on soil data of various layers and attributes from different rice-growing regions to obtain soil raster data for each region. The resolution of this raster data image is consistent with the resolution of the rice plant density image for the same rice-growing region. The formula for inverse distance weighted interpolation can be expressed as follows:

[0158]

[0159] Among them, Z * (x0) represents the predicted value of soil data Z for any sampling area x0 within the rice planting area, such as pH value, organic carbon content, etc., Z(x k ) represents the k-th sampling region x that is adjacent to sampling region x0. k The measured value of soil data Z, ω k Z(x) k ) for Z * The contribution weight of (x0), where N is the number of sampling regions adjacent to the sampling region x0.

[0160] In step 702, eight rice genetic traits can be used to reflect the influence of rice's own genetic characteristics on growth and development. The specific meanings and value ranges of these eight rice genetic traits are shown in the table below:

[0161] Table 4. Meaning and Value Range of Rice Genetic Characteristics

[0162]

[0163]

[0164] Based on field survey records of rice yield, plant biomass, stem biomass, leaf biomass, panicle biomass, leaf area index, and phenological stages in the sampling area, observational data files and time-series observational data files for the CERES-Rice model were created for model calibration. First, the genetic characteristic data of typical rice varieties were used as initial values ​​in the CERES-Rice model. The generalized likelihood uncertainty estimation method was used to debug the above eight genetic characteristics, with up to 8000 debugging attempts. Rice growth was simulated based on the genetic characteristic data obtained from each debugging attempt. The combination of genetic characteristic data with the smallest error between the simulated and actual rice growth was selected as the preliminary result. Then, a trial-and-error method was used to fine-tune this preliminary result, ensuring that the simulated rice growth values ​​matched the measured values ​​as closely as possible. For example, the absolute value of the error between the simulated and measured values ​​was made less than a second error threshold. The rice genetic characteristic data corresponding to this simulated value was then used as the genetic characteristic data to be processed. The second error threshold could be 10%.

[0165] It should be noted that rice growth status can be measured by rice yield, key growth period, dry matter weight, leaf area, etc. Among them, the measured value of key growth period can be obtained from the measured phenological period of rice, the measured value of dry matter weight can be obtained from the measured plant biomass, stem biomass, leaf biomass, panicle biomass, etc., and the measured value of leaf area can be obtained from the measured leaf area index. The CERES-Rice model is used to adjust the above eight rice genetic characteristic data so that the simulated rice growth status of multiple different rice growth statuses based on these eight rice genetic characteristic data can match the measured values ​​as much as possible.

[0166] It should be noted that the DSSAT model, or Decision Support System for Agrotechnology Transfer, is an agricultural decision service software. All crop models in the software are combined into modular crop system models, which are widely used in digital agriculture and smart agriculture fields such as quantitative assessment of climate effects, crop variety optimization design, and dynamic generation of management plans. Among them, the CERES-Rice model is the rice model in the DSSAT model, which simulates the growth, development and yield formation process of rice in daily increments.

[0167] The above methods were used to obtain the genetic characteristic data to be treated corresponding to different planting areas. Since different planting areas correspond to different planting methods, and different planting methods correspond to different rice varieties, the genetic characteristic data to be treated corresponding to different planting areas are the same as the genetic characteristic data to be treated corresponding to different rice varieties, as shown in the table below:

[0168] Table 5. Genetic characteristics of different rice varieties corresponding to the treatments.

[0169] Rice variety P1 P2O P2R P5 G1 G2 G3 G4 Indica rice 263.8 12.57 189.0 445.6 55.87 0.022 0.635 0.873 Japonica rice 392.5 11.73 134.3 338.5 67.72 0.027 0.715 0.925

[0170] Among them, indica rice corresponds to the transplanting method, and japonica rice corresponds to the broadcasting method.

[0171] In step 703, the daily meteorological data within the preset time period can be obtained from the daily observation data of the Huai'an base station in Jiangsu Province in 2012 provided by the China Meteorological Data Network. Specifically, this data can include daily maximum temperature, daily minimum temperature, daily precipitation, and daily sunshine hours. Since the meteorological station does not directly observe and record daily solar radiation, it can be calculated based on the daily sunshine hours. The specific calculation formula is as follows:

[0172] R S =R A *[a+b*(m / M)]; (7-2)

[0173] Among them, R S R is the daily solar radiation. A R represents the maximum possible solar radiation for a sunny day each day, m represents the actual sunshine hours for the day, M represents the possible sunshine hours for the day, and a and b are empirical coefficients, where a can take a value of 0.248 and b can take a value of 0.752. R is the solar radiation on any day within the preset time period. A It can be calculated using the following formula:

[0174]

[0175] in, XT represents the maximum possible solar radiation on a sunny day within a preset time period, t. t Let LAT be the true solar hour angle on day t within the preset time period, and LAT be the latitude of the observation area, which can be obtained through field surveys and records. t Let XT be the solar tilt angle on day t within a preset time period. t It can be calculated using the following formula:

[0176]

[0177] Among them, SD t It can be calculated using the following formula:

[0178]

[0179] The sunshine duration M on any day within the preset time period in formula (7-2) can be calculated using the following formula:

[0180]

[0181] Among them, M t This represents the number of hours of sunshine allowed on day t within a preset time period.

[0182] Based on the above methods, soil raster data, genetic characteristic data to be processed, and daily meteorological data within a preset time period were obtained in different rice planting areas. Combined with the previously obtained rice plant density, and with the assumption that the water and nutrients required for rice growth are sufficient to ensure that the rice is not limited by water and fertilizer during its growth, spatial expansion was implemented on the programming platform, and the DSSAT model was called pixel by pixel to estimate the potential yield of rice. The maximum potential yield of the transplanted planting area was found to be 11521 kg / ha, and the maximum potential yield of the broadcast planting area was 11493 kg / ha.

[0183] It should be noted that crop management information, including rice type, planting date, planting method, planting depth, fertilization and irrigation management, and harvest time, can also be simultaneously input into the DSSAT model to estimate the potential yield of rice.

[0184] This embodiment processes the collected data in various ways to obtain the soil raster data, genetic feature data to be processed, rice plant density, and daily meteorological data within a preset time period required for model estimation, which makes the estimation of potential rice yield more accurate and reliable.

[0185] In one embodiment, before obtaining the polarization backscattering coefficients from the fully polarimetric synthetic aperture radar data of the region to be estimated, the following may be included:

[0186] Radiometric calibration, geometric correction, and image filtering are performed on fully polarimetric synthetic aperture radar data.

[0187] Image filtering can be achieved using enhanced Lee filtering, with a filter window size of 5 pixels * 5 pixels.

[0188] This embodiment, by performing radiometric calibration, geometric correction, and image filtering on fully polarimetric synthetic aperture radar (SSAR) data, can more accurately identify various radar features from SSAR data.

[0189] The following describes the rice potential yield estimation device provided in the embodiments of this application. The rice potential yield estimation device described below can be referred to in correspondence with the rice potential yield estimation method described above.

[0190] Figure 8 A schematic diagram of the structure of the rice potential yield estimation device provided in an embodiment of this application. (Refer to...) Figure 8 This application provides a device for estimating the potential yield of rice, which may include:

[0191] The rice planting method identification module 801 is used to: identify the rice planting method in the area to be estimated based on synthetic aperture radar data;

[0192] The rice planting area division module 802 is used to: divide different rice planting areas according to different rice planting methods in the area to be estimated;

[0193] The rice plant density inversion module 803 is used to: perform rice plant density inversion for different rice planting areas to obtain the rice plant density corresponding to different rice planting areas.

[0194] The rice potential yield estimation module 804 is used to estimate the potential yield of rice in a corresponding rice planting area based on different rice plant densities.

[0195] The rice potential yield estimation device provided in this embodiment first identifies the rice planting method in the area to be estimated based on synthetic aperture radar data. Then, it divides the area into different rice planting regions based on the different planting methods. Next, it performs rice plant density inversion for each different rice planting region to obtain the corresponding rice plant density. Finally, it estimates the potential rice yield within the corresponding rice planting region based on the different rice plant densities. Since the synthetic aperture radar data is continuous, large-scale data within the area to be estimated, the rice plant density obtained from this inversion can accurately simulate the actual rice plant density in the corresponding rice planting region, thus enabling accurate estimation of the potential rice yield within the corresponding rice planting region based on this rice plant density.

[0196] Furthermore, since most current rice plant density inversion methods do not consider the impact of different planting methods, but the differences in the field distribution characteristics of plants caused by different planting methods will have a significant impact on their radar response characteristics, and thus have a significant impact on the rice plant density inversion through radar feature data, this embodiment first identifies the rice planting method in the area to be estimated, distinguishes different rice planting areas for different rice planting methods, and performs rice plant density inversion separately, which can make the rice density inversion of rice planting areas with different rice planting methods more accurate and reliable.

[0197] In one embodiment, the rice planting method identification module 801 is specifically used for:

[0198] The polarization backscattering coefficients are obtained from the fully polarized synthetic aperture radar data of the region to be estimated, and various matrices to be decomposed are generated.

[0199] Multiple polarization decomposition methods are used to perform polarization decomposition on the multiple matrices to be decomposed, resulting in multiple polarization decomposition components; the polarization decomposition methods correspond to the matrices to be decomposed.

[0200] The radar feature data of the area to be estimated is input into the land cover type identification model to obtain the land cover type of the area to be estimated; the radar feature data includes the polarization backscattering coefficient, the coherence matrix data in the multiple matrices to be decomposed, and the multiple polarization decomposition components.

[0201] The rice planting method in the area to be estimated is obtained based on the land cover type of the area to be estimated.

[0202] The land cover type identification model is obtained by training a random forest model with land cover type data and radar feature data of the sampled area within the area to be estimated.

[0203] In one embodiment, the land cover type identification model in the rice planting method identification module 801 is constructed based on the following steps:

[0204] The land cover type data and radar feature data of any sampling area within the area to be estimated are used as a sample to obtain multiple samples within the area to be estimated.

[0205] A first specific proportion of samples is extracted from the plurality of samples as a first training set, and the first training set is input into a random forest model to obtain multiple decision trees in the random forest model.

[0206] The multiple decision trees are fitted sequentially in order of height from low to high to obtain the trained random forest model;

[0207] A second specific proportion of samples is extracted from the plurality of samples as a first test set. The first test set is input into the trained random forest model. The mode of the classification results of multiple decision trees in the trained random forest model is determined as the land cover type to be processed. The sum of the second feature proportion and the first feature proportion is 1.

[0208] If the type of land cover to be processed is inconsistent with the actual land cover type, then return to the step of extracting a first specific proportion of samples from the multiple samples as the first training set, and inputting the first training set into the random forest model to obtain multiple decision trees in the random forest model.

[0209] If the type of land cover to be processed is consistent with the actual land cover type, then the trained random forest model at this time is determined to be a land cover type recognition model.

[0210] In one embodiment, the rice plant density inversion module 803 is specifically used for:

[0211] Calculate the Spearman correlation coefficient between the rice plant density and radar features in the sampled areas of different rice planting regions; the radar features are the features corresponding to the radar feature data.

[0212] If the absolute value of the Spearman correlation coefficient is greater than the coefficient threshold, then the radar feature corresponding to the Spearman correlation coefficient is determined as the plant density feature in the corresponding rice planting area.

[0213] The plant density characteristic data of different rice planting areas are input into the corresponding rice plant density inversion model to obtain the rice plant density corresponding to different rice planting areas.

[0214] The rice plant density inversion model is obtained by training an elastic network model based on rice plant density data and plant density characteristic data from sampled areas in different rice planting regions.

[0215] In one embodiment, the rice plant density inversion model in the rice plant density inversion module 803 is constructed based on the following steps:

[0216] The rice plant density data and plant density characteristic data of any sampling area within any rice planting area are used as a sample to obtain multiple samples within the same rice planting area.

[0217] A third specific proportion of samples is extracted from the plurality of samples as a second training set, and the second training set is input into the elastic network model;

[0218] Calculate the minimum value of the elastic network model, and determine the elastic network model corresponding to the minimum value as the trained elastic network model;

[0219] A fourth specific proportion of samples is extracted from the plurality of samples as a second test set. The second test set is input into the trained elastic network model, and the rice plant density output by the trained elastic network model is determined as the plant density to be treated. The sum of the fourth feature proportion and the third feature proportion is 1.

[0220] If the absolute value of the error between the plant density to be processed and the actual rice plant density is greater than or equal to the first error threshold, and the number of training times of the elastic network model is less than the number of training times threshold, then after allocating the error to the weight of each plant density feature, the process returns to the step of extracting a third specific proportion of samples from the multiple samples as the second training set and inputting the second training set into the elastic network model.

[0221] If the absolute value of the error between the plant density to be processed and the actual rice plant density is less than the first error threshold, or the number of training iterations of the elastic network model is greater than or equal to the number of iterations threshold, then the trained elastic network model at this time is determined to be the rice plant density inversion model corresponding to any rice planting area.

[0222] In one embodiment, the potential rice yield estimation module 804 is specifically used for:

[0223] Inverse distance weighted interpolation was performed on soil data from different rice-growing areas to obtain soil raster data for different rice-growing areas.

[0224] Rice growth is simulated based on rice genetic characteristic data from different rice planting areas. The rice genetic characteristic data is then adjusted using the generalized likelihood uncertainty estimation method and the trial-and-error method. If the absolute value of the error between the simulated value and the measured value of rice growth is less than the second error threshold, the rice genetic characteristic data corresponding to the simulated value is determined as the genetic characteristic data to be processed.

[0225] Based on the soil raster data, the genetic characteristic data to be processed, the rice plant density, and the daily meteorological data within a preset time period, the potential yield of rice in different rice planting areas is estimated.

[0226] In one embodiment, a radar data preprocessing module (not shown) is further included, for:

[0227] Radiometric calibration, geometric correction, and image filtering are performed on the fully polarimetric synthetic aperture radar data.

[0228] Figure 9 Example: A schematic diagram of the physical structure of an electronic device, such as... Figure 9 As shown, the electronic device may include a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 can call a computer program stored in the memory 930 to execute steps of a method for estimating the potential yield of rice, such as:

[0229] Identify rice planting methods in the area to be estimated based on synthetic aperture radar data;

[0230] Different rice planting areas are divided according to the different rice planting methods in the area to be estimated;

[0231] Rice plant density was inverted for different rice planting areas to obtain the corresponding rice plant density for different rice planting areas;

[0232] The potential yield of rice in a corresponding rice-growing area is estimated based on different rice plant densities.

[0233] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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 this application. 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.

[0234] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to perform the steps of the methods provided in the above embodiments, such as including:

[0235] Identify rice planting methods in the area to be estimated based on synthetic aperture radar data;

[0236] Different rice planting areas are divided according to the different rice planting methods in the area to be estimated;

[0237] Rice plant density was inverted for different rice planting areas to obtain the corresponding rice plant density for different rice planting areas;

[0238] The potential yield of rice in a corresponding rice-growing area is estimated based on different rice plant densities.

[0239] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0240] 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.

[0241] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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. Such 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 this application.

Claims

1. A method for estimating the potential yield of rice, characterized in that, include: Identify rice cultivation methods in the area to be estimated based on synthetic aperture radar data, including: The polarization backscattering coefficients are obtained from the fully polarized synthetic aperture radar data of the region to be estimated, and various matrices to be decomposed are generated. Multiple polarization decomposition methods are used to perform polarization decomposition on the multiple matrices to be decomposed, resulting in multiple polarization decomposition components; the polarization decomposition methods correspond to the matrices to be decomposed. The radar feature data of the area to be estimated is input into the land cover type identification model to obtain the land cover type of the area to be estimated; the radar feature data includes the polarization backscattering coefficient, the coherence matrix data in the multiple matrices to be decomposed, and the multiple polarization decomposition components. The rice planting method in the area to be estimated is obtained based on the land cover type of the area to be estimated. The land cover type identification model is obtained by training a random forest model with land cover type data and radar feature data of the sampled area within the area to be estimated. Different rice planting areas are divided according to the different rice planting methods in the area to be estimated; Rice plant density was inverted for different rice planting areas to obtain the corresponding rice plant density for different rice planting areas; The potential yield of rice in a corresponding rice-growing area is estimated based on different rice plant densities.

2. The method for estimating the potential yield of rice according to claim 1, characterized in that, The land cover type identification model is constructed based on the following steps: The land cover type data and radar feature data of any sampling area within the area to be estimated are used as a sample to obtain multiple samples within the area to be estimated. A first specific proportion of samples is extracted from the plurality of samples as a first training set, and the first training set is input into a random forest model to obtain multiple decision trees in the random forest model. The multiple decision trees are fitted sequentially in order of height from low to high to obtain the trained random forest model; A second specific proportion of samples is extracted from the plurality of samples as the first test set. The first test set is input into the trained random forest model. The mode of the classification results of multiple decision trees in the trained random forest model is determined as the land cover type to be processed. The sum of the second specific ratio and the first specific ratio is 1; If the type of land cover to be processed is inconsistent with the actual land cover type, then return to the step of extracting a first specific proportion of samples from the multiple samples as the first training set, and inputting the first training set into the random forest model to obtain multiple decision trees in the random forest model. If the type of land cover to be processed is consistent with the actual land cover type, then the trained random forest model at this time is determined to be a land cover type recognition model.

3. The method for estimating potential rice yield according to claim 1, characterized in that, The process of inverting rice plant density for different rice-growing areas to obtain the corresponding rice plant density for different rice-growing areas includes: Calculate the Spearman correlation coefficient between the rice plant density and radar features in the sampled areas of different rice planting regions; the radar features are the features corresponding to the radar feature data. If the absolute value of the Spearman correlation coefficient is greater than the coefficient threshold, then the radar feature corresponding to the Spearman correlation coefficient is determined as the plant density feature in the corresponding rice planting area. The plant density characteristic data of different rice planting areas are input into the corresponding rice plant density inversion model to obtain the rice plant density corresponding to different rice planting areas. The rice plant density inversion model is obtained by training an elastic network model based on rice plant density data and plant density characteristic data from sampled areas in different rice planting regions.

4. The method for estimating the potential yield of rice according to claim 3, characterized in that, The rice plant density inversion model is constructed based on the following steps: The rice plant density data and plant density characteristic data of any sampling area within any rice planting area are used as a sample to obtain multiple samples within the same rice planting area. A third specific proportion of samples is extracted from the plurality of samples as a second training set, and the second training set is input into the elastic network model; Calculate the minimum value of the elastic network model, and determine the elastic network model corresponding to the minimum value as the trained elastic network model; A fourth specific proportion of samples is extracted from the plurality of samples as a second test set. The second test set is input into the trained elastic network model, and the rice plant density output by the trained elastic network model is determined as the plant density to be treated. The sum of the fourth specific proportion and the third specific proportion is 1. If the absolute value of the error between the plant density to be processed and the actual rice plant density is greater than or equal to the first error threshold, and the number of training times of the elastic network model is less than the number of training times threshold, then after allocating the error to the weight of each plant density feature, the process returns to the step of extracting a third specific proportion of samples from the multiple samples as the second training set and inputting the second training set into the elastic network model. If the absolute value of the error between the plant density to be processed and the actual rice plant density is less than the first error threshold, or the number of training iterations of the elastic network model is greater than or equal to the number of iterations threshold, then the trained elastic network model at this time is determined to be the rice plant density inversion model corresponding to any rice planting area.

5. The method for estimating the potential yield of rice according to claim 1, characterized in that, The method of estimating the potential rice yield in a corresponding rice-growing area based on different rice plant densities includes: Inverse distance weighted interpolation was performed on soil data from different rice-growing areas to obtain soil raster data for different rice-growing areas. Rice growth is simulated based on rice genetic characteristic data from different rice planting areas. The rice genetic characteristic data is then adjusted using the generalized likelihood uncertainty estimation method and the trial-and-error method. If the absolute value of the error between the simulated value and the measured value of rice growth is less than the second error threshold, the rice genetic characteristic data corresponding to the simulated value is determined as the genetic characteristic data to be processed. Based on the soil raster data, the genetic characteristic data to be processed, the rice plant density, and the daily meteorological data within a preset time period, the potential yield of rice in different rice planting areas is estimated.

6. The method for estimating the potential yield of rice according to claim 1, characterized in that, Before obtaining the polarization backscattering coefficients based on the fully polarimetric synthetic aperture radar data of the region to be estimated, the following steps are included: Radiometric calibration, geometric correction, and image filtering are performed on the fully polarimetric synthetic aperture radar data.

7. A device for estimating the potential yield of rice, characterized in that, The method for estimating the potential yield of rice as described in claim 1 includes: The rice planting method identification module is used to: identify the rice planting method in the area to be estimated based on synthetic aperture radar data; The rice planting area division module is used to: divide different rice planting areas according to different rice planting methods in the area to be estimated; The rice plant density inversion module is used to: invert rice plant density for different rice planting areas and obtain the rice plant density corresponding to different rice planting areas. The rice potential yield estimation module is used to estimate the potential yield of rice in a corresponding rice planting area based on different rice plant densities.

8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the rice potential yield estimation method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the rice potential yield estimation method as described in any one of claims 1 to 6.

Images