Winter wheat yield estimation method and system based on synthetic aperture radar image, storage medium and electronic equipment

Abstract

The application discloses a high-resolution winter wheat yield estimation method and system based on a synthetic aperture radar image, a storage medium and an electronic device, and relates to the technical field of yield estimation. The application discloses a kind of based on synthetic aperture radar image high-resolution winter wheat yield estimation method, system, storage medium and electronic equipment, comprising: time series SAR image is preprocessed;Time series backscattering image is masked;After removing winter wheat, time series backscattering image is segmented into multiple blocks according to the administrative boundary in the yield estimation range, then the generated neural network sample is composed of time series vector sample and image sample;Gaussian process is introduced to model the spatial structure of data explicitly;The network model is trained, and the optimal network model is saved.The yield of winter wheat in the area to be estimated can be quickly and accurately obtained by the application, providing an understanding of yield and production capacity, helping decision makers, planners and stakeholders make informed decisions, optimize resource utilization, and promote sustainable economic and social development.At the same time, the application has good performance and is easy to implement in engineering.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a high-resolution winter wheat yield estimation method and system based on synthetic aperture radar images, a storage medium and an electronic device. BACKGROUND

[0002] At present, winter wheat is one of the main crops in China, and how to estimate crop yield in a timely and effective manner to obtain accurate grain yield information has always been the focus of the agricultural field. Using remote sensing images to quickly and accurately estimate yield can provide an understanding of yield and production capacity, helping decision makers, planners and stakeholders make informed decisions, optimize resource utilization, and promote sustainable economic and social development. Therefore, the yield estimation of winter wheat is very important.

[0003] In related research on crop yield estimation, traditional methods mainly use statistical means to obtain crop yield data, but such methods have difficulty in data collection and are severely wasteful of manpower and resources. Remote sensing technology has been widely used in crop yield estimation due to its wide monitoring range and stable monitoring cycle. With the development of artificial intelligence (AI), artificial neural networks (ANN) have been used to estimate yield on several different crops, such as CNN, RNN, etc., and the results show that machine learning methods can be superior to traditional regression methods. However, the current methods do not fully utilize the spatio-temporal features of remote sensing data, and there is room for further improvement in the accuracy and precision of yield estimation. Moreover, most of these applications only use optical data, but optical images are easily affected by clouds and rain, making it difficult to obtain complete time series of remote sensing images. Synthetic aperture radar (SAR) is a microwave remote sensing sensor with all-weather and all-day imaging capabilities, and can effectively observe the earth even in cloudy and rainy weather, which can well compensate for the shortcomings of optical remote sensing.

[0004] Therefore, it is of great significance to fully extract the features of time-series SAR data to realize winter wheat yield estimation. SUMMARY

[0005] The purpose of the present application is to provide a high-resolution winter wheat yield estimation method and system based on synthetic aperture radar images, a storage medium and an electronic device, which can fully utilize the powerful feature extraction capability of deep learning combined with Gaussian process to build a winter wheat yield estimation model with strong generalization and robustness, solving the drawbacks of current yield estimation methods.

[0006] The technical solution adopted by the present application is as follows:

[0007] A high-resolution winter wheat yield estimation method based on synthetic aperture radar images, comprising the following steps:

[0008] Step S101: Preprocess the time-series synthetic aperture radar image to obtain time-series backscattering images with different polarization modes;

[0009] Step S102: Mask the time series backscattered images with different polarization modes to remove non-winter wheat pixels;

[0010] Step S103: Divide the time series backscatter map, after removing non-winter wheat pixels, into blocks according to the administrative boundaries of the production area to be estimated;

[0011] Step S104: Use a histogram dimensionality reduction method to generate neural network samples, generating a time series vector sample and an image sample for each administrative region.

[0012] Step S105: Construct LSTM and CNN respectively to extract features from time series vector samples and image samples; this is used to effectively extract the spatial and temporal features of remote sensing images, so as to make full use of the spatiotemporal characteristics of remote sensing images.

[0013] Step S106: Introduce a Gaussian process to model the spatial structure of the features extracted by LSTM and CNN, thereby further improving the accuracy of yield estimation.

[0014] Step S107: Obtain ground-measured data of winter wheat in the production area to be estimated, divide it into training set and validation set, and train the model.

[0015] Step S108: Select the coefficient of determination For network models with a value greater than 0.65, output is estimated, where y i For real data, For predictive data, The mean of the real dataset is denoted as n, and the total number of data points is denoted as n. The yield results of winter wheat are shown in the graph.

[0016] In step S101, the preprocessing includes orbit correction, thermal noise removal, radiometric calibration, deburst, multi-view, filtering, and terrain correction. Through preprocessing, each pixel of the image represents the true radar backscattering coefficient, thereby forming backscattering images with different polarization modes.

[0017] In step S102, winter wheat planting distribution data provided by the National Science and Technology Infrastructure Platform - National Ecological Science Data Center are specifically used to extract winter wheat.

[0018] In step S103, the backscattered image is divided into blocks according to the administrative boundaries of the production area to be estimated. Specifically, the administrative region vector map is used to divide the time series backscattered coefficient image of the extracted winter wheat pixels into blocks.

[0019] The step S104 specifically comprises the following steps:

[0020] Firstly, the histogram dimensionality reduction is performed on the winter wheat time series backscattering images of the administrative regions.

[0021] Then, the reduced histogram is normalized, and the normalization formula is as follows:

[0022]

[0023] In the formula, h i is the pixel histogram generated after dimensionality reduction, H i is the pixel histogram vector after normalization.

[0024] Finally, the time series vector samples and image samples are generated, and each administrative region generates a neural network sample including a time series vector sample and an image sample.

[0025] In step S106, the Gaussian process model is introduced,

[0026] The mean function is linear with respect to the deep features, and the covariance kernel depends on the spatial structure, and the kernel function is as shown in formula 3:

[0027]

[0028] In the formula, g loc -g′ loc represents the distance between the training data and the test data, ||·||2 represents the L2 norm, σ and r loc are hyperparameters, is an additional Gaussian noise term, and I is the unit matrix.

[0029] The linear Gaussian process model expression is as follows:

[0030] y(x)=f(x)+h(x) T β (4)

[0031] In the formula, f(x) ~ gp(0, k(x, x′)); h(x) represents the feature vector extracted from the deep model based on the original data; β follows a Gaussian prior b is the weight vector obtained by connecting the feature vector extracted from the deep model with the output layer, B=σ b I, wherein σ b is a hyperparameter, and I is the unit matrix.

[0032] In step S107, the network model is trained, specifically comprising the following steps:

[0033] The obtained ground measured yield data is divided into a training set and a verification set in a ratio of 8:2 at random;

[0034] The constructed training set is input into the built network model for training, the network model training effect is judged through the precision evaluation index, the parameters in the network model are adjusted, and the optimal crop classification model is obtained.

[0035] A high-resolution winter wheat yield estimation system based on synthetic aperture radar images, comprising: a preprocessing unit configured to perform orbit correction, thermal noise removal, radiation calibration, Deburst, multi-view, filtering, terrain correction operations on the SAR image to obtain backscatter images of different polarization modes, and then mask them, remove non-winter wheat pixels and segment the processed result image according to the administrative boundary of the area to be estimated;

[0036] A sample generation unit configured to generate neural network samples using a histogram dimensionality reduction technique, including a time series vector sample and an image sample;

[0037] A model construction unit configured as an LSTM network and a CNN network to extract features from the time series vector sample and the image sample respectively; then a Gaussian process is introduced to realize spatial structure modeling of data;

[0038] A test unit configured to estimate the winter wheat yield based on the optimal network model and parameters based on the input time series synthetic aperture radar image, to obtain the winter wheat yield estimation result based on the time series synthetic aperture radar.

[0039] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to cause the device where the computer-readable storage medium is located to execute the high-resolution winter wheat yield estimation method based on synthetic aperture radar images.

[0040] An electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor, the processor executing the program to implement the high-resolution winter wheat yield estimation method based on synthetic aperture radar images.

[0041] The present application is achieved.

[0042] 1. Preprocess the time series SAR image (orbit correction, thermal noise removal, radiation calibration, Deburst, multi-view, filtering, terrain correction) to obtain backscatter images of different polarization modes;

[0043] 2. Mask the time series backscatter images of different polarization modes to remove non-winter wheat pixels;

[0044] 3. Using the administrative boundary vector map of the study area to block the time series backscatter coefficient image of the extracted winter wheat pixels;

[0045] 4. Histogram dimensionality reduction is performed on the winter wheat time series backscatter image, then the reduced histogram is normalized, and finally a time series vector sample and an image sample are generated. Each administrative area generates a neural network sample, which includes a time series vector sample and an image sample.

[0046] 6. CNN and LSTM are combined to extract image sample features and time series vector sample features, respectively;

[0047] 7. A Gaussian process component is introduced to model the extracted features in a spatial structure, thereby further improving the accuracy of yield estimation;

[0048] 8. The ground measured yield data obtained is randomly divided according to a ratio of 8:2 to form a training set and a validation set; the constructed training set is input into the network model to perform training, the model training effect is judged through the precision evaluation index, the parameters in the network are adjusted, and the optimal yield estimation model is obtained.

[0049] 9. The optimal model parameters are loaded, the time series SAR image is input into the network model, and the yield estimation result is obtained. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0051] Figure 1 The flowchart of the present application;

[0052] Figure 2 The SAR image provided by the embodiment of the present application;

[0053] Figure 3 The SAR image provided by the embodiment of the present application after removing non-winter wheat pixel points;

[0054] Figure 4 The neural network sample provided by the embodiment of the present application;

[0055] Figure 5 The yield estimation result provided by the embodiment of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0057] Using remote sensing images to quickly and accurately estimate yield can provide an understanding of yield and production capacity, help decision-makers, planners, and stakeholders make informed decisions, optimize resource utilization, and promote sustainable economic and social development.

[0058] The applicant found through research on the prior art that the current remote sensing yield estimation methods can be mainly divided into statistical model-based methods, physical model-based methods, and machine learning model-based methods.

[0059] The statistical model-based method is based on the statistical relationship between remote sensing data and ground observation data, and establishes a regression model or other statistical model of crop yield and remote sensing data. Common statistical models include linear regression models, multiple regression models, etc. However, these methods often have low precision.

[0060] The physical model-based method uses physical models to simulate crop growth and yield. These models are based on crop physiological characteristics, hydrological processes, radiation transmission, and other factors, and couple remote sensing data with physical models to estimate crop growth conditions and yield. Common physical models include biophysical models, crop models, etc. However, these methods have numerous parameters that are difficult to obtain.

[0061] The machine learning-based method uses remote sensing data and measured crop yield data to construct a model, and predicts and estimates crop yield by learning the relationship between the data. Through the training process, the machine learning algorithm automatically adjusts the model parameters to establish the mapping relationship between remote sensing data and crop yield in an optimal way. However, the current machine learning method fails to fully exploit the spatiotemporal characteristics of remote sensing information.

[0062] To this end, the applicant proposes a SAR image winter wheat yield estimation method based on deep learning and Gaussian process. The time series SAR image is preprocessed (orbit correction, thermal noise removal, radiation calibration, Deburst, multi-view, filtering, terrain correction) to obtain backscatter images of different polarization modes; the time series backscatter images of different polarization modes are masked to remove non-wheat pixel points; the time series backscatter images without non-wheat pixels are blocked according to the administrative boundary of the estimated yield; a histogram dimensionality reduction technique is used to generate neural network samples, and a time series vector sample and an image sample are generated for each administrative region; LSTM and CNN are constructed respectively to extract features from time series vector samples and image samples; a Gaussian process component is introduced to explicitly model the spatial structure of the data and further improve accuracy; obtain the ground measured data of winter wheat, and randomly divide it into training set and validation set according to the ratio of 8:2, train the model; select the model with higher accuracy to estimate the yield, and obtain the yield result map of winter wheat. This method effectively extracts the spatial and temporal features of remote sensing images, which not only uses CNN to fully extract the spatial information of time series remote sensing images, but also uses LSTM to fully extract the temporal information of time series remote sensing images. The Gaussian process component is also introduced to explicitly model the spatial structure of the data, thereby further improving the yield estimation accuracy. This method uses SAR data to achieve high-precision estimation of winter wheat yield.

[0063] Exemplary method

[0064] Figure 1 A flowchart of a SAR image winter wheat yield estimation method of the present application. As shown in the figure, the method comprises the following steps: Figure 1

[0065] Step S101: Preprocess the SAR image (orbit correction, radiation calibration, Deburst, multi-view, filtering, terrain correction).

[0066] By orbit correction of the SAR image, more accurate orbit files can be obtained, making subsequent processing more accurate; radiation calibration can make each pixel value of the image represent the true radar backscatter value, and then obtain accurate backscatter images; Deburst can effectively remove the signal-free part of the Sentinel-1IW SLC image; multi-view can effectively eliminate the speckle noise in the SAR image, and greatly reduce the subsequent data volume; filtering can further eliminate the speckle noise in the SAR image to reduce the influence of speckle noise on subsequent processing; terrain correction, in addition to geographic coding, also performs terrain radiation correction processing on the SAR image, making the preprocessed image more consistent with the actual ground object. After preprocessing, backscatter images of different polarization modes are obtained. ​

[0067] Step S102: Masking the time series backscattering images of different polarization modes to remove non-wheat pixels.

[0068] The purpose of masking the backscattering images in step S102 is to remove non-winter wheat pixels in the images. This method uses the winter wheat planting distribution data provided by the National Science and Technology Infrastructure Platform-National Ecological Science Data Center (http: / / www.nesdc.org.cn) to extract winter wheat.

[0069] Step S103: Dividing the time series backscattering images with removed non-winter wheat pixels into blocks according to the administrative boundaries of the estimated yield area.

[0070] This method estimates the average yield of winter wheat in each administrative area, so it needs to process the remote sensing images of each administrative area. The time series backscattering images with removed non-winter wheat pixels processed in S102 are segmented using the administrative boundary vector map of the estimated yield area.

[0071] Step S104: Using a histogram dimensionality reduction technique to generate neural network samples, generating a time series vector sample and an image sample for each administrative area.

[0072] After determining the histogram dimensionality reduction range, each image is divided into a reasonable number of intervals, and the number of pixels is discretized and counted to generate a pixel histogram. At the same time, the generated pixel histogram is normalized by formula 2.

[0073]

[0074] In the formula, h i is the pixel histogram vector generated after dimensionality reduction, H i is the normalized pixel histogram vector. A set of time series pixel histogram vectors can be generated for each administrative area corresponding to a set of time series remote sensing images; at the same time, a matrix can be generated after fusion on the time series. Thus, a time series vector sample and an image sample are generated for each administrative area.

[0075] Step S105: Constructing LSTM and CNN respectively to extract features from time series vector samples and image samples;

[0076] This method uses LSTM to extract features from time series vector samples, which mainly consists of an input layer, an LSTM unit layer, and a fully connected layer. Each LSTM unit receives the output of the previous time and the current time vector as input. Finally, a fully connected layer is added. L2 loss is used for regression tasks. To prevent overfitting, we regularize the network by adding a dropout layer with a dropout rate of 0.75 after each state transition.

[0077] The method also uses CNN to extract the features of the image samples, which mainly consists of an input layer, 7 convolutional layers, 7 activation layers, 7 batch normalization layers, 3 Dropout layers, and 1 fully connected layer. The number of convolution kernels of the convolutional layers C1-C7 is 64, 64, 128, 128, 256, 256, and 512, respectively, and the kernel size is 3x3. The sliding step is 1, 2, 1, 2, 1, 2, and 1, respectively, and each convolutional layer is padded with 1. At the same time, batch normalization and Relu function activation operations are performed on each convolutional layer, and some Dropout layers are added to prevent model overfitting.

[0078] Step S106: Introduce a Gaussian process component to explicitly model the spatial structure of the data and further improve accuracy;

[0079] The method designs a linear Gaussian process model, in which the mean function is linear with respect to the deep features, and the covariance kernel depends on the spatial structure. The kernel function is shown in equation 3.

[0080]

[0081] where g loc -g′ loc represents the distance between the training data and the test data, ||·||2 represents the L2 norm, σ and r loc are hyperparameters, is an additional Gaussian noise term, and I is the identity matrix.

[0082] The expression of the linear Gaussian process model designed in the experiment is as follows:

[0083] y(x)=f(x)+h(x) T β (4)

[0084] where f(x)~gp(0,k(x,x′)). h(x) represents the feature vector extracted from the deep model based on the original data. β follows a Gaussian prior b is the weight vector obtained by connecting the feature vector extracted from the deep model to the output layer, B=σ b I, where σ b is a hyperparameter, and I is the identity matrix.

[0085] Step S107: Obtain the winter wheat ground measured data, divide it into a training set and a validation set, and train the model;

[0086] The ground measured data of winter wheat is obtained, and is divided into a training set and a verification set according to a ratio of 8:2. The training set and the neural network sample are input into the model for training, and the verification set is used to verify the model accuracy. First, the batch_size is set to 20, the maximum iteration number is 1000, and the learning rate is 0.001; then the training data set and the sample set are input into the network model constructed in steps S105 to S106, the network is trained, and every 10 generations, the network performance is tested by using the verification data set and the model parameters obtained by the current training are saved.

[0087] Step S108: Selecting a model with higher accuracy to estimate the yield to obtain a yield result map of winter wheat. The optimal model parameters saved in step S107 are loaded, and then the sample is input into the trained network model, and the output is the yield estimation value of each administrative region.

[0088] Figure 1 The technical flowchart of the winter wheat yield estimation method based on the synthetic aperture radar image. Figure 2 One of the time series SAR images used in the present technology is a Sentinel-1 dual-polarized spaceborne SAR image, Figure 3 The image shown is one of the backscattering coefficient maps after image preprocessing and removal of non-winter wheat pixels. As can be seen from the figure, the image noise after preprocessing is small, and the corresponding non-winter wheat pixels have all become white. Figure 4 The time series vector sample and image sample formed after the histogram dimensionality reduction technique and the normalization operation. As can be seen from the figure, the time series vector sample is fused in time to become an image sample. Figure 5 The winter wheat yield estimation result is the average yield of winter wheat in each village of Weishi County. The method of the present application can effectively estimate the yield of winter wheat. Compared with the results of the verification set, the results verify that the yield of winter wheat estimated by the method has high precision, the root mean square error (RMSE) is 63.6 kg / mu, the mean absolute error (MAE) is 53.9 kg / mu, the determination coefficient (R 2 ) is 0.698, which fully proves the effectiveness of the method. The above is only a preferred embodiment of the present application, and is not used to limit the protection scope of the present application.

[0089] In the description of the present application, it should be noted that for the orientation words, such as the term "center", "transverse", "vertical

[0090] The terms "to", "length", "width", "thickness", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicating relative orientation and position are based on the orientation or position shown in the drawings, and are used only to facilitate the description of the present application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and cannot be understood as limiting the specific protection scope of the present application.

[0091] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of the present application are intended to cover not only inclusive but also exclusive processes, methods, systems, products, or devices containing those steps or units clearly listed, but also other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0092] Note that the above is only the preferred embodiment of the present application and the application of technical principles. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and those skilled in the art can make various obvious changes, readjustments and substitutions without departing from the scope of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the specific embodiments described herein, and can include more other effective embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for high-resolution winter wheat yield estimation based on synthetic aperture radar images, characterized in that: The method comprises the following steps: Step S101, preprocessing the time series synthetic aperture radar image to obtain time series backscatter images of different polarization modes; Step S102, masking the time series backscatter images of different polarization modes to remove non-winter wheat pixel points; Step S103, dividing the time series backscatter images from which the non-winter wheat pixel points are removed into blocks according to the administrative boundaries of the to-be-estimated yield area; Step S104, using a histogram dimension reduction method to generate a neural network sample, and generating a time series vector sample and an image sample for each administrative area; Step S105, constructing an LSTM and a CNN respectively to extract features of the time series vector sample and the image sample; the LSTM and the CNN are used to effectively extract spatial and temporal features of the remote sensing image, so as to fully utilize the spatial and temporal characteristics of the remote sensing image; Step S106, introducing a Gaussian process to model the spatial structure of the features extracted by the LSTM and the CNN, so as to further improve the accuracy of yield estimation; in step S106, the Gaussian process model is introduced, The average function is linear with respect to the deep features, and the covariance kernel depends on the spatial structure, and the kernel function is as shown in formula 3: where g loc -g′ loc denotes the distance between the training and test data, ||·||2denotes the L2norm, σ and r loc are hyperparameters, is an additional Gaussian noise term, I is the identity matrix; The linear Gaussian process model is expressed as follows: y(x) = f(x) + h(x) T β (4) where f(x) ~ gp(0, k(x, x’)); h(x) represents the feature vector extracted from the deep model based on the raw data; β follows a Gaussian prior b is the weight vector obtained by connecting the feature vector extracted in the deep model with the output layer, B = σ b I, where σ b is a hyperparameter, and I is the identity matrix; Step S107, obtaining the ground measured data of winter wheat in the to-be-estimated yield area, dividing the data into a training set and a validation set, and training the model; Step S108, selecting a determination coefficient The network model greater than 0.65 is used to estimate the yield, wherein y i is real data, is predicted data, is the mean value in the real data set, and n is the total number of data; and a yield result graph of winter wheat is obtained.

2. The method for high-resolution winter wheat yield estimation based on synthetic aperture radar images according to claim 1, characterized in that, In step S101, the preprocessing includes orbit correction, thermal noise removal, radiation calibration, Deburst, multi-view, filtering, terrain correction, so that each pixel of the image represents the true radar backscatter coefficient, thereby forming backscatter images of different polarization modes.

3. The method of claim 1, wherein the method is a method of estimating high-resolution winter wheat yield based on a synthetic aperture radar image, characterized by, In step S102, the winter wheat pixel points are extracted by using the winter wheat planting distribution data provided by the National Science and Technology Infrastructure- National Ecological Science Data Center.

4. The method for high-resolution winter wheat yield estimation based on synthetic aperture radar images according to claim 1, characterized in that, In step S103, the backscatter image is divided into blocks according to the administrative boundaries of the to-be-estimated yield area, and the winter wheat pixel time series backscatter coefficient image is divided into blocks by using the administrative area vector map.

5. The method for high-resolution winter wheat yield estimation based on synthetic aperture radar images according to claim 1, characterized in that, The step S104 specifically comprises the following steps: First, histogram dimension reduction is performed on the winter wheat time series backscatter image of the administrative area; Then, the reduced histogram is normalized; the normalization formula is: where h i is the pixel histogram after dimension reduction, H i is the pixel histogram vector after normalization; Finally, a time series vector sample and an image sample are generated, wherein each administrative area generates a neural network sample including a time series vector sample and an image sample.

6. The method of claim 1, wherein the method is a method of estimating high resolution winter wheat yield based on synthetic aperture radar images, characterized by, In step S107, the network model is trained, specifically comprising the following steps: The obtained ground measured yield data is randomly divided according to a ratio of 8:2 to form a training set and a validation set; The constructed training set is input into the built network model for training, the training effect of the network model is judged by an accuracy evaluation index, the parameters in the network model are adjusted, and an optimal crop classification model is obtained.

7. A system for high resolution winter wheat yield estimation based on the synthetic aperture radar image of claim 1, characterized by, It comprises: The preprocessing unit is configured to perform orbit correction, thermal noise removal, radiation calibration, Deburst, multi-view, filtering, terrain correction operation on the SAR image, obtain the backscattering image in different polarization modes, then perform masking on the backscattering image, remove non-winter wheat pixels, and segment the processing result image according to the administrative boundary of the to-be-estimated yield area. The sample generation unit is configured to generate neural network samples by using a histogram dimension reduction technique, including a time series vector sample and an image sample. The model construction unit is configured to perform feature extraction on the time series vector sample and the image sample by using an LSTM network and a CNN network, and then introduce a Gaussian process to realize spatial structure modeling of data. The test unit is configured to estimate the winter wheat yield based on the input time series synthetic aperture radar image by using an optimal network model and parameters, and obtain a winter wheat yield estimation result based on the time series synthetic aperture radar image.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor, and the computer readable storage medium causes the device to execute the high-resolution winter wheat yield estimation method based on the synthetic aperture radar image according to any one of claims 1-7.

9. An electronic device, comprising: Comprise: A memory, a processor, and a program stored in the memory and executable on the processor, and the processor executes the program to realize the high-resolution winter wheat yield estimation method based on the synthetic aperture radar image according to any one of claims 1-7.

Images