SAR intelligent anti-interference method based on time-frequency domain processing

Abstract

The invention discloses an SAR intelligent anti-interference method based on time-frequency domain processing, and belongs to the field of radar signal processing. For the problems of complex and changeable electromagnetic interference environments and the like, the method selects an auto-encoder model to construct a deep learning model based on a time-frequency domain, and realizes interference detection on SAR echo samples and interference removal. According to the method, time-frequency domain signal processing and the deep learning technology are combined, the data driving advantages of the time-frequency domain signal processing and the deep learning technology are fused into the SAR anti-interference field, the strong data driving advantages of the deep learning technology are utilized, interference characteristics are mined from mass data, and then interference is effectively removed. According to the application of the method, the anti-interference capability of the SAR under the complex electromagnetic interference condition is remarkably improved, and intelligent anti-interference of the SAR is achieved.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing, and in particular to a SAR intelligent anti-jamming method based on time-frequency domain processing. Background Art

[0002] Synthetic Aperture Radar (SAR) has been widely used in civilian applications such as resource remote sensing, geographical mapping, war zone detection, and precision guidance, and has become an indispensable key technology. However, with the rapid development of electronic technology, the electromagnetic interference environment has become increasingly complex and changeable. Traditional interference suppression methods are mainly divided into parametric methods or non-parametric methods, both of which have problems such as weak adaptability. These problems seriously affect the anti-jamming performance of SAR, making it face many challenges in practical applications. Once the interference signal enters the SAR system, it is very likely to damage the integrity and accuracy of the echo signal, which may lead to problems such as blurred imaging and loss of target features, seriously affecting the imaging quality.

[0003] Therefore, it is crucial to develop effective anti-jamming technologies for SAR. It is found that interference signals are concentrated in the time-frequency spectrum, which is significantly different from the time-frequency spectrum of the original echo signal. Therefore, interference detection and removal can be completed in the time-frequency domain. Moreover, the processing object of SAR interference suppression is essentially data processing. Deep learning algorithms can automatically learn interference features from a large amount of interference data and build a model adapted to complex interference environments without the need for manual pre-setting of complex model parameters and rules. Therefore, combining these emerging artificial intelligence technologies with SAR time-frequency domain signal processing is expected to break through the limitations of traditional methods and significantly improve the anti-jamming ability of SAR in complex electromagnetic environments. Summary of the Invention

[0004] The present invention discloses a SAR intelligent anti-jamming method based on time-frequency domain processing to cope with the current complex and changeable electromagnetic interference environment and improve the anti-jamming ability of SAR in the electromagnetic interference environment.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A SAR intelligent anti-jamming method based on time-frequency domain processing, the method comprising the following steps:

[0007] Step 1, constructing a SAR imaging and processing module; wherein, the SAR imaging and processing module is used to generate a SAR image of the input SAR echo signal and perform scale normalization processing on the generated SAR image;

[0008] In addition, in order to enhance data diversity, for the data set used in the model training stage, it further includes: performing image enhancement processing on the generated SAR image, including but not limited to operations such as rotating and scaling the image;

[0009] Step 2: Construct an adversarial perturbation module to achieve the interference function. The input of this adversarial perturbation module is the SAR echo signal, which is used to output the interfered SAR echo signal;

[0010] Step 3: Construct an interference detection module to achieve the interference detection function. This interference detection module includes an interference detection network based on an autoencoder; the interference detection module first extracts the echo time-frequency spectrum of the input SAR echo signal and uses it as the input of the interference detection network to extract the signal feature map of the SAR echo signal;

[0011] For any SAR echo signal to be detected, when there is a feature vector in its signal feature map that does not belong to the preset feature space, it is determined that the SAR echo signal is interfered; otherwise, it is determined that the SAR echo signal is not interfered;

[0012] Step 4: Construct an anti-interference module to achieve the anti-interference function; the input of this interference detection module is the SAR echo signal determined to be interfered, and it removes the interference component of the input SAR echo signal based on a filter;

[0013] Step 5: Build a SAR intelligent anti-interference system based on the adversarial perturbation module, the interference detection module, the anti-interference module and the SAR imaging and processing module;

[0014] Step 6: Input the original SAR echo signal x into the SAR imaging and processing module and the adversarial perturbation module of the SAR intelligent anti-interference system respectively at the same time; generate the original SAR image y and the interfered echo signal x';

[0015] Input the echo signals x and x' into the interference detection module, preset a feature space based on the signal feature map of the original SAR echo signal (this feature space is equivalent to a threshold, which is jointly composed of time and frequency spectrum. Signals within this threshold range are all normal signals; signals outside this threshold will be determined to be interfered. This feature space can be set based on the accumulated signal feature maps of the original SAR echo signals for interference detection in the test stage), and perform interference detection on it based on the signal feature map of the interfered echo signal; if the detection result is that it is interfered, input the echo signal x' into the anti-interference module to obtain the anti-interference signal x1 and input it into the SAR imaging and processing module to obtain the SAR anti-interference image y1;

[0016] Step 7, based on the set image quality evaluation index, evaluate the anti-interference effect of the original SAR image y and the SAR anti-interference image y1. Based on the anti-interference effect evaluation, train the network model parameters of the SAR intelligent anti-interference system, and use the trained interference detection module, anti-interference module, and SAR imaging and processing module for the SAR anti-interference generation of the target SAR echo signal. That is, first input the target SAR echo signal into the interference detection module. If it is detected that it is interfered, then send it to the anti-interference module, and finally send it to the SAR imaging and processing module; if it is not detected that it is interfered, then directly send it to the SAR imaging and processing module; based on the output of the SAR imaging and processing module, obtain the SAR image of the target SAR echo signal.

[0017] Further, in step 1, use the RD algorithm (Range-Doppler algorithm, a commonly used algorithm in synthetic aperture radar (SAR) imaging, mainly used to process SAR echo signals to generate high-resolution images) to obtain the SAR image of the SAR echo signal.

[0018] Further, in step 1, the normalization scale function used for scale normalization processing is: where v max is the maximum amplitude of the SAR echo signal in the training set used for training the SAR intelligent anti-interference system, R0 is the slant range of the echo signal corresponding to v max and R is the slant range of the echo signal to be scale-normalized.

[0019] Further, the anti-perturbation module can generate the interfered SAR echo signal based on the Gaussian white noise signal; or it can first generate the general perturbation in the SAR image domain (i.e., the anti-perturbation δ) based on the generator of ResNet (the trained generator G), and then combine the anti-perturbation δ and the SAR system parameters to obtain the interfered SAR echo signal by using the range convolution and azimuth product modulation interference method.

[0020] That is, first generate the original SAR image of each original SAR echo signal in the training set (used for training the SAR intelligent anti-interference system) based on the SAR imaging and processing module, and obtain an input template that conforms to the distribution characteristics of the SAR image based on the original SAR images of several training samples where n represents the number of samples, and x i represents the original SAR image of the i-th sample; then input the template An input generator to output an adversarial perturbation δ; that is, the generator implements a mapping through ResNet to obtain a general adversarial perturbation (the adversarial perturbation plus the input sample is the corresponding adversarial sample); then, based on the generated adversarial perturbation and the SAR system coefficients, a modulation interference algorithm of distance convolution and azimuth multiplication is used to obtain an interference signal, and the interference signal is output through the SAR imaging and processing module as the corresponding adversarial perturbation in the SAR image domain, so as to achieve the purpose of converting the adversarial perturbation in the SAR image domain generated by the generator into an interference signal in the time-frequency domain.

[0021] That is, in the present invention, the input template can be set in the following ways: generating the original SAR images of the original SAR echo signals in the training set of the SAR intelligent anti-jamming system through the SAR imaging and processing module; and generating an input template based on the mean value of the original SAR images in the training set. It can also directly use random Gaussian noise as the input template.

[0022] Furthermore, the loss function of the generator during training is set as:

[0023] L G = λ·L G1 +(1 - λ)L G2

[0024] where λ is a preset weight coefficient, L G1 represents the attack loss, and L G2 represents the interference concealment loss (that is, the degree of image distortion is measured by the p-norm), and its calculation formula is specifically:

[0025]

[0026] where, represents the adversarial sample obtained by adding the adversarial perturbation to the original SAR echo signal of the input, and f v (·) represents the logits output of the victim model, and the victim model is set as: inputting the input template into the generator, outputting the adversarial perturbation δ through the generator, and then obtaining the adversarial sample based on the original SAR echo signal and predicting the target type of the adversarial sample based on a classifier (such as a fully connected classifier with at least two layers, and the last fully connected layer has a softmax function) to represent the attacker type, C tr represents the true category (that is, the true category regarding the attacker type), i represents the target category in the dataset, and p represents the norm.

[0027] Further, in step 3, the interference detection network is the Location-Net model of the interference detection network constructed based on the autoencoder, which includes an encoder and a decoder. Among them, the encoder is used to extract the signal feature map of the echo time-frequency spectrum of the SAR echo signal, and the decoder is used to reconstruct the extracted signal feature map into the echo time-frequency spectrum of the SAR echo signal. Moreover, when the Location-Net model of the interference detection network is trained, white noise can be added to the input echo time-frequency spectrum to improve the robustness of the network.

[0028] In addition, during training, the Adam optimization algorithm can be used to complete the training of Location-Net, so as to realize the feature extraction of the signal feature map of the input SAR echo signal based on the feature mapping of the trained encoder.

[0029] Further, in step 5, the Location-Net model is specifically set as follows:

[0030] The encoder includes a convolutional layer and a downsampling inverted residual module. The convolutional kernel size of the convolutional layer is 3×3, and the convolutional stride is 2. The convolutional layer receives the input of the Location-Net model and completes the first downsampling. After convolution, the activation function ReLU is used. The downsampling inverted residual module realizes the second downsampling of the encoder.

[0031] The decoder includes two upsampling inverted residual modules and a convolutional layer. The first upsampling is completed by two upsampling inverted residual modules, which just corresponds to the two downsamplings in the encoder. Among them, the convolutional kernel size of the convolutional layer is also 3×3, which makes the shape of the network output data the same as the input data.

[0032] Further, in step 7, the adopted image quality evaluation indexes include: signal-to-noise ratio (SNR), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM).

[0033] The technical solution provided by the present invention at least brings the following beneficial effects:

[0034] Based on time-frequency domain signal processing, the present invention integrates the data-driven advantages of deep learning technology into the field of SAR anti-interference, significantly improves the anti-interference ability of SAR under complex electromagnetic interference conditions, and realizes intelligent anti-interference of SAR. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The above and / or additional aspects and advantages of the present invention will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, in which:

[0036] Figure 1 It is a schematic flowchart of a SAR intelligent anti-interference method based on time-frequency domain processing provided by an embodiment of the present invention.

[0037] Figure 2 Schematic diagram of the structure of the SAR intelligent anti-jamming system provided by the embodiment of the present invention. Detailed implementation manners

[0038] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of the present invention will be described in detail and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described by referring to the accompanying drawings are exemplary and are intended to explain the present invention, rather than being construed as a limitation to the present invention.

[0039] In the face of problems such as complex and changeable electromagnetic interference environments, the embodiment of the present invention provides a SAR intelligent anti-jamming method based on time-frequency domain processing. Based on the time-frequency domain, an autoencoder model is selected to construct a deep learning model to realize interference detection and removal of SAR echo samples. The method of the present invention combines time-frequency domain signal processing with deep learning technology, integrates the data-driven advantages of the former into the field of SAR anti-jamming, and utilizes the powerful data-driven advantages of deep learning technology to mine interference features from a large amount of data, thereby realizing effective removal of interference. Thereby significantly improving the anti-jamming ability of SAR under complex electromagnetic interference conditions, which is conducive to realizing intelligent anti-jamming of SAR.

[0040] In one embodiment, referring to Figure 1 , a SAR intelligent anti-jamming method based on time-frequency domain processing provided by the embodiment of the present invention is realized through the following steps:

[0041] S1. Construct a SAR imaging and processing module. Obtain the SAR raw echo signal, use the RD algorithm to obtain the SAR raw image sample, and preprocess the collected image samples, including but not limited to operations such as rotating and scaling the image for enhancement processing to enhance data diversity (this processing is mainly applicable to the model training stage, that is, image enhancement processing of the SAR raw image samples in the training set), and normalize the image data to a fixed range, where the normalization scale function is: The original data set of the image is obtained after preprocessing.

[0042] In this embodiment, the two-dimensional SAR echo processing is decomposed into a cascaded form of two one-dimensional processings. The feature is to perform FFT on the data after range compression along the azimuth direction, transform it to the range-Doppler domain, and then complete range migration correction and azimuth compression to finally obtain the SAR image sample. Then define the normalization scale function as: where, v max is the maximum value of the echo data amplitude in the training set, and R0 is v maxThe slant range of the echo data belonging to, R is the slant range of the echo data to be processed. The difference in the distribution of different echo data is reduced by dividing the current data by the normalization scale.

[0043] S2. Construct an adversarial perturbation network model. First, use an adversarial perturbation template generator based on ResNet to generate an input template that conforms to the distribution characteristics of SAR images by fusing the trained images, then input the input template into the trained generator G to obtain an adversarial perturbation, and finally convert the adversarial perturbation into an interference signal x' through the range convolution azimuth product modulation interference algorithm and SAR parameter information.

[0044] In this embodiment, to fully combine the characteristics of SAR images themselves, all the SAR images in the training set are added together and averaged to generate an input template, and then ResNet is selected as the network structure of the generator G. At the same time, a loss function is designed to optimize the parameters of the generator G. In terms of the effectiveness of interference, the attack loss where f v (·) is the logits output of the victim model, C tr represents the true class, represents the adversarial sample, and i represents the target class in the dataset. In terms of the concealment of interference, Finally, the loss function L G = λ·L G1 +(1 - λ)L G2 , where λ is the weight coefficient. Thus, the adversarial perturbation δ can be obtained.

[0045] After generating the adversarial perturbation δ, it is necessary to use an interference signal generator to combine the adversarial perturbation δ and SAR system parameters, and use the range convolution azimuth product modulation interference method to obtain the interference signal x'.

[0046] S3. Construct an interference detection module to implement the interference detection function. First, input the SAR echo signal, convert the time-domain problem into a time-frequency domain problem through signal processing, and finally obtain the time-frequency spectra of the echo signal and the interference signal. Secondly, construct an interference detection network LocNet model based on an autoencoder, and use the time-frequency spectrum of the original echo without interference as the input data and output label of the network for training. To improve the robustness of the network, white noise is added to the time-frequency spectrum input to the network during training. And use the Adam optimization algorithm to complete the training of LocNet. The features of the normal echo signal are extracted through feature mapping. A feature space is preset according to the extracted SAR original echo features. When there are feature vectors in the feature map F that do not belong to the feature space L, it is determined that the SAR echo signal is interfered; otherwise, it is determined that the SAR echo signal is not interfered.

[0047] In this embodiment, the short-time Fourier transform (STFT) is first used to complete the step of converting the echo from the time domain to the time-frequency domain. The parameters of the STFT are set as follows: the window function is the Hamming window, the window length is 31, the number of Fourier transform points is 32, and the sliding step of the window is 1.

[0048] Secondly, based on the overall structure of the autoencoder, an interference detection network LocNet is constructed with convolutional layers and inverted residual modules as the basic modules. The inverted residual module is a lightweight convolutional module used to replace the convolutional layers in the network and serve as the basic building block of the convolutional neural network, which greatly reduces the computational complexity of the network. Its main idea is effective information transmission and network nonlinear mapping. First, the feature map is upsampled to a higher number of channels, then the feature map with a high number of channels is subjected to network nonlinear processing, and after completion, the feature map is downsampled to save the effective information with a low-channel feature map. The LocNet network is mainly divided into two parts: an encoder and a decoder. The encoder consists of a convolutional layer and a downsampling inverted residual module. The size of the convolutional kernel is 3×3, and the convolutional stride is 2. The convolutional layer receives the input of the network and completes the first downsampling. After convolution, the activation function ReLU is used. The downsampling inverted residual module is responsible for the second downsampling of the encoder. The decoder contains two upsampling inverted residual modules and a convolutional layer. The first upsampling is completed by two upsampling inverted residual modules, which exactly corresponds to the two downsamplings in the encoder. The size of the convolutional kernel of a convolutional layer is also 3×3, which makes the shape of the network output data the same as that of the input data. After STFT, the time-frequency spectrum of the echo without interference is used as the input data and output label of the network to detect the interference time-frequency component N(t,f) = |x(t,f) - x'(t,f)| as the training target to participate in the training, where x(t,f) and x'(t,f) are the time-frequency components of the SAR original echo signal x and the interference signal x' respectively. During the training process, white noise is added to the time-frequency spectrum input to the network to improve the robustness of the network, and the Adam optimization algorithm is used to complete the training of LocNet.

[0049] S4. Construct an anti-interference module to implement the anti-interference function. The echo signal determined by the interference detection module to be interfered is input into this module, and the interference component is removed through a filter to achieve the anti-interference function.

[0050] In this embodiment, a threshold τ is set according to the interference signal x', and then the interference time-frequency component is located according to this threshold. When N(t,f) > τ, the interference time-frequency component location L(t,f) = 1, otherwise, it is 0. Then, a filter is designed according to the location L(t,f) of the interference time-frequency component to complete the removal of interference. After that, the inverse short-time Fourier transform is performed to complete the conversion of the signal from the time domain to the time-frequency domain.

[0051] S5. Integrate the adversarial perturbation module, interference detection module, anti-interference module with the SAR imaging and processing module to build an overall anti-interference system. By building a SAR intelligent anti-interference system (i.e., building an overall anti-interference network), the interaction of each module is completed.

[0052] In this embodiment, the system is built by cascading each module.

[0053] S6. Initialize the system parameters, input the original signal and the signal after anti-interference into the system, and finally obtain the SAR original image and the SAR anti-interference image, which specifically includes:

[0054] S61. Initialize the system parameters.

[0055] S62. Input the original SAR echo signal x into the SAR imaging and image processing module and the adversarial perturbation module respectively at the same time to generate the original SAR image y and the interfered signal x'.

[0056] S63. Input the interfered signal x' and x into the interference detection module together. If there is a feature vector in the feature map that does not belong to the feature space, then flag = 1, and it is determined that the SAR echo signal is interfered; otherwise, flag = 0 is determined that the SAR echo signal is not interfered.

[0057] S64. For the detected interfered signal x', input it into the anti-interference module to obtain the anti-interference signal x1, and then input x1 into the SAR imaging and processing module to obtain the SAR anti-interference image y1.

[0058] In this embodiment, the interference, interference detection, interference localization and removal of the original echo signal are completed in the time-frequency domain, and then the signal passes through the SAR imaging module to generate the final SAR anti-interference image. This can not only more intuitively observe the anti-interference effect of the method, but also facilitate the next step through the image quality evaluation index.

[0059] S7. Feedback the quality of the SAR anti-interference image according to the image quality evaluation index, so as to evaluate the effect of the SAR intelligent anti-interference method based on time-frequency domain processing. Commonly used indicators include signal-to-noise ratio (SNR), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), etc.

[0060] In this embodiment, the peak signal-to-noise ratio (PSNR), structural similarity (SSIM) and learned perceptual image patch similarity (LPIPS) are selected to measure the quality of the SAR anti-interference image y1 from three aspects of signal-to-noise ratio, image structural similarity and image perceptual similarity.

[0061] First, for PSNR, there is a formula It measures the difference between y1 and y. Among them, the mean square error has where the pixel size is m×n, the number of image bits is b, y(i,j), y1(i,j) are the pixel values at the pixel point (i,j). When the difference between y1 and y is smaller, the MSE is smaller and the PSNR is larger, which means the better the image quality and the better the anti-interference effect. Secondly, for SSIM, there is a formula The subscript i is used to represent the pixel point, N is the total number of pixel points. It measures the similarity between y1 and y from three aspects: brightness, contrast and structure. The more similar the images are, the closer the SSIM is to 1 and the better the anti-interference performance of the SAR image. Finally, for LPIPS, it uses deep features to measure the similarity between y1 and y. The higher the similarity, the smaller the difference between the deep features, and the smaller the output result of LPIPS, and the better the anti-interference effect of the SAR image. Finally, the model parameters of the SAR intelligent anti-interference system in the embodiments of the present invention are trained by comprehensively considering the three evaluation qualities. It stops when the preset training convergence conditions are met (such as the convergence of the comprehensive evaluation index or the iteration number reaches the upper limit, etc.), so as to generate the SAR anti-interference of the target SAR echo signal based on the trained interference detection module, anti-interference module, and the line of sight of the SAR imaging and processing module.

[0062] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0063] In addition, the descriptions such as "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first", "second", etc. can explicitly or implicitly include at least one of the features.

[0064] Any process or method description, whether in a flowchart or otherwise described in this specification, can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations, where functions may be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0065] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with suitable combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0066] Those of ordinary skill in the art of the present technology can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A SAR intelligent anti-jamming method based on time-frequency domain processing, characterized in that Including the following steps: Step 1, construct an SAR imaging and processing module, which is used to generate an SAR image of the input SAR echo signal and perform scale normalization processing on the generated SAR image; Step 2, construct an adversarial perturbation module, the input of which is the SAR echo signal and is used to output a perturbed SAR echo signal; Step 3, construct an interference detection module, which includes an interference detection network based on an autoencoder; the interference detection module first extracts the echo time-frequency spectrum of the input SAR echo signal and uses it as the input of the interference detection network to extract the signal feature map of the SAR echo signal; For any SAR echo signal to be detected, when there is a feature vector in its signal feature map that does not belong to the preset feature space, it is determined that the SAR echo signal is interfered; otherwise, it is determined that the SAR echo signal is not interfered; Step 4, construct an anti-interference module, the input of which is the SAR echo signal determined to be interfered, and it removes the interference component of the input SAR echo signal based on a filter; Step 5, build an SAR intelligent anti-interference system based on the adversarial perturbation module, the interference detection module, the anti-interference module and the SAR imaging and processing module; Step 6, input the original SAR echo signal x into the SAR imaging and processing module and the adversarial perturbation module of the SAR intelligent anti-interference system respectively at the same time; generate the original SAR image y and the perturbed echo signal x'; Input the echo signals x and x' into the interference detection module, preset the feature space based on the signal feature map of the original SAR echo signal, and perform interference detection based on the signal feature map of the perturbed echo signal; if the detection result is interfered, input the echo signal x' into the anti-interference module to obtain the anti-interference signal x1 and input it into the SAR imaging and processing module to obtain the SAR anti-interference image y1; Step 7, evaluate the anti-interference effect of the original SAR image y and the SAR anti-interference image y1 based on the set image quality evaluation index, train the network model parameters of the SAR intelligent anti-interference system based on the anti-interference effect evaluation, and use the trained interference detection module, anti-interference module, and SAR imaging and processing module for the generation of SAR anti-interference of the target SAR echo signal.

2. The method according to claim 1, wherein In Step 1, the distance-Doppler algorithm is used to obtain the SAR image of the SAR echo signal.

3. The method according to claim 1, characterized in that, In step 1, the normalization scale function used for scale normalization processing is as follows: where v max is the maximum amplitude of the SAR echo signal in the training set used for the training of the SAR intelligent anti-jamming system, R0 is the slant range of the echo signal corresponding to v max and R is the slant range of the echo signal to be processed by scale normalization.

4. The method according to claim 1, wherein The adversarial perturbation module generates a perturbed SAR echo signal based on Gaussian white noise signal.

5. The method according to claim 1, wherein The adversarial perturbation module obtains the perturbed SAR echo signal based on the generator of ResNet, specifically including: Set the input template: generate the original SAR images of the original SAR echo signals in the training set of the SAR intelligent anti-interference system through the SAR imaging and processing module; and generate the input template based on the mean value of the original SAR images in the training set; or directly use random Gaussian noise as the input template; Input the input template into the generator to output the adversarial perturbation δ; According to the adversarial perturbation δ and the SAR system coefficients, the disturbed SAR echo signal is obtained based on the modulation interference algorithm of range convolution and azimuth product.

6. The method according to claim 5, wherein The loss function of the generator during training is set as: L G = λ·L G1 + (1 - λ)L G2 Among them, λ is a preset weight coefficient, and L G1 represents the attack loss, and L G2 represents the interference concealment loss, and its calculation formula is specifically as follows: Among them, f v (·) represents the output of the victim model, and the victim model is set as follows: input the input template into the generator, output the adversarial perturbation δ through the generator, and add the adversarial perturbation δ to the original SAR echo signal of the input to obtain the adversarial sample Then, predict the adversarial sample based on the classifier for the target type, C tr represents the true class, i represents the target class in the dataset, and p represents the norm.

7. The method according to claim 1, wherein In step 3, the interference detection network is the Location-Net model of the interference detection network constructed based on the autoencoder, which includes an encoder and a decoder. Among them, the encoder is used to extract the signal feature map of the echo time-frequency spectrum of the SAR echo signal, and the decoder is used to reconstruct the extracted signal feature map into the echo time-frequency spectrum of the SAR echo signal; and when the Location-Net model of the interference detection network is trained, white noise is added to the input echo time-frequency spectrum.

8. The method according to claim 7, characterized in that In step 5, the Location-Net model is specifically set as: The encoder includes a convolutional layer and a downsampling inverted residual module. Among them, the convolutional kernel size of the convolutional layer is 3×3, and the convolutional stride is 2; the convolutional layer receives the input of the Location-Net model and completes the first downsampling. After convolution, the activation function is ReLU; the downsampling inverted residual module realizes the second downsampling of the encoder; The decoder includes two upsampling inverted residual modules and a convolutional layer. Among them, the convolutional kernel size of the convolutional layer is 3×3.

9. The method according to claim 1, wherein In step 7, the adopted image quality evaluation indexes include: signal-to-noise ratio, peak signal-to-noise ratio, and structural similarity index.

10. The method according to claim 1, characterized in that, For the training set used for the training of the SAR intelligent anti-jamming system, the SAR imaging and processing module further includes: performing image enhancement processing on the generated SAR image.

Images