SAR (Synthetic Aperture Radar) target data generation method and system fused with scattering feature optimization

Abstract

The invention relates to an SAR (Synthetic Aperture Radar) target data generation method and system fused with scattering feature optimization. The method comprises the following steps: extracting an attribute scattering center parameter through an OMP algorithm, and converting the attribute scattering center parameter into a scattering feature visual image; inputting the scattering feature visualization image and the SAR complex data image into an improved cyclic consistent generative adversarial network, and outputting a generated image; and calculating multiple losses between the original image and the generated image by using a loss function, and learning a scattering law by using the multiple losses to guide the improved cyclic consistent generative adversarial network so as to generate SAR target data. An attribute scattering center parameter is extracted through an OMP algorithm and converted into a scattering feature visual image, and then a generated image is obtained, so that the physical scattering feature of the generated image is more real; the improved cyclic consistent generative adversarial network is used to learn the scattering law, so that the image quality can be optimized, and the scattering feature distribution of the generated image is more similar to that of a real image.

Description

Technical Field

[0001] The present invention relates to the technical field of synthetic aperture radar data generation, and in particular to a SAR target data generation method and system integrating scattering feature optimization. Background Art

[0002] Synthetic Aperture Radar (SAR) has become a very important means of earth observation due to its unique imaging mechanism and is widely used in military and civilian fields. With the development of artificial intelligence, deep learning has been widely used in SAR interpretation research. Due to the high cost of SAR data acquisition, the data sets 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 are divided into two categories, one is the data enhancement method using geometric transformation, and the other is the image generation method based on data modeling; deep learning methods based on generative models mainly implement data augmentation based on restricted boltzmann machines (RBM), variational auto-encoder (VAE) and generative adversarial network (GAN).

[0003] However, data enhancement methods based on geometric transformation are usually implemented by image rotation, mirroring and other methods, without considering the imaging mechanism of radar, and simply increasing the diversity of image data; the data modeling method is to use computer-aided design (CAD) software to accurately model the target, and then obtain the target SAR image through electromagnetic calculation. However, building a target model not only requires expert knowledge, but also requires continuous correction of the model to ensure the quality of simulated SAR images. This method consumes a lot of manpower and time resources, and it is difficult to simulate the real environment of the target. Deep learning methods based on generative models also have certain limitations: RBM and VAE are not able to capture image details, resulting in low quality of generated SAR images, and the structural information of the target is quite different from the real data; traditional GAN ​​does not combine the scattering information of SAR imaging, resulting in the inability of simulated data to effectively represent the scattering characteristics of real targets.

[0004] Therefore, traditional SAR image data expansion methods often have problems with poor data generation quality and deviation from the actual scattering law, resulting in low image quality. Summary of the invention

[0005] Based on this, in order to solve the above technical problems, a SAR target data generation method and system integrating scattering feature optimization are provided, which can improve the quality of generated images and make the scattering feature distribution of the generated images more similar to the real images.

[0006] A SAR target data generation method integrating scattering feature optimization, the method comprising:

[0007] Acquire a SAR complex data image, extract attribute scattering center parameters of the SAR complex data image by using an OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image;

[0008] Inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and outputting a generated image;

[0009] Acquire an original image, use a loss function to calculate multiple losses between the original image and the generated image, and use the multiple losses to guide the improved cycle-consistent generative adversarial network to learn a scattering law to obtain a target generative adversarial network;

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

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

[0012] Inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and outputting a generated image, including:

[0013] Inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and extracting fused scattering features through the encoder;

[0014] The fused scattering feature is converted by the converter, shallow features and deep features are retained, and a converted fused scattering feature is obtained;

[0015] The decoder maps the converted fused scattering feature distribution back to the data image and outputs a 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 mirror filling on all sides of the input image and uses a convolution kernel to expand the number of channels; the downsampling module includes a convolution layer, an instance normalization layer, a linear rectification function, and a discard 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 rectification function, and a discard regularization layer; the output module includes a reflection filling layer, a convolution layer, and a Tanh activation layer.

[0020] In one of the embodiments, a converter is provided between the encoder and the decoder, and the converter includes a residual module.

[0021] In one of the embodiments, the improved cycle-consistent generative adversarial network further includes a discriminator;

[0022] The discriminator is used to divide the input image into blocks and perform true and false discrimination on the divided blocks;

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

[0024] In one embodiment, extracting attribute scattering center parameters of the SAR complex data image by using an OMP algorithm, and converting the attribute scattering center parameters into a scattering feature visualization image, includes:

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

[0026] According to the sum of the responses of each scattering center, the parameters of each attribute scattering center are obtained;

[0027] The position parameters in the attribute scattering center parameters are extracted, and the remaining attribute scattering center parameters are visualized as a scattering feature image according to the position parameters.

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

[0029] A SAR target data generation system integrating scattering feature optimization, the system comprising:

[0030] An image conversion module is used to obtain 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;

[0031] An image generation module, used for inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and outputting a generated image;

[0032] A generative adversarial network adjustment module is used to obtain an original image, calculate multiple losses between the original image and the generated image using a loss function, and use the multiple losses to guide the improved cycle-consistent generative adversarial network to learn a scattering law to obtain a target generative adversarial network;

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

[0034] The above-mentioned SAR target data generation method and system with integrated scattering feature optimization extracts the attribute scattering center parameters through the OMP algorithm and converts them into a scattering feature visualization image, thereby obtaining a generated image, so that the physical scattering characteristics of the generated image are more realistic; using an improved cycle-consistent generative adversarial network to learn the scattering law, the image quality can be optimized, so that the scattering feature distribution of the generated image is more similar to that of the real image. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A diagram of an application environment of a SAR target data generation method for fusion scattering feature optimization in one embodiment;

[0036] Figure 2 A schematic diagram of a flow chart of a method for generating SAR target data by integrating scattering feature optimization in one embodiment;

[0037] Figure 3 Visualize images for scattering features;

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

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

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

[0041] Figure 7 A flow chart of a method for generating SAR target data by fusing scattering feature optimization in one embodiment;

[0042] Figure 8 A structural block diagram of a SAR target data generation system for fusion scattering feature optimization in one embodiment;

[0043] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] The SAR target data generation method provided by the embodiment of the present application for fusion scattering feature optimization can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can obtain a SAR complex data image, extract the attribute scattering center parameters of the SAR complex data image through the OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image; the computer device 110 can input the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and output a generated image; the computer device 110 can obtain an original image, use a loss function to calculate multiple losses between the original image and the generated image, and use multiple losses to guide the improved cycle-consistent generative adversarial network to learn the scattering law, and obtain a target generative adversarial network; the computer device 110 can generate SAR target data through the target generative adversarial network. Among them, the computer device 110 can be, but is not limited to, various personal computers, laptops, robots, unmanned aerial vehicles and other devices.

[0046] In one embodiment, Figure 2 As shown, a SAR target data generation method with fused scattering feature optimization is provided, comprising the following steps:

[0047] Step 202: Acquire a SAR complex data image, extract attribute scattering center parameters of the SAR complex data image by using an OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image.

[0048] The computer device can obtain a SAR complex data image, and extract the attributed scattering center (ASC) of the SAR target from the SAR complex data image using a physical model method based on orthogonal matching pursuit (OMP). Considering the complexity of the attribute scattering center parameters and the existence of estimation errors, the fusion method of the attribute scattering center is optimized in this embodiment, and its position and other parameters are used to convert it into a scattering feature visualization image, and then the scattering feature visualization image is fused to guide the generation of the SAR target, so that the physical scattering characteristics of the generated image are more realistic.

[0049] Specifically, in one embodiment, a SAR target data generation method for fused scattering feature optimization is provided, which may also include a process of extracting attribute scattering center parameters and converting them into a visualized image. The specific process includes: based on the SAR complex data image, using the OMP algorithm to calculate the sum of the responses of each scattering center; according to the sum of the responses of each scattering center, obtaining each attribute scattering center parameter; extracting the position parameters in the attribute scattering center parameters, and according to the position parameters, visualizing the remaining attribute scattering center parameters as a scattering feature image.

[0050] In this embodiment, the computer device can use the OMP algorithm to estimate the ASC parameters of the SAR target. The OMP algorithm achieves accurate estimation of the sparse parameter vector by iteratively selecting the basis vector most relevant to the residual and updating the support set and the residual. 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 attribute scattering center theory, the backscattering of the target can be approximated as the sum of the responses of each scattering center: Where i represents the i-th scattering center, f represents the radar frequency, represents the azimuth, q represents 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 expressed as:

[0052]

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

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

[0055] Step 204: input the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and output a generated image.

[0056] Then, the computer device can concatenate the scattering feature visualization image and the SAR complex data image as input to the generator, which is an improved cycle-consistent generative adversarial network.

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

[0058] In one embodiment, a SAR target data generation method for fused scattering feature optimization is provided, which may also include a process of outputting a generated image, and the specific process includes: inputting a scattering feature visualization image and a SAR complex data image into an improved cycle-consistent generative adversarial network, and extracting a fused scattering feature through an encoder; converting the fused scattering feature through a converter, retaining shallow features and deep features, and obtaining a converted fused scattering feature; and mapping the converted fused scattering feature distribution back to the data image through a decoder, and outputting a generated image.

[0059] In order to improve the characterization ability of the cycle-consistent generative adversarial network for SAR scattering features, the scattering feature visualization image and the SAR complex data image can be cascaded to optimize the fusion of scattering features. Among them, the improved cycle-consistent generative adversarial network CycleGAN constructs a new generation model. On the basis of the adversarial loss, style conversion loss and style consistency loss are added to guide the model to learn the imaging laws of SAR targets. In the data generation process, the source domain and target domain of the generation network are the MSTAR dataset and the SAMPLE dataset, respectively.

[0060] The improved cycle-consistent generative adversarial network can be divided into three parts: encoder, converter and decoder. The encoder extracts the fusion features of the input image data, the converter transforms the fusion features while retaining the shallow features and deep features, and the decoder is responsible for mapping the converted feature distribution back to the target image. The generator can learn the conversion relationship between MSTAR data and SAMPLE data, and finally generate 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 mirror padding on all sides of the input image and uses a convolution kernel to expand the number of channels; the downsampling module includes a convolution layer, an instance normalization layer, a linear rectification function, and a discard regularization layer.

[0062] In one embodiment, Figure 4 As shown, the decoder includes an upsampling module and an output module; wherein the upsampling module includes a deconvolution layer, an instance normalization layer, a linear rectification function, and a discard regularization layer; the output module includes a reflection filling layer, a convolution layer, and a Tanh activation layer.

[0063] Specifically, Figure 4 As shown in Figure 1, the encoder consists of an initial convolution module and two downsampling modules; the decoder consists of two upsampling modules and an output module. The initial convolution module includes a reflection padding layer, a convolution layer, an instance normalization layer, and a linear rectification function.

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

[0065] For SAR target images of different categories, their scattering feature distributions are very different. However, during the training process of CycleGAN, the target category of the generated SAR image is often inconsistent with the target category of the input image. To solve this problem, this embodiment combines the SAR target image and its corresponding scattering feature image to construct the cascade feature input of the network, so that the network can learn richer feature representations, distinguish the feature differences between targets of different categories, and help maintain the consistency of the target category when generating images. Among them, the size of the input image is set to 128×128.

[0066] The structures of the encoder's initial convolution module and downsampling module are different. The purpose of constructing the initial convolution module is to enable the network to effectively extract low-level features from the input image while maintaining the edge information of the image and providing appropriate feature representation for subsequent network layers. To maintain the edge information of the image, the initial convolution module first fills the input image with an axisymmetric mirror image on all sides; then a convolution kernel with a size of 7×7 and a step size of 1 is used to expand the number of channels; finally, the low-level features of the image are obtained through an instance normalization layer and a linear rectification function. The downsampling modules are composed of a convolution layer, an instance normalization layer, a linear rectification function, and a drop regularization layer, where: the convolution kernel size of the convolution layer is 3×3 and the step size is 2; the drop regularization layer is introduced into the method of this embodiment to improve the generalization ability of the model and prevent the network from overfitting. The downsampling module gradually reduces the spatial resolution of the feature map while increasing the depth of the feature map to effectively extract high-level features of the image.

[0067] The converter converts the extracted feature vector into the feature vector of the target domain image while ensuring that the scale of the feature vector remains unchanged, which can improve the network depth and optimize the network gradient propagation.

[0068] The upsampling module of the decoder is obtained by replacing the convolutional layer of the encoder downsampling module with the corresponding deconvolutional layer. The deconvolutional layer also uses a convolutional kernel of size 3×3 and stride 2. The output module of the decoder consists of a reflection padding layer, a convolutional layer with a convolutional kernel of size 7×7 and stride 1, and a Tanh activation layer. In the output module, the purpose of using the reflection padding layer is to maintain the coherence of the SAR image content and the integrity of the information.

[0069] In one embodiment, the improved cycle-consistent generative adversarial network also includes a discriminator; the discriminator is used to divide the input image into blocks and perform true and false discrimination on the divided blocks; the discriminator includes a convolution module and an output layer.

[0070] Among them, the discriminator network can be used to distinguish the difference between the multi-dimensional features of real data and generated data. The discriminator divides the input image into smaller blocks and distinguishes true from false for each block, which helps to enhance the model's capture of local details, improve training efficiency and model stability. Figure 6As shown in the figure, the discriminator consists of three convolution modules and an output layer. Among them, the three convolution modules of the discriminator are different. The first convolution module was designed without considering the normalization operation in order to retain the original distribution of the input features and avoid sample oscillation; in addition, the convolution modules all use normalization and discard regularization operations for a single sample, making the network training process more stable and improving the generalization of the model. The output layer is designed with only a single convolution operation, so that the network can complete local discrimination while ensuring the effective transfer of gradients, simplifying the discrimination task, and better capturing the local and texture features of the SAR target image, achieving fine-grained discrimination of the target image features, and improving the computational efficiency of the network.

[0071] During network training, the discriminator continuously optimizes parameters to learn to distinguish between real SAMPLE images and generated SAR images, thereby providing feedback to the generator, prompting it to continuously optimize and generate high-quality data that is more consistent with the distribution of SAMPLE data. At the same time, the discriminator constrains the generator to improve the authenticity of local details and texture structures of the generated images. Ultimately, through the optimization of the discriminator, the generated data is closer to the SAMPLE data in terms of visual and statistical characteristics.

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

[0073] Step 206, obtaining the original image, using the loss function to calculate the multiple losses between the original image and the generated image, and using the multiple losses to guide the improved cycle-consistent generative adversarial network to learn the scattering law, thereby obtaining the target generative adversarial network.

[0074] Among them, the loss function can be used to measure the difference between the generated image and the original image. On the basis of adversarial loss, the loss function adds style conversion loss and style consistency loss to guide the model to learn the imaging rules of SAR targets, optimize image quality and maintain target category consistency, thereby improving the overall generation effect.

[0075] In one embodiment, the loss function includes adversarial loss, cycle consistency loss, and identity loss. Among them, 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 conversion process; identity loss ensures that the generated image is consistent with the input image when the input image already belongs to the target domain image.

[0076] Specifically, adversarial loss is an important constraint to ensure that the generator and the discriminator maintain game confrontation during the training process and eventually reach an equilibrium state. Adversarial loss motivates the generator to generate realistic images while helping the discriminator to correctly distinguish between real and generated images, so that the generated images are closer to the statistical grayscale distribution of the real data. In order to improve the authenticity of the distribution of scattering features of the generated SAR image, this embodiment uses the scattering center parameters of the SAR image to construct the image scattering feature similarity loss on the basis of fusing scattering features, and guides the model to better learn the scattering mechanism of the target. The construction process of adversarial loss can be specifically expressed as:

[0077] in, and are the four attribute scattering center parameters of a pair of real and generated data, L A ,L α ,L L , Represents the absolute error loss of the four attribute scattering center parameters, L DIF-ASC is the weighted loss function of the four sub-loss functions, and the weights of each parameter loss are β1=4, β2=2, β3=β4=1. On this basis, the generator and discriminator losses can be constructed, which are specifically expressed as follows:

[0078] in, G Y The loss function of the generator, D Y The loss function of the discriminator is G X The loss function of the generator, D X The loss function of the discriminator, E represents expectation, P data (x) is the statistical grayscale distribution of sample X, P data (y) is the statistical grayscale distribution of sample Y. The final adversarial loss can be expressed as:

[0079] Cycle consistency loss is the core part of CycleGAN. Cycle consistency loss enables CycleGAN to complete image conversion tasks without paired training data. At the same time, during the image conversion process, the structure and content of the image should remain consistent. During the network training process, the source domain image X is converted to the target domain image Y. fake , and then converted to the source domain image X rec ; The target domain image Y is converted twice to obtain the reconstructed image Y recIn order to establish the symmetric relationship between the source domain and the target domain, the reconstructed image should be similar to the original image. Therefore, the specific expression of the cycle consistency loss is constructed as:

[0080] The purpose of constructing the identity loss is to assist in enhancing 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 expressed as:

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

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

[0083] In one embodiment, a SAR target data generation method for fusion scattering feature optimization is provided as follows: Figure 7 As shown in the figure: First, the attribute scattering center parameters of the SAR complex data image are extracted through the OMP algorithm and converted into a scattering feature visualization image, which has high computational efficiency and strong robustness, and the final scattering center result is more accurate; secondly, in order to improve the model's ability to characterize SAR scattering features, the cascade scattering feature visualization image and the SAR complex data image are used to optimize the fusion scattering features; finally, CycleGAN is improved to construct a new generation model, which includes a generator and a discriminator. On the basis of the adversarial loss, style conversion loss and style consistency loss are added to guide the model to learn the imaging rules of SAR targets. In the data generation process, the source domain and target domain of the generation network are the MSTAR dataset and the SAMPLE dataset, respectively.

[0084] Among them, G Y and G X is the generator; D Y and D X is the discriminator; Y fake and Y rec Generate and reconstruct images that conform 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 various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0086] In one embodiment, Figure 8 As shown, a SAR target data generation system with fused scattering feature optimization is provided, including: 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] An image conversion module 810 is used to obtain 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;

[0088] An image generation module 820 is used to input the scattering feature visualization image and the SAR complex data image into the improved cycle-consistent generative adversarial network and output a generated image;

[0089] A generative adversarial network adjustment module 830 is used to obtain an original image, calculate multiple losses between the original image and the generated image using a loss function, and use the multiple losses to guide the improved cycle-consistent generative adversarial network to learn a scattering law to obtain a 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 cycle-consistent generative adversarial network is a generative network composed of an encoder, a converter, and a decoder; the image generation module 820 is also used to input the scattering feature visualization image and the SAR complex data image into the improved cycle-consistent generative adversarial network, and extract the fused scattering features through the encoder; the fused scattering features are converted through the converter to retain the shallow features and the deep features to obtain the converted fused scattering features; the converted fused scattering feature distribution is mapped back to the data image through the decoder, and the generated image is output.

[0092] In one embodiment, the encoder includes an initial convolution module and a downsampling module; wherein the initial convolution module performs axisymmetric mirror padding on the input image on all sides and uses a convolution kernel to expand the number of channels; the downsampling module includes a convolution layer, an instance normalization layer, a linear rectification function, and a discard 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 rectification function, and a discard regularization layer; the output module includes a reflection filling layer, a convolution 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 cycle-consistent generative adversarial network also includes a discriminator; the discriminator is used to divide the input image into blocks and perform true and false discrimination on the divided blocks; the discriminator includes a convolution module and an output layer.

[0096] In one embodiment, the image conversion module 810 is also used to calculate the sum of the responses of each scattering center based on the SAR complex data image using the OMP algorithm; obtain each attribute scattering center parameter according to the sum of the responses of each scattering center; extract the position parameters in the attribute scattering center parameters, and visualize the remaining attribute scattering center parameters as a scattering feature image according to 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. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a SAR target data generation method with fused scattering feature optimization is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0099] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0100] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of a method for generating SAR target data with optimized fused scattering characteristics are implemented.

[0101] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for generating SAR target data with optimized fused scattering characteristics are implemented.

[0102] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database 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. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0103] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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 above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A SAR target data generation method integrating scattering feature optimization, characterized in that: The method comprises: Acquire a SAR complex data image, extract attribute scattering center parameters of the SAR complex data image by using an OMP algorithm, and convert the attribute scattering center parameters into a scattering feature visualization image; Inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and outputting a generated image; Acquire an original image, use a loss function to calculate multiple losses between the original image and the generated image, and use the multiple losses to guide the improved cycle-consistent generative adversarial network to learn a scattering law to obtain a target generative adversarial network; SAR target data is generated by the target generation adversarial network.

2. The SAR target data generation method of fused scattering feature optimization according to claim 1, characterized in that: The improved cycle-consistent generative adversarial network is a generative network composed of an encoder, a converter, and a decoder; Inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and outputting a generated image, including: Inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and extracting fused scattering features through the encoder; The fused scattering feature is converted by the converter, shallow features and deep features are retained, and a converted fused scattering feature is obtained; The decoder maps the converted fused scattering feature distribution back to the data image and outputs a generated image.

3. The SAR target data generation method of fused scattering feature optimization according to claim 2, characterized in that: The encoder includes an initial convolution module and a downsampling module; wherein, The initial convolution module performs axisymmetric mirror filling on all sides of the input image and uses a convolution kernel to expand the number of channels; the downsampling module includes a convolution layer, an instance normalization layer, a linear rectification function, and a discard regularization layer.

4. The SAR target data generation method of fused scattering feature optimization according to claim 2, characterized in that: The decoder includes an upsampling module and an output module; wherein, The upsampling module includes a deconvolution layer, an instance normalization layer, a linear rectification function, and a discard regularization layer; the output module includes a reflection filling layer, a convolution layer, and a Tanh activation layer.

5. The SAR target data generation method of fused scattering feature optimization according to claim 2, characterized in that: A converter is arranged between the encoder and the decoder, and the converter includes a residual module.

6. The SAR target data generation method of fused scattering feature optimization according to claim 2, characterized in that: The improved cycle-consistent generative adversarial network also includes a discriminator; The discriminator is used to divide the input image into blocks and perform true and false discrimination on the divided blocks; The discriminator includes a convolution module and an output layer.

7. The SAR target data generation method of fused scattering feature optimization according to claim 1, characterized in that: Extracting attribute scattering center parameters of the SAR complex data image by using an OMP algorithm, and converting the attribute scattering center parameters into a scattering feature visualization image, including: Based on the SAR complex data image, the sum of responses of each scattering center is calculated using the OMP algorithm; According to the sum of the responses of each scattering center, the parameters of each attribute scattering center are obtained; The position parameters in the attribute scattering center parameters are extracted, and the remaining attribute scattering center parameters are visualized as a scattering feature image according to the position parameters.

8. The SAR target data generation method of fused scattering feature optimization according to claim 1, characterized in that: The loss functions include adversarial loss, cycle consistency loss, and identity loss.

9. A SAR target data generation system integrating scattering feature optimization, characterized in that: The system comprises: An image conversion module is used to obtain 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, used for inputting the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and outputting a generated image; A generative adversarial network adjustment module is used to obtain an original image, calculate multiple losses between the original image and the generated image using a loss function, and use the multiple losses to guide the improved cycle-consistent generative adversarial network to learn a scattering law to obtain a target generative adversarial network; A data generation module is used to generate SAR target data through the target generation adversarial network.

10. The SAR target data generation system with fused scattering feature optimization according to claim 9, characterized in that: The improved cycle-consistent generative adversarial network is a generative network composed of an encoder, a converter, and a decoder; The image generation module is also used to: input the scattering feature visualization image and the SAR complex data image into an improved cycle-consistent generative adversarial network, and extract the fused scattering feature through the encoder; The fused scattering feature is converted by the converter, shallow features and deep features are retained, and a converted fused scattering feature is obtained; The decoder maps the converted fused scattering feature distribution back to the data image and outputs a generated image.

Images