SAR deception jamming effect evaluation method based on visual network

Abstract

The application discloses a method for evaluating the effect of SAR deception jamming based on a visual network, and belongs to the field of electronic countermeasures. The method can solve the problem that the existing method for evaluating the effect of SAR deception jamming cannot evaluate the reconnaissance error of different parameters with different degrees. The method specifically comprises the following steps: step 1, generating jamming samples containing SAR parameter reconnaissance error, taking the generated jamming samples as input in a training set, and taking the parameter error type and parameter error degree corresponding to the jamming samples as output in the training set; step 2, building a visual network model; step 3, training the built visual network model by using the training set; and step 4, inputting a jamming sample to be evaluated into the trained visual network model, and outputting the evaluation result of the parameter error type and parameter error degree of the jamming sample to be evaluated by the trained visual network model. The method can be applied to the evaluation of the effect of SAR deception jamming.

Description

Technical Field

[0001] This invention belongs to the field of electronic warfare, specifically relating to a method for evaluating the effectiveness of SAR deception jamming based on visual networks. Background Technology

[0002] After a jammer performs deception jamming on a Synthetic Aperture Radar (SAR), it is necessary to effectively assess the jamming effect in order to make correct strategic decisions. This includes assessing the effectiveness of the jamming, the quality of the jammed imaging, whether there are parameter reconnaissance errors, and whether it is necessary to re-reconnoiter the parameters and implement jamming again. Deception jamming effect assessment is a crucial technology driving the continuous development of SAR electronic countermeasures. Ensuring the effectiveness and accuracy of the assessment technology is an important prerequisite for further advancements in electronic countermeasures.

[0003] In the process of generating SAR deception jamming signals, various factors, such as reconnaissance, inevitably lead to errors of different types and degrees in various parameters, affecting the effectiveness of SAR deception jamming and thus degrading the performance of the jammer. In the current complex and intelligent electromagnetic environment, traditional indicator evaluation methods are generally limited in scope and lack universality. They cannot accurately evaluate SAR deception jamming imaging results under different reconnaissance error conditions, resulting in significant errors and uncertainties, making them unsuitable for the needs of current electronic warfare. Therefore, researching and utilizing current deep learning technology to scientifically, reliably, and accurately evaluate SAR deception jamming results containing different error types and degrees can significantly enhance SAR electronic warfare capabilities.

[0004] The paper "Performance evaluation of deception against synthetic aperture radar based on multifeature fusion" combines explicit evaluation metrics with latent features extracted by neural networks to classify and evaluate interference images generated by different degrees of motion parameter errors. The results show that deep learning networks can achieve better evaluation results than traditional methods. However, this method only considers motion errors and not other parameter error types, resulting in an overly idealized evaluation type. The paper "Evaluation of Deceptive Jamming Effect on SAR Based on Visual Consistency" systematically integrates three visual levels—detection, recognition, and semantics—to propose an efficient evaluation framework for SAR deception jamming effects based on visual consistency. While it achieves better evaluation results for error outcomes, it also fails to consider the evaluation of reconnaissance errors with different parameters. The paper "A SAR Jamming Effect Evaluation Method Based on AHP and Target Detection and Recognition Performance" proposes an effective combination of the analytic hierarchy process (AHP) and currently popular target detection and recognition methods, achieving good results in simulation experiments. However, it still cannot achieve evaluation of deceptive targets under conditions of reconnaissance errors.

[0005] The evaluation of SAR deception jamming effectiveness mainly assesses the jamming effect by judging the quality of the imaging results, but there is a lack of methods to evaluate the deterioration of jamming effect caused by parameter reconnaissance errors. Although existing methods have some connection with deep learning theory, they are limited to target recognition and the evaluation of error severity under a single error type, without in-depth research on different error types.

[0006] In summary, existing methods for evaluating the effectiveness of SAR deception and jamming cannot assess situations where different parameter types exhibit varying degrees of reconnaissance error. Summary of the Invention

[0007] The purpose of this invention is to address the problem that existing SAR deception jamming effect evaluation methods cannot evaluate situations where different parameter types have varying degrees of reconnaissance errors, and to propose a SAR deception jamming effect evaluation method based on visual networks.

[0008] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0009] A method for evaluating the effectiveness of SAR deception jamming based on visual networks, the method specifically includes the following steps:

[0010] Step 1: Generate interference samples containing SAR parameter reconnaissance errors. Use the generated interference samples as input to the training set, and use the parameter error type and parameter error degree corresponding to the interference samples as output to the training set.

[0011] Step 2: Construct a visual network model, which includes a visual feature extraction module, a Global Average Pooling (GAP) layer, and a Linear layer.

[0012] Step 3: Train the constructed visual network model using the training set;

[0013] Step 4: Input the interference samples to be evaluated into the trained visual network model, and output the evaluation results of the parameter error type and parameter error degree of the interference samples to be evaluated through the trained visual network model.

[0014] Furthermore, in step one, generating interference samples containing SAR parameter reconnaissance errors involves the following specific process:

[0015] Select the SAR parameters used to generate interference samples, and represent the value of the nth selected SAR parameter in the error-free case as σ. n n = 1, 2, ..., N, where N represents the number of SAR parameters selected;

[0016] For each selected SAR parameter, K error levels are set, and the value of the nth SAR parameter with error level k is represented as σ. ＜n,k＞ k = 1, 2, ..., K;

[0017] According to σ ＜n,k＞ Generate SAR deception jamming signal s(t), n=1,2,…,N, k=1,2,…,K;

[0018] Then based on the error-free parameter σ n Image the SAR spoofing jamming signal CS (Chirp Scaling, CS) to form jamming sample X. ＜n,k＞ .

[0019] Furthermore, the SAR parameters selected for generating interference samples are range modulation frequency, carrier frequency, velocity, position, and pulse repetition frequency.

[0020] Furthermore, the error levels are 1%, 3%, 5%, 7%, and 9%, respectively, i.e., K = 5.

[0021] An error margin of 1% refers to the parameter value σ used when generating the SAR deception jamming signal. ＜n,1＞ , σ ＜n,1＞ =σn ±0.01σ n .

[0022] Furthermore, the interference sample X ＜n,k＞ After inputting the constructed visual network model, the output of the visual network model is:

[0023] F out =Linear{GAP[VAN(X ＜n,k＞ )]}

[0024] Among them, F out VAN(·) is the output of the visual network model, and VAN(·) is the visual feature extraction module.

[0025] The visual feature extraction module not only considers spatial adaptability but also captures channel-dimensional adaptability, which is ignored by the self-attention mechanism. Therefore, when evaluating the effectiveness of deception interference at the SAR two-dimensional image level, the visual network model of this invention has significant advantages and is beneficial for a deeper understanding of SAR image information.

[0026] Furthermore, the visual feature extraction module includes four units, each of which includes a downsampling layer, a VAN block, and an LN.

[0027] The beneficial effects of this invention are:

[0028] This invention utilizes inversion to convert a large number of SAR images into SAR interference samples containing different parameter error types and error degrees. Then, the interference samples are used to train a visual network model. Finally, the interference samples to be evaluated are input into the trained visual network model. This solves the problem that existing SAR deception interference effect evaluation methods cannot evaluate the parameter error type and error degree. The evaluation results of this invention for SAR deception interference effect are consistent with the real results, and can achieve better interference effect evaluation results.

[0029] This invention utilizes a visual network model to enhance intelligence, and can quickly, accurately, and reliably determine what kind of parameter errors and what degree of reconnaissance error exist in the SAR deception jamming results, so as to adjust electronic countermeasures strategies and overcome the limitations of current SAR deception jamming effect evaluation. Attached Figure Description

[0030] Figure 1 This is a structural diagram of the SAR deception interference sample generation method of the present invention;

[0031] Figure 2 This is a diagram of the VAN network (visual network) model constructed in this invention;

[0032] In the figure, H and W represent the height and width of the image, respectively, and GAP (Global Average Pooling) represents global average pooling.

[0033] Figure 3 This is a graph showing the evaluation results of parameter error types in this invention;

[0034] The figure shows the specific evaluation accuracy of various types of errors, where the vertical axis represents the true error type and the horizontal axis represents the evaluated and predicted error type. According to the results in the figure, the model can identify almost all five error types successfully, enabling the model to identify error types in the test interference image data. Although the classification results show that the carrier frequency error classification accuracy is 0.97, with 1% and 2% being misclassified as velocity error and closest slant distance error, respectively, overall, the method of this invention can accurately and effectively evaluate error types.

[0035] Figure 4 This is a graph showing the evaluation results of the parameter error level in this invention;

[0036] The results show that SAR deception jamming images at each error level can almost guarantee no erroneous distinctions, and the error level evaluated by the model is basically consistent with the actual error level, achieving a good error level evaluation effect.

[0037] Figure 5 This is a test result diagram of the SAR deception interference effect evaluation based on visual networks in this invention;

[0038] In the generation of interference samples, it is assumed that the enemy SAR radar is performing frontal and side imaging, and that the transmitted signal is a linear frequency modulated signal s(t). Where T is the signal pulse width, taken as 2.5μs, the carrier frequency is 5.3GHz, and the distance chrip modulation frequency is 20×10. 12 Hz / s, the nearest slant distance R0 in the simulation is 2×10 4 m and v are set to 150 m / s, and the angle of inclination is 0°;

[0039] The simulation in this invention uses the publicly available MSTAR dataset, and the selected parameter types include range modulation frequency Kr, carrier frequency f0, velocity v, and position R. N Five reconnaissance parameters, including pulse repetition frequency (PRF), were set with five error levels: 1%, 3%, 5%, 7%, and 9%. An SAR deception jamming signal with errors was generated using an ICS (Inverse Chirp Scaling) inversion-based SAR deception jamming method. Then, the CS image of the deception jamming signal was formed using error-free parameters to create jamming samples.

[0040] In the figure, the labels on the left side of the image represent the true error type and error level of the image, while the labels below the image represent the error type and error level predicted by the model for the SAR jamming image. The results in the figure show that all predicted labels are consistent with the true labels of the jamming image, indicating that the model trained in this invention can effectively evaluate the error type and error level of the deception jamming results, achieving a reliable, accurate, and effective evaluation of the SAR deception jamming effect. The SAR deception jamming effect evaluation based on a visual network designed in this invention can directly evaluate jamming samples with error types and error levels, and its evaluation performance is good. Detailed Implementation

[0041] The method of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Specific Implementation Method 1: The SAR deception interference effect evaluation method based on visual networks described in this implementation method specifically includes the following steps:

[0043] Step 1, such as Figure 1 As shown, SAR parameter reconnaissance error interference samples are generated. The generated interference samples are used as inputs to the training set, and the parameter error type and parameter error degree corresponding to the interference samples are used as outputs to the training set.

[0044] Specifically:

[0045] Select the SAR parameters used to generate interference samples, and represent the value of the nth selected SAR parameter in the error-free case as σ. n n = 1, 2, ..., N, where N represents the number of SAR parameters selected;

[0046] For each selected SAR parameter, K error levels are set, and the value of the nth SAR parameter with error level k is represented as σ. ＜n,k＞ k = 1, 2, ..., K;

[0047] According to σ ＜n,k＞ Generate SAR deception jamming signal s(t), n=1,2,…,N, k=1,2,…,K;

[0048] Then based on the error-free parameter σ n Image the SAR spoofing jamming signal CS (Chirp Scaling, CS) to form jamming sample X. ＜n,k＞ .

[0049] In this invention, 200 images of each of the 10 target classes in the MSTAR training dataset are randomly selected first, resulting in a total of 2000 original SAR images. Then, using the controlled variable method, that is, while keeping other reconnaissance parameters constant, the five parameters that have a significant impact on interference imaging (range modulation frequency Kr, carrier frequency f0, velocity v, position R) are analyzed. N Each reconnaissance parameter (Pulse Repetition Frequency, PRF) is used to generate SAR deception jamming signals with different error levels (1%, 3%, 5%, 7%, 9%) through SAR inversion. Finally, experimental jamming sample images are generated using the CS imaging algorithm. Similarly, an average of 40 images of each target type are selected from the SAR test set, resulting in 400 original SAR images. Corresponding test jamming samples are generated using the above method, with 5dB Gaussian noise added during sample generation to verify the algorithm's anti-jamming capability. In the acquired sample X... ＜n,k＞ The CCP's data includes 50,000 training samples, 10,000 validation samples, and 10,000 test samples.

[0050] Step 2: Construct a visual network model, which includes a visual feature extraction module, a global average pooling (GAP) layer, and a Linear layer.

[0051] The visual network model constructed in this invention is as follows: Figure 2 As shown. The entire model consists of downsampling, visual structure (VANblock), LayerNorm (LN), Global Average Pooling (GAP), and Linear operations. For the input interference sample X, the output of the visual network model can be expressed as: F out =Linear{GAP[VAN(X)]}, where Linear and GAP represent Linear and GAP operations respectively, and VAN represents a visual feature extraction network integrated by downsampling, VAN block, and LN. Given an input image of H×W×3, it mainly accumulates features and grasps information through 4 stages, while realizing dimensionality reduction in image size and dimensionality increase in channels.

[0052] The feature maps for each stage are {1 / 4, 1 / 8, 1 / 16, 1 / 32} of the original image. The output of each stage in the VAN can be represented as follows: Where i represents the stage number, VAN_b refers to the VAN block network structure, D represents the downsampling overlap patch merging operation, and T represents the tensor before each input stage. <H i W i C i> represents the dimension of the input tensor of the i-th layer. The core LKA (Large Kernel Attention) in VAN_b can be represented as: This represents the element-wise multiplication operation between Att and X, using an attention graph. This can be represented as: Att = Conv 1×1 {DW-D-Conv[DWConv(X)]}, where Conv 1×1 DW-D-Conv represents a 1×1 convolution, DWConv represents a depthwise extended convolution, and DWConv represents a depthwise convolution.

[0053] Step 3: Input the training and validation sets from the interference samples into the visual network model for training.

[0054] The obtained 50,000 interference training samples and 10,000 validation samples were fed into the completed visual network model. The training learning rate was set to 0.001, the loss function was cross-entropy error, and Adam was used for optimization. The batch size was set to 16. The model learned the features in the interference samples and verified the evaluation of the model through the validation set. A total of 100 training iterations were performed to achieve effective and accurate evaluation of the test interference samples.

[0055] Evaluation results as follows Figure 3 , Figure 4 and Figure 5 As shown.

[0056] Step 4: Evaluate the test samples based on the results of the optimal network model.

[0057] The test interference samples to be evaluated are fed into the trained network model to obtain the evaluation results of the model's interference effect on the test samples.

[0058] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for evaluating the effectiveness of SAR deception interference based on visual networks, characterized in that, The method specifically includes the following steps: Step 1: Generate interference samples containing SAR parameter reconnaissance errors. Use the generated interference samples as input to the training set, and use the parameter error type and parameter error degree corresponding to the interference samples as output to the training set. The specific process for generating interference samples containing SAR parameter reconnaissance errors is as follows: Select the SAR parameters used to generate interference samples, and select the first... The values ​​of the SAR parameters under error-free conditions are expressed as follows: , , Indicates the number of SAR parameters selected; For each selected SAR parameter, set K error levels, and assign each error level to... The first time The values ​​of the SAR parameters are represented as follows: , ; according to Generate SAR deception jamming signal , , ; Then based on error-free parameters CS imaging of SAR deception jamming signals forms jamming samples ; Step 2: Construct a visual network model, which includes a visual feature extraction module, a global average pooling (GAP) layer, and a Linear layer. Step 3: Train the constructed visual network model using the training set; Step 4: Input the interference samples to be evaluated into the trained visual network model, and output the evaluation results of the parameter error type and parameter error degree of the interference samples to be evaluated through the trained visual network model.

2. The method for evaluating the SAR deception interference effect based on visual networks according to claim 1, characterized in that, The SAR parameters selected for generating interference samples are range modulation frequency, carrier frequency, velocity, position, and pulse repetition frequency.

3. The method for evaluating the SAR deception interference effect based on visual networks according to claim 2, characterized in that, The error levels are 1%, 3%, 5%, 7% and 9%, respectively, i.e., K=5.

4. The method for evaluating the SAR deception interference effect based on visual networks according to claim 3, characterized in that, The interference sample After inputting the constructed visual network model, the output of the visual network model is: in, The output of the visual network model. This is the visual feature extraction module.

5. The method for evaluating the SAR deception interference effect based on visual networks according to claim 4, characterized in that, The visual feature extraction module includes four units, each of which includes a downsampling layer, a VAN block, and an LN.

Images