A time-frequency feature-based composite deception jamming intelligent identification method, system, device and medium

Abstract

The application discloses a composite deception jamming intelligent identification method and system based on time-frequency characteristics, equipment and a medium, and belongs to the field of radar anti-jamming. The application firstly performs pulse compression and coherent accumulation processing on the composite deception jamming signal to obtain a Doppler unit where the maximum energy spectrum of the jamming signal is located, then only performs coherent accumulation processing on the jamming signal, extracts a time domain slice of the Doppler unit where the maximum energy spectrum is located, so that the time-frequency diagram of the jamming signal after the jam-to-noise ratio is improved can be obtained; the application has good generalization ability under the condition that multiple deception jammers are dynamically superimposed and combined; the identification accuracy of the composite deception jamming under a low jam-to-noise ratio can be greatly improved, reliable technical support is provided for the field of radar anti-jamming; and the application is easy to implement in engineering and fast.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of radar anti-jamming, and particularly relates to a composite deception jamming intelligent identification method and system based on time-frequency characteristics, a device and a medium. BACKGROUND

[0002] Due to the important role of radar, it often becomes a priority target. In a complex and changeable environment, it is increasingly important to improve the anti-jamming capability of radar. With the emergence and popularity of digital radio frequency memory (DRFM) technology, the jammer can quickly generate new jamming with complex modulation mode and variable parameters, making it difficult for the radar to work normally and bringing new challenges to the development of radar anti-jamming measures. Effective identification of jamming is an important prerequisite for interference suppression. Only by determining the specific type of jamming signal can appropriate and reasonable anti-jamming measures be taken.

[0003] The existing jamming signal identification technology has two problems: first, under the condition of low jamming noise ratio, the strength of the jamming signal is close to or lower than the background noise level, resulting in low jamming identification accuracy; second, there is less research on the identification of composite deception jamming superimposed with multiple deception jammers. SUMMARY

[0004] The purpose of the present application is to provide a composite deception jamming intelligent identification method and system based on time-frequency characteristics, which effectively improves the identification accuracy of composite jamming under low jamming noise ratio and is easy to implement in engineering.

[0005] The purpose of the present application is achieved by the following technical solutions:

[0006] A composite deception jamming intelligent identification method based on time-frequency characteristics, the specific steps are as follows:

[0007] Step 1: Use the jamming signal simulation module to simulate M pulse jamming signals;

[0008] Step 2: Use the composite deception jamming generation module and the composite deception jamming signal preprocessing module to preprocess the jamming signals obtained in step 1, and then use the feature extraction module and the dataset construction module to extract features from the jamming signals to generate a dataset;

[0009] Step 3: Use the input module to input the dataset generated in step 2 into the YOLOv8 neural network inference detection model and train it;

[0010] Step 4: Based on the trained model in step 3, use the test set generation module and the evaluation module to detect the composite deception jamming under different jamming noise ratios.

[0011] Further, the step 1 simulates the slice reconstruction interference of M pulses, the intermittent sampling direct forwarding interference, the intermittent sampling cyclic forwarding interference, the comb spectrum interference by using the interference signal simulation module, the radar pulse repetition frequency is 3 kHz, the working frequency is 10 GHz, the signal pulse width is 70 μs, the bandwidth is 25 MHz, the sampling frequency is 60 MHz, and the number of coherent accumulation pulses M is 64.

[0012] Further, the step 2 is specifically:

[0013] Step 2.1: The four kinds of deception jamming signals in step 1 are additively compounded by the composite deception jamming generation module to generate 11 kinds of composite deception jamming signals;

[0014] Step 2.2: The M composite deception jamming signals in one CPI are pulse compressed and phase accumulation processed by the composite deception jamming signal preprocessing module, the maximum value is searched on the obtained range-Doppler domain result to obtain the Doppler unit f where the interference is located; only the M composite deception jamming signals in one CPI are phase accumulation processed, and the time domain slice of the Doppler unit f where the composite deception jamming is located is extracted from the obtained result;

[0015] Step 2.3: The extracted time domain slice is processed by short-time Fourier transform by the feature extraction module and the data set construction module, the time domain slice is first framed and windowed, then the windowed signal is discrete Fourier transformed, and finally the transformed result is summed to generate a time-frequency graph, the jamming-to-noise ratio, the distance and velocity of the interference are changed to obtain the composite deception jamming time-frequency graph data set.

[0016] Further, the data set constructed in step 2 is input into the YOLOv8 network by the input module in step 3, a plurality of training samples are synthesized in one time-frequency graph by using the Mosaic data enhancement technology; the backbone network uses the C2f module; the detection head adopts the SPPF module and the FPN+PAN module, and the C3 module in YOLOv5 is also replaced with the C2f module to enhance the network feature fusion capability, and the structure of the decoupling head is adopted, so that two parallel branches respectively extract the class features and position features of the interference signal, and then one layer of 1×1 convolution is used to complete the classification and positioning of the interference signal.

[0017] Further, the step 4 is specifically:

[0018] Step 4.1: The composite deception jamming signal time-frequency graph under different jamming-to-noise ratios in the range of-28 to-15 dB is generated as a test set by using the test set generation module;

[0019] Step 4.2: use the evaluation module, and based on precision P, recall R, mAP50-95 as the performance evaluation index of the composite fraud interference recognition classification; the higher P is, the more reliable the detection result is, and the less false detection is; the higher R is, the more the algorithm can detect all interference signals, and the less the missed detection is; the greater the value of mAP is, the higher the detection accuracy of the algorithm is.

[0020] Further, the calculation formulas of P, R and mAP are as follows:

[0021]

[0022]

[0023] Wherein, TP is the number of correct interference signal detection, FP is the number of false interference signal detection, FN represents the number of missed detection; AP is the accuracy, which is obtained by the area of P-R curve; mAP is the average accuracy of all interference signal categories; the precision P represents the proportion of real positive samples in the model predicted as positive samples, that is, the ratio of the number of correctly predicted positive samples to the number of all predicted positive samples; the recall R represents the ratio of the number of positive samples correctly predicted by the model to the number of real positive samples, that is, the ratio of the number of positive samples successfully found by the model to the number of all real positive samples; mAP50-95 represents the average accuracy of the model in the IoU threshold range of 50% to 95%.

[0024] A computer device / equipment / system, comprising a memory, a processor and a computer program stored on the memory, the processor executes the computer program to realize the steps of a composite fraud interference intelligent identification method based on time-frequency features.

[0025] A computer readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, realizes the steps of a composite fraud interference intelligent identification method based on time-frequency features.

[0026] A computer program product comprising a computer program / instruction, which, when executed by a processor, realizes the steps of a composite fraud interference intelligent identification method based on time-frequency features.

[0027] An electronic device comprising:

[0028] A memory for saving a computer program;

[0029] A processor for executing the computer program to realize an instruction tracking method of a composite fraud interference intelligent identification method based on time-frequency features.

[0030] The beneficial effects of the present application are:

[0031] The present application firstly carries out pulse compression and coherent accumulation processing on the composite deception jamming signal to obtain the Doppler unit where the maximum energy spectrum of the jamming signal is located, then only carries out coherent accumulation processing on the jamming signal, extracts the time domain slice of the Doppler unit where the maximum energy spectrum is located, so that the jamming signal after improving the jam-to-noise ratio can be obtained; has good generalization ability under the condition of dynamically superimposing multiple deception jammers; can greatly improve the recognition accuracy of composite deception jamming under low jam-to-noise ratio, provides reliable technical support for the field of radar anti-jamming; easy to implement in engineering, fast. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 is the overall flowchart of the present application;

[0033] Figure 2 is the time-frequency feature extraction flowchart;

[0034] Figure 3 is the YOLOv8 detection flowchart;

[0035] Figure 4 is the slice reconstruction jamming range-Doppler plot;

[0036] Figure 5 is the slice reconstruction jamming time-frequency plot;

[0037] Figure 6 is the intermittent sampling direct forwarding jamming range-Doppler plot;

[0038] Figure 7 is the intermittent sampling direct forwarding jamming time-frequency plot;

[0039] Figure 8 is the intermittent sampling cyclic forwarding jamming range-Doppler plot;

[0040] Figure 9 is the intermittent sampling cyclic forwarding jamming time-frequency plot;

[0041] Figure 10 is the comb spectrum jamming range-Doppler plot;

[0042] Figure 11 is the comb spectrum jamming time-frequency plot. DETAILED DESCRIPTION

[0043] The present application will be further described below in conjunction with the drawings.

[0044] The present application is a kind of composite deception jamming intelligent recognition system based on time-frequency feature, contains interference signal simulation module, composite deception jamming generation module, composite deception jamming signal preprocessing module, feature extraction module, data set construction module, input module, test set generation module, evaluation module, the functions of each module are as follows:

[0045] Interference signal simulation module: simulate the slice reconstruction interference of M pulses, intermittent sampling direct forwarding interference, intermittent sampling cycle forwarding interference, comb spectrum interference, assuming that the radar pulse repetition frequency is 3 kHz, the working frequency is 10 GHz, the signal pulse width is 70 μs, the bandwidth is 25 MHz, the sampling frequency is 60 MHz, and the number of coherent accumulation pulses M is 64.

[0046] Composite deception jamming generation module: additively composite the four kinds of deception jamming signals to generate 11 kinds of composite deception jamming signals.

[0047] Composite deception jamming signal preprocessing module: pulse compression and coherent accumulation processing are performed on the M composite deception jamming signals in one CPI, maximum value search is performed on the obtained range-Doppler domain results to obtain the Doppler unit f where the jamming is located; only the coherent accumulation processing is performed on the M composite deception jamming signals of one CPI, and the time domain slice of the Doppler unit f where the composite deception jamming is located is extracted from the obtained results.

[0048] Feature extraction module: short-time Fourier transform processing is performed on the extracted time domain slice, the time domain slice is first framed and windowed, then discrete Fourier transform is performed on the windowed signal, and finally the transformed results are summed to generate a time-frequency graph.

[0049] Dataset construction module: change the jamming-to-noise ratio, distance and velocity of the interference to obtain a composite deception jamming time-frequency graph dataset.

[0050] Input module: input the constructed dataset into the YOLOv8 network, use the Mosaic data enhancement technology to synthesize multiple training samples in one time-frequency graph, and improve the adaptability of the model to complex interference signal scenes. For the interference detection task, the data enhancement method can transform the original sample to cover more scenes, thereby enhancing the robustness of the algorithm, reducing the overfitting situation, and improving the generalization ability of the model. Backbone network part: use C2f module to realize further lightweight. Detection head part: use SPPF module and FPN+PAN module, and also replace the C3 module in YOLOv5 with C2f module with richer gradient flow, enhance the network feature fusion capability, and adopt the structure of decoupled head, two parallel branches respectively extract the class features and position features of the interference signal, and then use a layer of 1×1 convolution to complete the classification and positioning tasks of the interference signal.

[0051] Test set generation module: generate composite deception jamming signal time-frequency graphs under different jamming-to-noise ratios in the range of -28 to -15 dB as test sets.

[0052] The evaluation module: the present application adopts precision (P), recall (R) and mAP50-95 as the performance evaluation indexes of the composite deception jamming recognition classification. The higher P represents that the detection result is more reliable and the false detection is less; the higher R represents that the algorithm can detect all the jamming signals as much as possible and the missed detection is less; the larger the value of mAP represents the higher the detection accuracy of the algorithm. The calculation formulas of P, R and mAP are as follows:

[0053]

[0054] Wherein, TP is the number of correct interference signal detection, FP is the number of false detection of interference signal, and FN represents the number of missed detection. AP is the accuracy obtained from the area of the P-R curve. mAP is the average accuracy of all interference signal categories. That is, the precision (P) represents the proportion of real positive samples in the model predicted as positive samples, that is, the ratio of the number of correct prediction as positive samples to the number of all prediction as positive samples; the recall (R) represents the ratio of the number of correct prediction as positive samples by the model to the number of real positive samples, that is, the ratio of the number of positive samples successfully found by the model to the number of all real positive samples; and mAP50-95 represents the average accuracy of the model in the IoU threshold range of 50% to 95%.

[0055] According to Figure 1 , the present application proposes a composite deception jamming intelligent recognition method based on time-frequency characteristics, and the specific steps include:

[0056] Step S1: simulate multiple pulse interference signals;

[0057] Step S2: pre-process the interference signals to generate a data set;

[0058] Step S3: train a YOLOv8 neural network inference detection model based on the generated data set;

[0059] Step S4: detect the composite deception jamming under different jamming-to-noise ratios based on the trained model.

[0060] The system input is a radar receiving signal containing composite jamming. The radar signal is pre-processed. First, pulse compression and coherent accumulation processing are performed to obtain the Doppler unit where the signal maximum energy spectrum is located. Then, only coherent accumulation processing is performed on the signal, the time domain slice of the Doppler unit where the maximum energy spectrum is located is extracted, feature analysis is performed, and the time-frequency joint distribution graph feature is selected. The time domain slices of single-class interference and composite interference are subjected to short-time Fourier transform to obtain time-frequency graphs as the input data set of the YOLOv8 neural network for training, and the features of the interference signals in the time-frequency graphs are extracted. The composite deception jamming signals are identified by using the automatically extracted network features, the precision, recall and average accuracy are given, and the positioning of the interference signals is completed by using the anchor frame.

[0061] The specific implementation steps are as follows:

[0062] Step S1: simulate a plurality of pulse interference signals, refer to the accompanying drawings Figure 2 .

[0063] Further, simulate M pulse slice reconstruction interference, intermittent sampling direct forwarding interference, intermittent sampling cycle forwarding interference, comb spectrum interference, assume that the radar pulse repetition frequency is 3kHz, the working frequency is 10GHz, the signal pulse width is 70μs, the bandwidth is 25MHz, the sampling frequency is 60MHz, and the number of coherent accumulation pulses M is 64.

[0064] Assume that the radar transmitted signal is a linear frequency modulation (LFM) signal:

[0065]

[0066] In the formula, T p is the signal pulse width; f c is the signal carrier frequency; μ=B / T p is the frequency modulation slope of the signal; and B is the signal bandwidth.

[0067] Slice reconstruction interference is a kind of dense false target interference, and its principle is: the jammer stores the received radar transmitted signal, reads the signal and uniformly and equally spaced samples, and then copies the sampled signal for forwarding. The radar received one coherent processing interval (CPI) interference signal echo baseband signal is:

[0068]

[0069] In the formula, t,t m are fast time and slow time respectively, t m =mT r , m=0, 1, …, M-1, M is the number of coherent accumulation of interference signals, T r is the pulse repetition interval; R(t m )=R-v t t m is the radial distance function of the jammer and the radar, R is the initial distance, v t is the radial velocity of the jammer, c is the speed of light, and f(t m ) is the Doppler frequency function modulated by the jammer.

[0070] Intermittent sampling direct forwarding interference refers to that between two sampling operations, the jammer directly forwards the sampled signal. The model of M intermittent sampling direct interference signals in fast-slow time domain is:

[0071] J2(t,m·T r )=[J2(t,0·T r ),J2(t,1·T r ),…,J2(t,(M-1)·T r )]

[0072] In the formula, m = 0, 1, ..., M-1.

[0073] Intermittent sampling cyclic forwarding processes and forwards sampled signals sequentially according to a pre-defined order. The model for M intermittent sampling cyclic interference signals in the fast-slow time domain is as follows:

[0074] J3(t,m·T r )=[J3(t,0·T r ),J3(t,1·T r ),…,J3(t,(M-1)·T r )]

[0075] In the formula, m = 0, 1, ..., M-1.

[0076] Comb spectrum interference primarily involves the product modulation of radar transmitted signals by comb spectrum signals. The resulting comb spectrum interference signals are deceptive and suppressive. The model for M comb spectrum interference echo signals in the fast-slow time domain is as follows:

[0077] J4(t,m·T r )=

[0078] [J4(t,0·T r ),J4(t,1·T r ),…,J4(t,(M-1)·T r )]

[0079] Step S2: Preprocess the interference signal to generate a dataset. (See attached document.) Figure 2 The module for extracting the time-frequency distribution features of fast-slow time-domain interference signals.

[0080] Furthermore, the four types of deception interference signals are additively combined to generate 11 composite deception interference signals.

[0081] Furthermore, pulse compression and coherent accumulation are performed on M composite spoofing interference signals within a single CPI. A maximum value search is then performed on the obtained range-Doppler domain results to obtain the Doppler cell f where the interference resides. Coherent accumulation is then performed only on the M composite spoofing interference signals within a single CPI, and the time-domain slice of the Doppler cell f where the composite spoofing interference resides is extracted from the result. (Refer to the appendix...) Figure 4 , 6 8 and 10.

[0082] Further, the extracted time domain slices are subjected to short-time Fourier transform processing, first the time domain slices are subjected to frame windowing, then the windowed signal is subjected to discrete Fourier transform, and finally the transformed results are summed to generate a time-frequency graph, the clutter-to-noise ratio, the distance and the speed of the interference are changed, and a composite deception jamming time-frequency graph dataset is obtained, refer to the attached Figure 5 、 7 , 9 and 11.

[0083] Step S3: training the YOLOv8 neural network inference detection model based on the generated dataset, refer to the attached Figure 3 .

[0084] Further, the constructed dataset is input into the YOLOv8 network, and a plurality of training samples are synthesized in one time-frequency graph by using the Mosaic data enhancement technology, which improves the adaptability of the model to complex scenes of interference signals. For the interference detection task, the data enhancement method can transform the original sample to cover more scenes, thereby enhancing the robustness of the algorithm, reducing the overfitting condition, and improving the generalization ability of the model.

[0085] Further, the backbone network uses a C2f module to achieve further lightweight.

[0086] Further, the detection head adopts an SPPF module and an FPN+PAN module, and at the same time, the C3 module in YOLOv5 is replaced by a C2f module with richer gradient flow, which enhances the network feature fusion capability, and adopts a decoupling head structure, two parallel branches respectively extract the class features and position features of the interference signal, and then each uses a layer of 1x1 convolution to complete the classification and positioning tasks of the interference signal.

[0087] Step S4: detecting the composite deception jamming under different clutter-to-noise ratios based on the trained model, refer to the attached Figure 3 .

[0088] Further, the composite deception jamming signal time-frequency graph under different clutter-to-noise ratios is generated as a test set in the range of -28 to -15 dB of the clutter-to-noise ratio.

[0089] Further, the present application adopts precision (P), recall rate (R) and mAP50-95 as the performance evaluation indexes of composite deception jamming recognition classification. The higher the P is, the more reliable the detection result is, and the less the false detection is; the higher the R is, the more the algorithm can detect all the interference signals, and the less the missed detection is; the larger the value of mAP is, the higher the detection accuracy of the algorithm is. The calculation formulas of P, R and mAP are as follows:

[0090]

[0091] Wherein, TP is the number of correct detection of interference signals, FP is the number of false detection of interference signals, FN represents the number of missed detection. AP is the accuracy, which is obtained from the area of P-R curve. mAP is the average accuracy of all interference signal categories. That is, the precision (P) represents the proportion of true positive samples in the positive samples predicted by the model, that is, the ratio of the number of correctly predicted positive samples to the number of all predicted positive samples; the recall rate (R) represents the proportion of the number of correctly predicted positive samples to the number of true positive samples, that is, the ratio of the number of positive samples successfully found by the model to the number of all true positive samples; mAP50-95 represents the average accuracy of the model in the IoU threshold range of 50% to 95%.

[0092] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A time-frequency feature-based composite deception jamming intelligent identification method, characterized in that: The specific steps are as follows: Step 1: Use the interference signal simulation module to simulate the slice reconstruction interference of M pulse interference signals, intermittent sampling direct forwarding interference, intermittent sampling cyclic forwarding interference, and comb spectrum interference; Step 2: Use the composite deception jamming generation module and the composite deception jamming signal preprocessing module to preprocess the interference signals obtained in step 1, and then use the feature extraction module and the data set construction module to extract features from the interference signals to generate a data set; Step 2.1: Additively combine the four kinds of deception jamming signals described in step 1 by using the composite deception jamming generation module to generate 11 kinds of composite deception jamming signals; Step 2.2: Pulse compression and coherent accumulation are performed on the M complex deceptive jamming signals in one CPI by the complex deceptive jamming signal preprocessing module, maximum value search is performed on the obtained range-Doppler domain results, and the Doppler unit where the jamming is located is obtained ; Only M composite deception jamming signals of one CPI are accumulated in phase to obtain the Doppler unit where the composite deception jamming signal is located Time domain slice Step 2.3: The extracted time domain slices are processed by the feature extraction module and the data set construction module. First, the time domain slices are framed and windowed, then the windowed signals are subjected to discrete Fourier transform, and finally the transformed results are summed to generate a time-frequency graph. By changing the jam-to-noise ratio, the distance and speed of the interference, the composite deception jamming time-frequency graph data set is obtained; Step 3: Use the input module to input the data set generated in step 2 into the YOLOv8 neural network inference detection model and train it; Step 4: Based on the model trained in step 3, use the test set generation module and the evaluation module to detect composite deception jamming under different jam-to-noise ratios.

2. The method of claim 1, wherein the method comprises: The radar pulse repetition frequency in step 1 is 3 kHz, the working frequency is 10 GHz, the signal pulse width is , the bandwidth is 25 MHz, the sampling frequency is 60 MHz, and the number of coherent cumulative pulses M is 64.

3. The method of claim 1, wherein the method further comprises: In step 3, the data set constructed in step 2 is input into the YOLOv8 network through the input module. Mosaic data augmentation technology is used to synthesize multiple training samples in one time-frequency graph. The backbone network uses C2f module. The detection head uses SPPF module and FPN+PAN module. The C3 module in YOLOv5 is replaced by C2f module to enhance the network feature fusion capability. The structure of decoupling head is adopted to make two parallel branches extract the class features and position features of the interference signal respectively, and then use a 1x1 convolution layer to complete the classification and positioning of the interference signal.

4. The method of claim 1, wherein the method further comprises: Step 4 specifically includes: Step 4.1: Use the test set generation module to generate composite deception jamming signal time-frequency graphs under different jam-to-noise ratios in the range of -28~-15dB as test sets; Step 4.2: Use the evaluation module and take precision P, recall R and mAP50-95 as performance evaluation indexes for composite deception jamming recognition and classification. The higher the P value, the more reliable the detection result and the fewer the false positives. The higher the R value, the more interference signals the algorithm can detect and the fewer the false negatives. The larger the mAP value, the higher the detection accuracy of the algorithm.

5. The method of claim 4, wherein the method further comprises: The calculation formulas of P, R and mAP are as follows: ； ； ； ； wherein, is the number of correct detections of interference signals, is the number of false detections of interference signals, represents the number of missed detections; is the precision, calculated by the area of the curve; is the average precision of all interference signal classes; precision P represents the proportion of true positive samples in the positive samples predicted by the model, i.e., the ratio of the number of correctly predicted positive samples to the number of all predicted positive samples; recall R represents the proportion of the number of correctly predicted positive samples to the number of true positive samples, i.e., the ratio of the number of positive samples successfully found by the model to the number of all true positive samples; mAP50-95 represents the average precision of the model in the IoU threshold range of 50%~95%.

6. A computer apparatus / device / system comprising a memory, a processor, and a computer program stored on the memory, characterized in that: The processor executes the computer program to realize the steps of the method of any one of claims 1 to 5.

7. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that: The computer program / instructions are executed by the processor to realize the steps of the method of any one of claims 1 to 5.

8. A computer program product comprising computer programs / instructions, characterized in that: The computer program / instructions are executed by the processor to realize the steps of the method of any one of claims 1 to 5.

9. An electronic device, comprising: It includes: Memory for saving computer programs; Processor for executing the computer program to realize the method of any one of claims 1 to 5.

Images