A SAR image azimuth blur suppression method, system and device based on a guided CycleGAN network

Abstract

This application discloses a method, system, and device for azimuth blur suppression of SAR images based on a guided CycleGAN network, relating to the field of remote sensing images. The method includes: acquiring radar operating parameters and constructing a training dataset based on these parameters; the training dataset contains pairs of SAR image samples with and without azimuth blur; constructing a guided CycleGAN network; the guided CycleGAN network includes a first generator, a second generator, a first discriminator, and a second discriminator; training the guided CycleGAN network using the training dataset; introducing a minimum energy weighting criterion for optimal estimation during training, serving as the guiding direction; and processing the SAR image with azimuth blur using the trained guided CycleGAN network to output a reconstructed scene image after azimuth blur suppression. This application can improve the azimuth blur suppression capability of SAR images while maintaining resolution and is applicable to all SAR operating modes.

Description

Technical Field

[0001] This application relates to the field of remote sensing imagery, and in particular to a method, system, and device for suppressing azimuth blur in SAR images based on a guided CycleGAN network. Background Technology

[0002] The azimuth sampling rate of spaceborne Synthetic Aperture Radar (SAR) is typically 1.1 to 1.2 times the 3dB Doppler bandwidth, much smaller than the entire Doppler spectrum. According to the Nyquist sampling theorem, this causes aliasing in the Doppler spectrum, i.e., azimuth ambiguity. It also means that the echoes received by the antenna sidelobes are superimposed on the echoes received by the main lobe. Although the sidelobes are weaker than the main lobe, when the main lobe and sidelobes observe weak and strong targets respectively, azimuth ambiguity can mask the true weak targets. To suppress azimuth ambiguity, Moreira et al. (see "Suppressing the azimuth ambiguities in synthetic aperture radar images") designed a pair of filters using precise prior knowledge of spaceborne SAR systems. The outputs of the two filters were canceled to filter out the ambiguity. However, this method is only applicable to isolated ambiguity areas. Guarnieri et al. (see "Adaptive removal of azimuth ambiguities in SAR images") estimated local ambiguity using antenna patterns and weakened ambiguity in the frequency domain by constructing bandpass filters. However, this method loses resolution and is only applicable to stripe mode. Summary of the Invention

[0003] The purpose of this application is to provide a method, system, and device for suppressing azimuth blur in SAR images based on a guided CycleGAN network. This method can improve the ability to suppress azimuth blur in SAR images while ensuring resolution, and is applicable to all working modes of SAR.

[0004] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for suppressing SAR image azimuth blur based on a guided CycleGAN network, including: The radar operating parameters are obtained, and a training dataset is constructed based on the radar operating parameters; the training dataset contains pairs of SAR image samples with azimuth ambiguity and SAR image samples without azimuth ambiguity. A guided CycleGAN network is constructed; the guided CycleGAN network includes a first generator, a second generator, a first discriminator, and a second discriminator. The guided CycleGAN network is trained using the training dataset; during the training process, a minimum energy weighting criterion is introduced for optimal estimation as a guiding direction. The trained guided CycleGAN network is used to process SAR images with orientation blur, and outputs reconstructed scene images after orientation blur suppression.

[0005] Secondly, this application provides a SAR image orientation blur suppression system based on a guided CycleGAN network, comprising: A training dataset construction module is used to obtain radar operating parameters and construct a training dataset based on the radar operating parameters; the training dataset contains pairs of SAR image samples with azimuth ambiguity and SAR image samples without azimuth ambiguity. The model building module is used to build a guided CycleGAN network; the guided CycleGAN network includes a first generator, a second generator, a first discriminator, and a second discriminator. The training module is used to train the guided CycleGAN network using the training dataset; during the training process, a minimum energy weighting criterion is introduced to perform optimal estimation as a guiding direction; The orientation blur suppression module is used to process SAR images containing orientation blur using a trained guided CycleGAN network, and outputs a reconstructed scene image after orientation blur suppression.

[0006] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for suppressing SAR image azimuth blur based on a guided CycleGAN network.

[0007] According to the specific embodiments provided in this application, this application has the following technical effects: This application constructs paired training datasets with and without azimuth ambiguity, employs a guided CycleGAN network structure containing dual generators and dual discriminators, and introduces the minimum energy weighting criterion as a guiding direction to achieve optimal estimation during training. This enables adaptive suppression of azimuth ambiguity without sacrificing image resolution, effectively eliminating the masking effect of strong target sidelobes on weak real targets. It also eliminates the dependence on precise system prior knowledge, breaking through the limitations of traditional methods that are only applicable to isolated blurred regions and stripe patterns. This approach has a wider range of applications and greater versatility, ultimately outputting high-quality azimuth ambiguity-suppressed and reconstructed SAR images. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating a SAR image orientation blur suppression method based on a guided CycleGAN network, provided as an embodiment of this application; Figure 2 A schematic diagram of the training process of a guided CycleGAN network; Figure 3 This is a schematic diagram of the encoder structure; Figure 4 This is a schematic diagram of the residual module. Figure 5 This is a schematic diagram of the decoder structure; Figure 6 This is a schematic diagram of the discriminator. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and 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.

[0011] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] In one exemplary embodiment, such as Figure 1 As shown, a method for suppressing directional blur in SAR images based on a guided CycleGAN network is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using a server as an example, and includes the following steps S1 to S4.

[0013] S1: Obtain radar operating parameters and construct a training dataset based on the radar operating parameters; the training dataset contains pairs of SAR image samples with azimuth ambiguity and SAR image samples without azimuth ambiguity.

[0014] S2: Construct a guided CycleGAN network. For example... Figure 2As shown, the guided CycleGAN network includes a first generator. Second generator First discriminator Second discriminator .

[0015] S3: Train the guided CycleGAN network using the training dataset; during the training process, introduce the minimum energy weighting criterion for optimal estimation as the guiding direction.

[0016] S4: The trained guided CycleGAN network is used to process SAR images with orientation blur and output the reconstructed scene image after orientation blur suppression.

[0017] Implementing steps S1 to S4 above can improve the ability to suppress azimuth blur in SAR images while ensuring resolution, and is applicable to all SAR operating modes.

[0018] In one specific embodiment, step S1 specifically includes the following steps.

[0019] S11: Divide the SAR image into image blocks of a preset size, and select image blocks that meet the preset conditions to form an image dataset; the image dataset includes multiple SAR image samples with orientation ambiguity.

[0020] Multiple large-size SAR images were sorted into 128-bit formats. The dataset is divided into 128 segments. Image patches with obvious target features (such as ships or urban areas on the ocean) are selected, while image patches without obvious target features (such as ocean scenes without ships) are removed. This dataset constitutes the image dataset. .

[0021] S12: Based on the radar operating parameters, obtain the echo dataset corresponding to the image dataset.

[0022] (1) Obtain radar operating parameters. The radar operating parameters include: scene center reference slant range. Satellite flight speed Azimuth antenna dimensions Number of transmitted pulses Frequency modulation of radar transmitted signals Radar operating wavelength Time vector of radar transmitted pulse ;in Indicates the SAR launch number The sampling time of each pulse This indicates the transpose operation.

[0023] (2) The observation scene Deployed in a 3D coordinate system, with scene point spacing set simultaneously. For example, the scene pixel pitch It can be set to one-sixth the size of the azimuth antenna. / 6. Based on the pixel spacing settings for this scenario, assuming a total of... Calculate the scene points. middle The coordinates of each scene point.

[0024] Assuming the initial location of satellite sampling is Then the coordinates of the satellite's sequential sampling positions are: Finally, according to the first The observation scene is determined by the coordinates of each scene point and the coordinates of the satellite sampling location. Slope distance of all scene points at each sampling time .

[0025] (3) Based on the radar working principle and training data, the observation scenario is concentrated. Constructing the echo matrix from the slant range: This two-dimensional matrix represents the observation scene. The original echo signal matrix. Wherein, Indicates the azimuth sampling time as Distance to sampling time The echo, also known as the echo matrix Among them, the first line, number The elements of the column. Among them: Let c represent the step function, and c be the speed of light. For the first The closest slant distance between the scene point and the satellite.

[0026] S13: The echo dataset is imaged using a back projection algorithm to obtain a reference dataset; the reference dataset includes multiple SAR image samples without orientation blur.

[0027] Determine the observation scene Corresponding echo matrix Imaging results based on back projection (BP) in the Staggered SAR system (reference dataset). It should be noted that obtaining imaging results based on the echo matrix is ​​a mature existing technology and will not be elaborated upon in this application.

[0028] S14: Construct a training dataset based on the image dataset and the reference dataset.

[0029] The acquired dataset The dataset is divided into a test dataset (used to test the trained guided CycleGAN network) and a training dataset in a 9:1 ratio.

[0030] In one specific embodiment, the network structures of the first generator (for mapping a blurred image to an unblurred image) and the second generator (for mapping an unblurred image back to a blurred image) are the same, both including an encoder, multiple residual modules and a decoder.

[0031] The encoder consists of an initial convolutional module and two downsampling modules. The initial convolutional module captures low-level features in the image, such as edges and textures. Figure 3 As shown, the input image is first mirror-filled to preserve edge information, and then a 7-step process with a stride of 1 is used. A 7-kernel convolution expands the number of channels, and finally, a non-linear (ReLU) function combined with dimensionality normalization is used to extract low-level features of the input image. The downsampling module extracts multi-dimensional high-level features of the target image by reducing the resolution of the feature map. It consists of convolutional layers, dimensionality normalization, a ReLU function, and a Dropout layer. The convolutional kernels in the convolutional layers have a size of 3. 3. With a step size of 2, the Dropout layer can increase the model's generalization ability and prevent overfitting.

[0032] To enable the network to learn more complex target details from SAR images, several residual modules were added. Figure 4 There are 3 residual modules in the middle, each residual module is mirror-filled and has a size of 3. 3. It consists of convolution with a stride of 1, dimension normalization, and the ReLU function.

[0033] The decoder consists of two upsampling modules and one output module. For example... Figure 5 As shown, the upsampling module consists of a step size of 2 and a size of 3. It consists of a 3x3 deconvolutional layer, dimension normalization, ReLU function, and Dropout layer. The output module consists of a mirror padding layer with a stride of 1 and a size of 7. It consists of 7 deconvolutional layers, dimension normalization, and the ReLU function.

[0034] Specifically, due to the large parameters and network structure model in the generator and discriminator, the input image size cannot be set too large; in this application, it is set to 128. 128.

[0035] The first discriminator (distinguishing between real and fake target domain images) and the second discriminator (distinguishing between real and fake source domain images) have the same network structure, both including multiple convolutional modules and one output module. For example... Figure 6 As shown, the convolutional module consists of a size of 4 4. The module consists of convolutional layers with a stride of 2, dimension normalization, Leaky non-linear activation layers, and Dropout layers. The output module is a single unit with a size of 4. 4. Composed of convolutional layers with a stride of 2.

[0036] In one specific embodiment, when training the guided CycleGAN network using the training dataset, an alternating iterative strategy is employed to update the parameters of the guided CycleGAN network. In each iteration, a batch of source domain images (including SAR image samples with azimuth blur) and target domain images (SAR image samples without azimuth blur) are randomly selected from the training dataset. Each training iteration executes the following steps: (1) Input the image dataset into the first generator to obtain the preliminary generated image.

[0037] Will Input the first generator To obtain a preliminary generated image .

[0038] (2) Based on the preliminary generated image, the minimum energy weighting criterion is used to make the optimal estimate and obtain the optimal estimated image.

[0039] Using the minimum energy weighting criterion Perform optimal estimation to obtain the optimal estimated image. .

[0040] (3) Input the optimal estimated image and the reference dataset into the first discriminator to calculate the first adversarial loss.

[0041] First confrontation loss The calculation formula is: This represents the first discriminator. This indicates the expected operation.

[0042] (4) Input the reference dataset into the second generator to obtain the pseudo source domain image.

[0043] Reference dataset Input to the second generator The pseudo-source domain image is obtained. .

[0044] (5) Input the pseudo-source domain image and the image dataset together into the second discriminator and calculate the second adversarial loss.

[0045] Secondary combat losses The calculation formula is: This represents the second discriminator.

[0046] (6) Input the optimal estimated image into the second generator to obtain the reconstructed source domain image. , This is the optimal estimation operation based on the minimum energy weighting criterion.

[0047] (7) The reference dataset is passed through the second generator and the first generator in sequence to obtain the reconstructed target domain image. .

[0048] (8) Calculate the cycle consistency loss based on the reconstructed source domain image and the reconstructed target domain image.

[0049] Cyclic consistent loss The calculation formula is as follows: .

[0050] (9) Construct a total loss function by combining the first adversarial loss, the second adversarial loss, and the cycle consistent loss.

[0051] The expression for the total loss function is: For the total loss, These are the weighting coefficients for the cycle-consistent loss. The weighting coefficient for the first adversarial loss. The weighting coefficients for the second adversarial loss. and Same. In this embodiment, , .

[0052] (10) Based on the total loss function, backpropagate to update the parameters of the first generator, the second generator, the first discriminator and the second discriminator.

[0053] Repeat the above steps until the preset training epochs (epoch=200) are reached, at which point training should be stopped.

[0054] In the training process described above, the minimum energy weighting criterion is used for optimal estimation. The specific process for obtaining the optimal estimated image is as follows: (1) Perform azimuth Fourier transform on the initially generated image to obtain the azimuth frequency domain signal.

[0055] By generator generated image Perform the following operations: in, It is the azimuth bandwidth. and The real goal and the first The azimuth antenna pattern of the ambiguous region. and They represent the first The distances of each ambiguous region relative to the main region in the azimuth and range directions. Indicates a filter. It is a mixing factor in the slip-cohesion mode. It is the azimuth frequency. It is the azimuth sampling frequency. It is the satellite's flight speed. It is the signal wavelength. Indicates the size of the azimuth antenna. This represents the sinc function. This represents a rectangular window. It represents the azimuth and distance of the target in the image from the center of the scene. Represents the original image After generator In position The image generated on express In matrix form. express In matrix form. Representing time-domain images The orientation Fourier transform. , , All of these are intermediate variables. and This represents the local orientation ambiguity.

[0056] The specific operation method for this location Fourier transform is as follows: Extraction The first column, the first column of this column vector Each element and the following The elements are swapped, and then a Fast Fourier Transform (FFT) is performed on the resulting column vector. After the FFT operation, the first half of the column matrix is ​​swapped with the second half again, thus completing the transformation of the original image. The first column of data undergoes an orientation-to-Fourier transform. This is then applied sequentially to the original image. Performing the above operations on each column of data completes the azimuth-direction Fourier transform of the original image. Let the azimuth-direction frequency domain signal be denoted as... .

[0057] (2) Based on the azimuth frequency domain signal, calculate the proportion of unambiguous energy to the total image energy.

[0058] The local azimuth ambiguity-to-signal ratio (localAASR) is calculated. .

[0059] , All of these are intermediate variables.

[0060] according to Seeking The ratio of unblurred energy at a location to the energy of the entire image : (3) Determine the energy estimate based on the aforementioned proportionality coefficient.

[0061] remember and These represent the energy estimates for the main region and the fuzzy region, respectively.

[0062] (4) Calculate the phase error based on the energy estimation, and determine the optimal frequency threshold according to the preset maximum phase error limit.

[0063] make ,and The corresponding frequency domain range From the frequency domain range The phase error is calculated, and the maximum phase error cannot exceed [a certain value]. Determine an optimal threshold. .

[0064] (5) Based on the optimal frequency threshold, the minimum energy weighting criterion is used to make the optimal estimation and obtain the optimal estimated image.

[0065] Assuming intermediate variables : So .in The floor symbol indicates rounding down. Indicates based on the optimal threshold The corresponding frequency range. Solving based on the minimum energy weighting criterion. The image obtained after blur suppression (i.e., the optimal estimated image) is: ,in , Represents the conjugate transpose of a matrix. This represents a matrix formed by using each element of a vector as a value on the diagonal axis.

[0066] Given the advantages of deep learning in image enhancement, this application utilizes the CycleGAN network to automatically mine azimuth blur features in SAR images and leverages the network's strong transfer learning capabilities across image domains to transform the azimuth blur suppression problem of SAR images into an image transfer problem involving SAR images with and without azimuth blur. Specifically, the SAR image with azimuth blur is used as the source domain of the CycleGAN network, and the SAR image with azimuth blur suppression is used as the target domain. Through adversarial training of two pairs of generators and discriminators within the network, image transfer from the source domain to the target domain is achieved, thus completing the training of the SAR image azimuth blur suppression model. However, the CycleGAN network's ability to learn image features is relatively limited, leading to incomplete feature extraction in complex environments. To further improve the azimuth blur suppression capability of the CycleGAN network, this application incorporates antenna azimuth map weighted prior information into the traditional CycleGAN network. While maintaining image resolution, the optimal estimate after blur suppression is explicitly expressed based on the minimum weighted energy criterion, thereby further guiding the network to better suppress azimuth blur. In addition, since the antenna azimuth map describes spatial variations, it is applicable to a variety of arbitrary operating modes (strip, scan, slide-focus, spot-focus, TOPs).

[0067] Based on the same inventive concept, this application also provides a system for implementing the above-mentioned SAR image azimuth blur suppression method based on a guided CycleGAN network. The solution provided by this system is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the SAR image azimuth blur suppression system based on a guided CycleGAN network provided below can be found in the limitations of the SAR image azimuth blur suppression method based on a guided CycleGAN network described above, and will not be repeated here.

[0068] In one exemplary embodiment, a SAR image orientation blur suppression system based on a guided CycleGAN network is provided, comprising the following modules.

[0069] A training dataset construction module is used to obtain radar operating parameters and construct a training dataset based on the radar operating parameters; the training dataset contains pairs of SAR image samples with azimuth ambiguity and SAR image samples without azimuth ambiguity.

[0070] The model building module is used to build a guided CycleGAN network; the guided CycleGAN network includes a first generator, a second generator, a first discriminator, and a second discriminator.

[0071] The training module is used to train the guided CycleGAN network using the training dataset; during the training process, the minimum energy weighting criterion is introduced to perform optimal estimation as a guiding direction.

[0072] The orientation blur suppression module is used to process SAR images containing orientation blur using a trained guided CycleGAN network, and outputs a reconstructed scene image after orientation blur suppression.

[0073] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device stores data to be processed. The I / O interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with an external terminal via a network connection. When the computer program is executed by the processor, it implements the steps in the above-described method embodiments.

[0074] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0075] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0076] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for SAR image orientation based on a guided CycleGAN network The blur suppression method is characterized by comprising: The radar operating parameters are obtained, and a training dataset is constructed based on the radar operating parameters; the training dataset contains pairs of SAR image samples with azimuth ambiguity and SAR image samples without azimuth ambiguity. A guided CycleGAN network is constructed; the guided CycleGAN network includes a first generator, a second generator, a first discriminator, and a second discriminator. The guided CycleGAN network is trained using the training dataset; during the training process, a minimum energy weighting criterion is introduced for optimal estimation as a guiding direction. The trained guided CycleGAN network is used to process SAR images with orientation blur, and outputs reconstructed scene images after orientation blur suppression.

2. The SAR image orientation based on a guided CycleGAN network as described in claim 1 The fuzziness suppression method is characterized by, A training dataset is constructed based on the aforementioned radar operating parameters, specifically including: The SAR image is divided into image blocks of a preset size, and image blocks that meet preset conditions are selected to form an image dataset; the image dataset includes multiple SAR image samples with orientation ambiguity. Based on the radar operating parameters, obtain the echo dataset corresponding to the image dataset; The echo dataset is imaged using a back projection algorithm to obtain a reference dataset; the reference dataset includes multiple SAR image samples without azimuth blur. A training dataset is constructed based on the image dataset and the reference dataset.

3. The SAR image orientation blur suppression method based on guided CycleGAN network according to claim 1, characterized in that, The first generator and the second generator have the same network structure, both including an encoder, multiple residual modules and a decoder; the first discriminator and the second discriminator have the same network structure, both including multiple convolutional modules and an output module.

4. The SAR image orientation based on a guided CycleGAN network as described in claim 2 The fuzziness suppression method is characterized by, When training the guided CycleGAN network using the training dataset, the parameters of the guided CycleGAN network are updated using an alternating iterative strategy. Each training iteration performs the following steps: The image dataset is input into the first generator to obtain a preliminary generated image; Based on the initially generated image, the minimum energy weighting criterion is used to perform optimal estimation, resulting in the optimal estimated image. The optimal estimated image and the reference dataset are input together into the first discriminator to calculate the first adversarial loss; The reference dataset is input into the second generator to obtain a pseudo-source domain image; The pseudo-source domain image and the image dataset are input together into the second discriminator to calculate the second adversarial loss; The optimal estimated image is input into the second generator to obtain the reconstructed source domain image; The reference dataset is passed sequentially through the second generator and the first generator to obtain the reconstructed target domain image; Calculate the cycle consistency loss based on the reconstructed source domain image and the reconstructed target domain image; A total loss function is constructed by combining the first adversarial loss, the second adversarial loss, and the cycle-consistent loss. Based on the total loss function, the parameters of the first generator, the second generator, the first discriminator, and the second discriminator are updated through backpropagation.

5. The SAR image orientation based on a guided CycleGAN network as described in claim 4 The fuzziness suppression method is characterized by, Based on the initially generated image, the minimum energy weighting criterion is used to perform optimal estimation, resulting in the optimal estimated image, specifically including: Perform an azimuth Fourier transform on the initially generated image to obtain an azimuth frequency domain signal; Based on the azimuth frequency domain signal, calculate the proportion of unambiguous energy to the total image energy; Energy estimation is determined based on the aforementioned proportionality coefficient; The phase error is calculated based on the energy estimation, and the optimal frequency threshold is determined according to the preset maximum phase error limit. Based on the optimal frequency threshold, the minimum energy weighting criterion is used to perform optimal estimation, resulting in the optimal estimated image.

6. The SAR image orientation blur suppression method based on guided CycleGAN network according to claim 4, characterized in that, The expression for the total loss function is: in, For the total loss, For cycle-consistent loss, As the first instance of combat losses, For the second confrontation loss, These are the weighting coefficients for the cycle-consistent loss. The weighting coefficient for the first adversarial loss. The weighting coefficients for the second adversarial loss. For reference dataset, For image datasets, Indicates the first generator. Indicates the second generator. This is the optimal estimation operation based on the minimum energy weighting criterion. This represents the expectation operation. This represents the first discriminator. This represents the second discriminator. This is the optimal estimated image.

7. A method for SAR image orientation based on a guided CycleGAN network A fuzziness suppression system, characterized in that it comprises: A training dataset construction module is used to obtain radar operating parameters and construct a training dataset based on the radar operating parameters; the training dataset contains pairs of SAR image samples with azimuth ambiguity and SAR image samples without azimuth ambiguity. The model building module is used to build a guided CycleGAN network; the guided CycleGAN network includes a first generator, a second generator, a first discriminator, and a second discriminator. The training module is used to train the guided CycleGAN network using the training dataset; during the training process, a minimum energy weighting criterion is introduced to perform optimal estimation as a guiding direction; The orientation blur suppression module is used to process SAR images containing orientation blur using a trained guided CycleGAN network, and outputs a reconstructed scene image after orientation blur suppression.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the SAR image orientation blur suppression method based on a guided CycleGAN network as described in any one of claims 1-6.

Images