Sar target data generation method and system fusing scattering feature optimization

Abstract

The application relates to a SAR target data generation method and system fusing optimized scattering characteristics. The method comprises the following steps: converting attribute scattering center parameters into a scattering characteristic visual image by using an OMP algorithm; inputting the scattering characteristic visual image and a SAR complex data image into an improved cycle-consistent generative adversarial network to output a generated image; calculating multiple losses between an original image and the generated image by using a loss function, and guiding the improved cycle-consistent generative adversarial network to learn scattering rules by using the multiple losses, and then generating SAR target data. The attribute scattering center parameters are extracted by using the OMP algorithm and are converted into the scattering characteristic visual image, and then the generated image is obtained, so that the physical scattering characteristics of the generated image are more real. The improved cycle-consistent generative adversarial network is used to learn the scattering rules, the image quality can be optimized, and the scattering characteristic distribution of the generated image is more similar to that of a real image.

Description

Technical Field

[0001] This invention relates to the field of synthetic aperture radar data generation technology, and in particular to a method and system for generating SAR target data with optimized fusion scattering characteristics. Background Technology

[0002] Synthetic Aperture Radar (SAR) has become a crucial Earth observation tool due to its unique imaging mechanism, with widespread applications in both military and civilian fields. With the development of artificial intelligence, deep learning has been widely applied in SAR interpretation research. However, due to the high cost of SAR data acquisition, the datasets currently available for deep model training are very limited. Data augmentation methods include traditional data generation methods and deep learning methods based on generative models. Traditional SAR image data augmentation methods fall into two categories: data augmentation methods utilizing geometric transformations and image generation methods based on data modeling. Deep learning methods based on generative models primarily utilize restricted Boltzmann machines (RBMs), variational autoencoders (VAEs), and generative adversarial networks (GANs) to achieve data augmentation.

[0003] However, data augmentation methods based on geometric transformations typically employ image rotation and mirroring, failing to consider the imaging mechanisms of radar and merely increasing the diversity of image data. Data modeling methods utilize computer-aided design (CAD) software to accurately model the target, subsequently obtaining the target SAR image through electromagnetic calculations. However, constructing the target model requires not only expert knowledge but also continuous model refinement to ensure the quality of the simulated SAR image. This method consumes significant human and time resources and struggles to simulate the target's real-world environment. Deep learning methods based on generative models also have limitations: RBMs and VAEs lack the ability to capture image details, resulting in low-quality generated SAR images and significant discrepancies between the target's structural information and real data; traditional GANs do not incorporate scattering information from SAR imaging, leading to simulated data that cannot effectively represent the scattering characteristics of real targets.

[0004] Therefore, traditional SAR image data augmentation methods often suffer from poor data generation quality and deviation from the true scattering patterns, resulting in low image quality. Summary of the Invention

[0005] Based on this, in order to solve the above-mentioned technical problems, a method and system for generating SAR target data by fusing scattering feature optimization is provided, which can improve the quality of the generated image and make the scattering feature distribution of the generated image more similar to that of the real image.

[0006] A method for generating SAR target data by fusing scattering feature optimization, the method comprising:

[0007] Acquire SAR complex data images, extract attribute scattering center parameters from the SAR complex data images using the OMP algorithm, and transform the attribute scattering center parameters into a scattering feature visualization image;

[0008] The scattering feature visualization image and the SAR complex data image are input into an improved cyclic consistent generative adversarial network, and the generated image is output.

[0009] The original image is acquired, a loss function is used to calculate the multiple loss between the original image and the generated image, and the multiple loss is used to guide the improved recurrent consistent generative adversarial network to learn the scattering law, thereby obtaining the target generative adversarial network;

[0010] SAR target data is generated through the target generation adversarial network.

[0011] In one embodiment, the improved cyclic consistent generative adversarial network is a generative network consisting of an encoder, a converter, and a decoder;

[0012] The scattering feature visualization image and the SAR complex data image are input into an improved cyclic consistent generative adversarial network, and the generated image is output, including:

[0013] The scattering feature visualization image and the SAR complex data image are input into an improved cyclic consistent generative adversarial network, and the encoder extracts the fused scattering features.

[0014] The fused scattering features are converted by the converter, retaining both shallow and deep features, to obtain the converted fused scattering features;

[0015] The decoder maps the converted fused scattering feature distribution back to the data image, and outputs the generated image.

[0016] In one embodiment, the encoder includes an initial convolution module and a downsampling module; wherein,

[0017] The initial convolution module performs axisymmetric mirroring on the input image around its perimeter and uses convolution kernels to increase the number of channels; the downsampling module includes a convolutional layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer.

[0018] In one embodiment, the decoder includes an upsampling module and an output module; wherein,

[0019] The upsampling module includes a deconvolution layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer; the output module includes a reflection fill layer, a convolutional layer, and a Tanh activation layer.

[0020] In one embodiment, a converter is provided between the encoder and the decoder, the converter including a residual module.

[0021] In one embodiment, the improved cyclic consistent generative adversarial network further includes a discriminator;

[0022] The discriminator is used to divide the input image into blocks and to distinguish between real and fake blocks.

[0023] The discriminator includes a convolutional module and an output layer.

[0024] In one embodiment, the attribute scattering center parameters of the SAR complex data image are extracted using the OMP algorithm, and the attribute scattering center parameters are converted into a scattering feature visualization image, including:

[0025] Based on the SAR complex data image, the sum of the responses of each scattering center is calculated using the OMP algorithm;

[0026] The parameters of each scattering center are obtained by summing the responses of each scattering center.

[0027] The position parameters of the attribute scattering center parameters are extracted, and the remaining attribute scattering center parameters are visualized as scattering feature images based on the position parameters.

[0028] In one embodiment, the loss function includes adversarial loss, cycle consistency loss, and identity loss.

[0029] A SAR target data generation system with optimized scattering characteristics, the system comprising:

[0030] The image conversion module is used to acquire SAR complex data images, extract the attribute scattering center parameters of the SAR complex data images using the OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image.

[0031] An image generation module is used to input the scattering feature visualization image and the SAR complex data image into an improved cyclic consistent generative adversarial network and output a generated image.

[0032] A generative adversarial network (GAN) adjustment module is used to acquire the original image, calculate the multiple loss between the original image and the generated image using a loss function, and use the multiple loss to guide the improved cyclic consistent GAN to learn the scattering law, thereby obtaining the target GAN.

[0033] The data generation module is used to generate SAR target data through the target generation adversarial network.

[0034] The aforementioned SAR target data generation method and system with optimized scattering features extracts attribute scattering center parameters using the OMP algorithm and transforms them into a scattering feature visualization image, thereby obtaining a generated image that makes the physical scattering features of the generated image more realistic. By using an improved cyclic consistent generative adversarial network to learn scattering rules, the image quality can be optimized, making the scattering feature distribution of the generated image more similar to that of the real image. Attached Figure Description

[0035] Figure 1 This is an application environment diagram of a SAR target data generation method that integrates scattering feature optimization in one embodiment;

[0036] Figure 2 This is a flowchart illustrating a SAR target data generation method based on fused scattering feature optimization in one embodiment.

[0037] Figure 3 Visualize the scattering characteristics of the image;

[0038] Figure 4 This is a schematic diagram of the generator network structure in one embodiment;

[0039] Figure 5 This is a schematic diagram of the residual module structure in one embodiment;

[0040] Figure 6 This is a schematic diagram of the discriminator network structure in one embodiment;

[0041] Figure 7 This is a flowchart of a SAR target data generation method based on fused scattering feature optimization in one embodiment.

[0042] Figure 8 This is a block diagram of a SAR target data generation system that integrates scattering feature optimization in one embodiment.

[0043] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] The SAR target data generation method with optimized fusion scattering features provided in this application can be applied to, for example... Figure 1 The application environment shown. For example... Figure 1 As shown, the application environment includes computer device 110. Computer device 110 can acquire SAR complex data images, extract attribute scattering center parameters from the SAR complex data images using the OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image. Computer device 110 can input the scattering feature visualization image and the SAR complex data image into an improved recurrent consistent generative adversarial network (RCA), and output a generated image. Computer device 110 can acquire the original image, calculate the multiple loss between the original image and the generated image using a loss function, and use the multiple loss to guide the improved RCA to learn scattering patterns, thus obtaining a target generative adversarial network (PGA). Computer device 110 can generate SAR target data through the target GPA. The computer device 110 can be, but is not limited to, various personal computers, laptops, robots, unmanned aerial vehicles, etc.

[0046] In one embodiment, such as Figure 2 As shown, a method for generating SAR target data by fusing scattering feature optimization is provided, including the following steps:

[0047] Step 202: Obtain SAR complex data images, extract attribute scattering center parameters of SAR complex data images using the OMP algorithm, and convert attribute scattering center parameters into scattering feature visualization images.

[0048] Computer equipment can acquire SAR complex data images and extract the attributed scattering centers (ASCs) of SAR targets from these images using a physical model method based on orthogonal matching pursuit (OMP). Considering the complexity of the ASC parameters and the existence of estimation errors, this embodiment optimizes the fusion method of the ASCs. Utilizing their position and other parameters, they are converted into a scattering feature visualization image. This visualization image is then fused to guide SAR target generation, resulting in a more realistic physical scattering feature in the generated image.

[0049] Specifically, in one embodiment, a method for generating SAR target data with optimized fusion scattering features may further include the process of extracting attribute scattering center parameters and converting them into a visualization image. The specific process includes: calculating the sum of responses of each scattering center based on the SAR complex data image using the OMP algorithm; obtaining attribute scattering center parameters based on the sum of responses of each scattering center; extracting the position parameters from the attribute scattering center parameters; and visualizing the remaining attribute scattering center parameters as a scattering feature image based on the position parameters.

[0050] In this embodiment, the computer device can use the OMP algorithm to estimate the SAR target ASC parameters. The OMP algorithm achieves accurate estimation of sparse parameter vectors by iteratively selecting the basis vectors most relevant to the residuals and updating the support set and residuals. Therefore, the OMP method is superior to other methods in terms of computational efficiency, accuracy and robustness, and the final scattering center result is more accurate.

[0051] Specifically, based on the theory of attribute scattering centers, the backscattering of a target can be approximated as the sum of the responses of each scattering center: Where i represents the i-th scattering center, and f represents the radar frequency. Let θ represent the azimuth angle, q represent the number of scattering centers contained in the target, and the parameter set Θ = {θ} i}(i=1,2,...,q); For a single scattering center point, it can be represented as:

[0052]

[0053] Among them, f c Where C is the radar center frequency, and C is the electromagnetic wave propagation speed; parameters x i ,y i Indicates the azimuth and distance directions, A i Indicates amplitude, α i Indicates frequency dependence, L i and γ represents length and direction, respectively. i This indicates the directional dependency of the local ASC, and j represents the parameter.

[0054] The OMP algorithm achieves accurate estimation of sparse parameter vectors by iteratively selecting the basis vectors most relevant to the residuals and updating the support set and residuals. Because γ i The impact on the generative model is relatively small, so only the remaining 6 parameters of ASC are considered. This is based on the position parameter x of each scattering center. i ,y i The remaining four parameters are visualized as four electromagnetic scattering feature images. The visualization images converted from the parameter matrix are as follows: Figure 3As shown, two types of targets are illustrated: the M1 main battle tank and the M60 main battle tank. (a) is a real SAR image, (b) is an electromagnetic scattering feature image with amplitude parameter, (c) is an electromagnetic scattering feature image with frequency-dependent parameter, (d) is an electromagnetic scattering feature image with length parameter, and (e) is an electromagnetic scattering feature image with azimuth parameter. By jointly comparing these four electromagnetic scattering feature images, the estimation error in actual calculations is reduced, and the fusion method of attribute scattering centers is optimized.

[0055] Step 204: Input the scattering feature visualization image and the SAR complex data image into the improved cyclic consistent generative adversarial network, and output the generated image.

[0056] Next, the computer device can concatenate the scattering feature visualization image with the SAR complex data image, using them together as input to the generator, which is then an improved cyclic consistent generative adversarial network.

[0057] In one embodiment, the provided SAR target data generation method with fused scattering feature optimization may further include an improved cyclic consistent generative adversarial network process. Specifically, the improved cyclic consistent generative adversarial network is a generative network composed of an encoder, a converter, and a decoder.

[0058] In one embodiment, a method for generating SAR target data with optimized fusion scattering features may further include a process of outputting a generated image. The specific process includes: inputting a scattering feature visualization image and a SAR complex data image into an improved cyclic consistent generative adversarial network, extracting fusion scattering features through an encoder; converting the fusion scattering features through a converter, retaining shallow and deep features to obtain the converted fusion scattering features; and mapping the converted fusion scattering feature distribution back to the data image through a decoder to output the generated image.

[0059] To enhance the ability of Cyclic Generative Adversarial Networks (CGNs) to represent SAR scattering features, scattering feature visualization images and SAR complex data images can be cascaded, thereby optimizing the fusion of scattering features. Specifically, an improved CycleGAN is used to construct a new generative model. In addition to adversarial loss, style transfer loss and style consistency loss are added to guide the model in learning the imaging patterns of SAR targets. During data generation, the source and target domains of the generative network are the MSTAR dataset and the SAMPLE dataset, respectively.

[0060] The improved Cyclic Consistent Generative Adversarial Network (CGN) can be divided into three parts: an encoder, a transformer, and a decoder. The encoder extracts the fused features from the input image data, the transformer transforms the fused features while retaining both shallow and deep features, and the decoder is responsible for mapping the transformed feature distribution back to the target image. The generator learns the transformation relationship between MSTAR data and SAMPLE data, and finally generates SAR target simulation data that conforms to the SAMPLE data distribution.

[0061] like Figure 4 As shown, in one embodiment, the encoder includes an initial convolution module and a downsampling module; wherein, the initial convolution module performs axisymmetric mirroring on the input image around its perimeter and uses a convolution kernel to increase the number of channels; the downsampling module includes a convolutional layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer.

[0062] In one embodiment, such as Figure 4 As shown, the decoder includes an upsampling module and an output module; the upsampling module includes a deconvolution layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer; the output module includes a reflection padding layer, a convolutional layer, and a Tanh activation layer.

[0063] Specifically, such as Figure 4 As shown, the encoder consists of an initial convolutional module and two downsampling modules; the decoder consists of two upsampling modules and an output module. The initial convolutional module includes a reflection padding layer, a convolutional layer, an instance normalization layer, and a linear rectified function.

[0064] In one embodiment, a converter is provided between the encoder and decoder, and the converter includes residual modules. That is, in this embodiment, to enable the network to learn more complex features, a converter section is added to the network. The converter consists of multiple residual modules, such as... Figure 5 As shown, the residual module includes two reflection fill layers, two convolutional layers, two instance normalization layers, and a linear rectified function.

[0065] The scattering feature distributions of SAR target images vary significantly across different categories. During training, CycleGAN frequently encounters discrepancies between the target category of the generated SAR image and the target category of the input image. To address this issue, this embodiment constructs a cascaded feature input for the network by combining the SAR target image and its corresponding scattering feature image. This allows the network to learn richer feature representations, distinguishing feature differences between different target categories and helping to maintain target category consistency when generating images. The input image size is set to 128×128.

[0066] The encoder's initial convolutional module and downsampling module have different structures. The initial convolutional module is designed to enable the network to effectively extract low-level features from the input image while preserving edge information and providing appropriate feature representations for subsequent network layers. To preserve edge information, the initial convolutional module first performs axisymmetric mirroring on the input image; then, it uses a 7×7 kernel with a stride of 1 to increase the number of channels; finally, it passes through an instance normalization layer and a linear rectified function to obtain the low-level features of the image. The downsampling module consists of a convolutional layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer. The convolutional kernel size is 3×3 with a stride of 2. The dropout regularization layer is introduced in this embodiment to improve the model's generalization ability and prevent overfitting. The downsampling module gradually reduces the spatial resolution of the feature map while increasing its depth, effectively extracting high-level features from the image.

[0067] The converter transforms the extracted feature vectors into feature vectors of the target domain image while maintaining the scale of the feature vectors, which can increase the network depth and optimize network gradient propagation.

[0068] The decoder's upsampling module is obtained by replacing the convolutional layers of the encoder's downsampling module with corresponding deconvolutional layers. The deconvolutional layers also use a 3×3 kernel with a stride of 2. The decoder's output module consists of a reflection fill layer, a convolutional layer with a 7×7 kernel and a stride of 1, and a Tanh activation layer. The reflection fill layer in the output module is used to maintain the coherence and integrity of the SAR image content.

[0069] In one embodiment, the improved recurrent consistent generative adversarial network further includes a discriminator; the discriminator is used to divide the input image into blocks and distinguish between real and fake blocks; the discriminator includes a convolutional module and an output layer.

[0070] The discriminator network can be used to distinguish between multidimensional features of real data and generated data. The discriminator divides the input image into smaller blocks and classifies each block as real or fake, which helps enhance the model's ability to capture local details, improve training efficiency, and increase model stability. Figure 6As shown, the discriminator consists of three convolutional modules and one output layer. The three convolutional modules differ in that the first module does not incorporate normalization to preserve the original distribution of input features and avoid sample oscillations. Furthermore, each convolutional module employs normalization and dropout regularization operations on a per-sample basis, making the network training process more stable and improving the model's generalization ability. The output layer uses only a single convolutional operation, ensuring effective gradient propagation while performing local discrimination. This simplifies the discrimination task, better captures local and texture features of SAR target images, achieves fine-grained discrimination of target image features, and improves the network's computational efficiency.

[0071] During network training, the discriminator learns to distinguish between real sample images and generated SAR images by continuously optimizing its parameters. This provides feedback to the generator, prompting it to continuously optimize and generate high-quality data that better matches the distribution of the sample data. Simultaneously, the discriminator constrains the generator to improve the realism of local details and texture structure in the generated images. Ultimately, through the optimization of the discriminator, the generated data becomes visually and statistically closer to the sample data.

[0072] In this embodiment, by improving the generator model in CycleGAN, a converter is added between the encoder and decoder. By utilizing multiple residual modules, the network gradient propagation is optimized, which helps to simultaneously preserve shallow and deep features and distinguish feature differences between different categories.

[0073] Step 206: Obtain the original image, calculate the multiple loss between the original image and the generated image using the loss function, and use the multiple loss to guide the improved recurrent consistent generative adversarial network to learn the scattering law, thus obtaining the target generative adversarial network.

[0074] The loss function measures the difference between the generated image and the original image. Building upon adversarial loss, the loss function incorporates style transfer loss and style consistency loss to guide the model in learning the imaging patterns of SAR targets, optimizing image quality, maintaining target category consistency, and improving overall generation performance.

[0075] In one embodiment, the loss function includes adversarial loss, cycle consistency loss, and identity loss. Adversarial loss improves the quality and realism of the generated image; cycle consistency loss ensures that the target category of the image remains unchanged during the transformation process; and identity loss ensures that the generated image remains consistent with the input image when the input image already belongs to the target domain.

[0076] Specifically, adversarial loss is a crucial constraint that ensures the generator and discriminator maintain a game-like adversarial relationship during training, ultimately reaching an equilibrium state. Adversarial loss incentivizes the generator to produce realistic images while simultaneously helping the discriminator correctly distinguish between real and generated images, making the statistical grayscale distribution of the generated images closer to that of the real data. To improve the realism of the scattering feature distribution of the generated SAR images, this embodiment constructs an image scattering feature similarity loss based on the scattering center parameters of the SAR image, in addition to fusing scattering features, to guide the model in better learning the target's scattering mechanism. The construction process of adversarial loss can be specifically represented as follows:

[0077] in, and L represents the four attribute scattering center parameters of a pair of real and generated data. A ,L α ,L L , L represents the absolute error loss of the four attribute scattering center parameters respectively. DIF-ASC This is a weighted loss function summed from four sub-loss functions, with weights for each parameter being β1 = 4, β2 = 2, β3 = β4 = 1. Based on this, generator and discriminator losses can be constructed, as shown below:

[0078] in, For G Y The generator's loss function D Y The loss function of the discriminator For G X The generator's loss function D X The loss function of the discriminator, where E represents the expectation and P... data (x) represents the statistical gray distribution of sample X, P data (y) represents the statistical gray-level distribution of sample Y. The final adversarial loss can be expressed as:

[0079] Cycle consistency loss is the core component of CycleGAN. It enables CycleGAN to complete image transformation tasks without paired training data, while maintaining consistency in the structure and content of the images during the transformation process. During network training, the source domain image X is transformed into the target domain image Y. fake Then convert it into the source domain image X rec The target domain image Y is transformed twice to obtain the reconstructed image Y. recTo establish a symmetric relationship between the source and target domains, the reconstructed image should approximate the original image. Therefore, the specific representation of the cycle consistency loss is as follows:

[0080] The purpose of identity loss is to enhance the generator's ability to maintain image style consistency. Identity loss ensures that when the input image already belongs to the target domain, the generated image is as consistent as possible with the input image. Specifically, it is expressed as follows:

[0081] In summary, in this embodiment, the final loss of the network can be expressed as: L total =L adv +λ1L cycle +λ2L id ; where λ1 and λ2 are the weights of the cycle consistency loss and the identity loss, respectively.

[0082] Step 208: Generate SAR target data using a target generation adversarial network.

[0083] In one embodiment, a method for generating SAR target data by fusing scattering features and optimizing them is provided, such as... Figure 7 As shown: First, the OMP algorithm is used to extract the attribute scattering center parameters of SAR complex data images and convert them into scattering feature visualization images. This method is computationally efficient and robust, and the final scattering center results are more accurate. Second, to improve the model's ability to represent SAR scattering features, the scattering feature visualization images and SAR complex data images are cascaded, thereby optimizing the fusion of scattering features. Finally, CycleGAN is improved to construct a new generative model, which includes a generator and a discriminator. In addition to adversarial loss, style transfer loss and style consistency loss are added to guide the model in learning the imaging patterns of SAR targets. During the data generation process, the source domain and target domain of the generative network are the MSTAR dataset and the SAMPLE dataset, respectively.

[0084] Among them, G Y and G X For generator; D Y and D X For discriminator; Y fake and Y rec Generated and reconstructed images conforming to the distribution of the SAMPLE dataset; X fake and X rec Generate and reconstruct images that conform to the distribution of the MSTAR dataset.

[0085] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0086] In one embodiment, such as Figure 8 As shown, a SAR target data generation system with optimized scattering features is provided, comprising: an image conversion module 810, an image generation module 820, a generative adversarial network adjustment module 830, and a data generation module 840, wherein:

[0087] The image conversion module 810 is used to acquire SAR complex data images, extract the attribute scattering center parameters of the SAR complex data images through the OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image.

[0088] Image generation module 820 is used to input the scattering feature visualization image and the SAR complex data image into an improved cyclic consistent generative adversarial network and output the generated image.

[0089] Generative Adversarial Network Adjustment Module 830 is used to acquire the original image, calculate the multiple loss between the original image and the generated image using a loss function, and use the multiple loss to guide the improved Cyclic Consistent Generative Adversarial Network to learn the scattering law, thereby obtaining the target Generative Adversarial Network.

[0090] The data generation module 840 is used to generate SAR target data through a target generation adversarial network.

[0091] In one embodiment, the improved Cyclic Consistent Generative Adversarial Network (CGIRN) is a generative network consisting of an encoder, a converter, and a decoder. The image generation module 820 is further configured to input the scattering feature visualization image and the SAR complex data image into the improved CGIRN, extract the fused scattering features through the encoder, transform the fused scattering features through the converter, retaining both shallow and deep features to obtain the transformed fused scattering features, and map the transformed fused scattering feature distribution back to the data image through the decoder to output the generated image.

[0092] In one embodiment, the encoder includes an initial convolution module and a downsampling module; wherein, the initial convolution module performs axisymmetric mirroring on the input image around its perimeter and uses a convolution kernel to increase the number of channels; the downsampling module includes a convolutional layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer.

[0093] In one embodiment, the decoder includes an upsampling module and an output module; wherein the upsampling module includes a deconvolution layer, an instance normalization layer, a linear rectified function, and a dropout regularization layer; and the output module includes a reflection padding layer, a convolutional layer, and a Tanh activation layer.

[0094] In one embodiment, a converter is provided between the encoder and the decoder, and the converter includes a residual module.

[0095] In one embodiment, the improved recurrent consistent generative adversarial network further includes a discriminator; the discriminator is used to divide the input image into blocks and distinguish between real and fake blocks; the discriminator includes a convolutional module and an output layer.

[0096] In one embodiment, the image conversion module 810 is further configured to calculate the sum of responses of each scattering center based on the SAR complex data image using the OMP algorithm; obtain the parameters of each attribute scattering center based on the sum of responses of each scattering center; extract the position parameters from the attribute scattering center parameters; and visualize the remaining attribute scattering center parameters as a scattering feature image based on the position parameters.

[0097] In one embodiment, the loss function includes adversarial loss, cycle consistency loss, and identity loss.

[0098] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for generating SAR target data with optimized fused scattering characteristics. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0099] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0100] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a SAR target data generation method with optimized fused scattering features.

[0101] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program being executed by a processor to implement the steps of a SAR target data generation method with optimized fused scattering features.

[0102] Those skilled in the art will understand that all or part of the processes in the methods of 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, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0103] 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.

[0104] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for generating SAR target data with optimized scattering characteristics, characterized in that, The method comprises: acquiring a SAR complex data image, extracting attribute scattering center parameters of the SAR complex data image through an OMP algorithm, and converting the attribute scattering center parameters into a scattering feature visualization image; an improved cycle-consistent generative adversarial network is a generative network composed of an encoder, a converter and a decoder; the converter is arranged between the encoder and the decoder, and the converter comprises a residual module; the scattering feature visualization image and the SAR complex data image are input into the improved cycle-consistent generative adversarial network, the scattering feature visualization image and the SAR complex data image are concatenated, and the two are jointly used as an input part of a generator to output a generated image, comprising: inputting the scattering feature visualization image and the SAR complex data image into the improved cycle-consistent generative adversarial network, extracting fused scattering features through the encoder; converting the fused scattering features through the converter to retain shallow features and deep features, and obtaining converted fused scattering features; and distributing the converted fused scattering features back to a data image through the decoder to output a generated image; acquiring an original image, calculating multiple losses between the original image and the generated image by using a loss function, and guiding the improved cycle-consistent generative adversarial network to learn scattering rules by using the multiple losses to obtain a target generative adversarial network; generating SAR target data through the target generative adversarial network.

2. The method of claim 1, wherein the method further comprises: The encoder comprises an initial convolution module and a down-sampling module; wherein the initial convolution module performs four-around axisymmetric mirror filling on an input image, and expands the number of channels by using a convolution kernel; the down-sampling module comprises a convolution layer, an instance normalization layer, a linear rectifier function and a dropout regularization layer.

3. The method of claim 1, wherein the method further comprises: The decoder comprises an up-sampling module and an output module; wherein the up-sampling module comprises an inverse convolution layer, an instance normalization layer, a linear rectifier function and a dropout regularization layer; and the output module comprises a reflection padding layer, a convolution layer and a Tanh activation layer.

4. The method of claim 1, wherein the method further comprises: The improved cycle-consistent generative adversarial network further comprises a discriminator; the discriminator is used to divide an input image into blocks and distinguish the divided blocks as true or false; the discriminator comprises a convolution module and an output layer.

5. The method of claim 1, wherein the method further comprises: extracting attribute scattering center parameters of the SAR complex data image through an OMP algorithm, and converting the attribute scattering center parameters into a scattering feature visualization image, comprising: based on the SAR complex data image, calculating the sum of responses of each scattering center by using an OMP algorithm; obtaining each attribute scattering center parameter according to the sum of responses of each scattering center; extracting a position parameter in the attribute scattering center parameter, and visualizing the remaining attribute scattering center parameters into a scattering feature image according to the position parameter.

6. The method of claim 1, wherein the method further comprises: The loss function comprises an adversarial loss, a cycle consistency loss and an Identity loss.

7. A system for generating SAR target data with optimized fusion scattering characteristics, the system comprising: a SAR target data generator configured to generate SAR target data; and a fusion scattering characteristics optimizer configured to optimize the fusion scattering characteristics of the SAR target data. The system comprises: An image conversion module is configured to acquire a SAR complex data image, extract attribute scattering center parameters of the SAR complex data image through an OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image; An image generation module is configured to input the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, cascade the scattering feature visualization image and the SAR complex data image, and take the scattering feature visualization image and the SAR complex data image as input parts of a generator to output a generated image. The improved cycle-consistent generative adversarial network is a generative network composed of an encoder, a converter and a decoder. The converter is arranged between the encoder and the decoder, and the converter includes a residual module. The image generation module is further configured to input the scattering feature visualization image and the SAR complex data image into the improved cycle-consistent generative adversarial network, extract fused scattering features through the encoder, convert the fused scattering features through the converter to retain shallow features and deep features, obtain converted fused scattering features, and distribute the converted fused scattering features back to a data image through the decoder to output the generated image. A generative adversarial network adjustment module is configured to acquire an original image, calculate multiple losses between the original image and the generated image by using a loss function, and guide the improved cycle-consistent generative adversarial network to learn scattering rules by using the multiple losses to obtain a target generative adversarial network. A data generation module is configured to generate SAR target data through the target generative adversarial network.

Images