Composite jamming recognition method based on priori matrix embedding

Abstract

A composite interference recognition method based on prior matrix embedding, comprising: obtaining multi-domain information of radar echo signals before and after pulse compression and after moving target detection; obtaining a feature vector through convolution processing and splicing; performing discriminant probability calculation on multiple channels to obtain interference classification results respectively; and performing weighted fusion on the classification results by using a prior matrix to output a multi-label composite interference recognition result. The present application can simultaneously recognize multiple interferences in a composite scene, embed a prior matrix into a fusion link, differently weight the contribution of interferences according to channels, improve recognition accuracy under low signal-to-noise ratio, is compatible with mainstream radar processing procedures, and can be directly used in a main and auxiliary radar cooperative node.

Description

Technical Field

[0001] This invention relates to the field of radar jamming signal identification technology, specifically to a composite jamming identification method based on prior matrix embedding, which performs differentiated weighted fusion of identification results from multiple transform domain channels to achieve composite jamming type identification. Background Technology

[0002] With the rapid development of electronic warfare technology, the electromagnetic environment of modern battlefields is becoming increasingly complex, and radar systems face more diverse forms of active interference. In a primary and secondary radar node collaborative system, effective interference type identification is a prerequisite for implementing targeted suppression measures and fully leveraging collaborative effectiveness. Secondary radar nodes have relatively fewer antenna elements and wider beam main lobe angles. Multiple active interference signals often act simultaneously within the same beam, resulting in complex interference phenomena. Compared to primary nodes, secondary nodes are more susceptible to the superposition of various interference patterns, placing higher demands on the accuracy and robustness of the interference identification system.

[0003] Composite interference is more complex in terms of time-frequency representation and energy distribution. Traditional single-label classification networks can only output one category label at a time, which is insufficient to meet the requirement of simultaneous identification of multiple interferences in composite interference scenarios. Composite interference is more complex than single interference in terms of time-frequency representation and energy distribution.

[0004] Existing radar interference identification methods are mainly divided into two categories. The first category is traditional identification methods based on artificial features. These methods extract manually designed statistical features from a single transform domain, such as the time domain, frequency domain, or time-frequency domain, and then use classifiers such as support vector machines and decision trees to determine the interference type. While these methods are simple to implement, feature construction relies on a single transform domain, failing to effectively utilize the differences and complementarities between multiple transform domains. Furthermore, traditional classifiers output single-category labels, which cannot meet the need for simultaneous identification of multiple interference types in complex interference scenarios. The second category is identification methods based on deep learning. These methods automatically extract signal features through structures such as convolutional neural networks, achieving good results in single interference identification tasks. However, existing methods generally use uniform weighting when fusing multi-domain features, failing to utilize the differences in the contribution of each transform domain to the discrimination of different interference categories for differentiated weighting. Therefore, the identification accuracy remains significantly insufficient in low-noise-interference-ratio complex interference scenarios.

[0005] Therefore, how to design new identification methods for composite interference to utilize multi-transform domain feature information, support composite interference and perform intelligent identification has become a technical problem that needs to be solved by existing technologies. Summary of the Invention

[0006] The purpose of this invention is to propose a composite interference identification method based on prior matrix embedding. Based on multi-transform domain feature information, a prior matrix established using different interference features is used to achieve composite interference identification.

[0007] To achieve this objective, the present invention adopts the following technical solution: A method for identifying composite interference based on prior matrix embedding, characterized by comprising the following steps: Multi-transform domain dataset construction step S110: Acquire radar interference echo signals containing multiple pulses over a specific time series. The radar interference echo signal is processed by pulse compression and moving target detection to obtain the feature information of seven types of transform domains before and after pulse compression, including time domain, frequency domain, time-frequency domain, and moving target detection, as a dataset. Multi-channel combined feature extraction step S120: The feature information of the seven transform domains is extracted by convolution of corresponding dimensions and unified into vectors of equal length; the combined features of the seven channels are obtained by combining seven-bit binary masks. ; Multi-channel independent classification probability calculation step S130: The combined features of the seven channels are fed into independent classification branches, mapped through a fully connected layer and activated by a sigmoid function to obtain the pairs of all channels. The predicted probability of each type of interference is calculated, and the final output is a seven-channel independent probability vector of the interference category. Each channel's probability vector contains a pair of... The predicted probability of interference; Composite interference identification step S140: With prior matrix As fusion weights, for each type of interference, the predicted probabilities of all seven transform domain combined channels are weighted and fused in one go using a prior weight matrix to obtain the comprehensive confidence level of that type of interference. A threshold is then used to discriminate the interference result based on this comprehensive confidence level. The interference types are sequentially subjected to comprehensive confidence calculation and interference result discrimination to obtain the composite interference type identification result.

[0008] Optionally, in step S110, Assuming radar interference echo signal The time series length is ,Include Multi-pulse interference echo signal; radar echo signal Time-domain features are used to extract the real and imaginary parts of the signal, resulting in... , radar echo signal Frequency domain features, after Fourier transform, extract the real and imaginary parts of the signal to obtain... , radar echo signal The time-frequency domain features are analyzed, and after short Fourier transform, the real and imaginary parts of the signal are extracted to obtain... , The original radar echo signal is pulse-compressed to obtain the pulse-compressed signal. : Radar echo signal after pulse compression Time-domain features are used to extract the real and imaginary parts of the signal, resulting in... , Radar echo signal after pulse compression Frequency domain features, after Fourier transform, extract the real and imaginary parts of the signal to obtain... , Radar echo signal after pulse compression The time-frequency domain features are analyzed, and after short Fourier transform, the real and imaginary parts of the signal are extracted to obtain... , pulse compression signal Moving target detection processing is performed to obtain the range Doppler domain. The real and imaginary parts of the signal are extracted to obtain the range Doppler domain features. .

[0009] Optionally, in step S120, One-dimensional features in the time and frequency domains are extracted using one-dimensional convolution Conv1d, while two-dimensional features in the time-frequency domain and the range-Doppler domain are extracted using two-dimensional convolution Conv2d. All single-domain features are then unified into equal-length feature vectors after global pooling.

[0010] Optionally, step S120 specifically includes: First, convolutional feature extraction is performed on each transform domain separately to obtain single-domain feature vectors of the same dimension: , in For the convolutional feature extraction network corresponding to the transform domain, For the same feature dimension, all seven single-domain features are concatenated to obtain the basic vector:

[0011] Apply a seven-bit binary mask to each channel Indicates whether to use the feature information carried by different transform domains. Indicates the first Should this combination be included in the first? The characteristics of the transform domain are obtained to obtain the first Combined characteristics of each channel: .

[0012] Optionally, in step S130, The fully connected layer is mapped as follows: Features of each channel Logistic regression output is obtained through independent fully connected layer mapping. : , Number of interference categories; in and For the first Learnable parameters of a channel classifier; The Sigmoid activation is: Logistic regression output After Sigmoid activation, this channel is obtained for all... Predicted probability of interference: , For each type of interference, each channel independently outputs its own probability vector, resulting in a total of independent interference category probability vectors across all seven channels. Each of them .

[0013] Optional, each Length is , including The probability of each type of interference.

[0014] Optionally, the prior matrix Regarding the first The interference type has a weight sum of 1. The prior matrix is ​​constructed based on the prior of multi-transform domain feature statistical analysis, or it is a uniform weight matrix. The prior matrix The The term represents the first Transformation domain combination pairs of the first Prior discriminant contribution of interference-like phenomena. For each interference category ,by The The column serves as the fusion weight for the 7 channels, among which .

[0015] Optionally, in step S140, For each type of interference The predicted probabilities of the 7 channels are then fused according to their weights: , All The overall prediction probability vector is obtained by summing the comprehensive prediction probabilities of various types of interference: , right Each component is determined by setting a decision threshold. Perform independent binary decision for each type of interference: , Finally, the composite interference multi-label recognition result vector is obtained. : , in Indicates the existence of the first Interference-like This indicates that the interference does not exist, enabling the identification of each type of interference in composite interference.

[0016] The present invention further discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described composite interference identification method based on prior matrix embedding.

[0017] The present invention also discloses a composite interference identification device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the composite interference identification method based on prior matrix embedding described above.

[0018] In summary, the present invention has the following advantages: 1. This invention adopts a multi-label classification architecture and multi-transform domain joint input, which can simultaneously identify multiple interferences in complex scenarios, solve the problem that traditional single-label networks can only output a single category and do not make full use of features, and the identification is more in line with the actual adversarial environment.

[0019] 2. This invention embeds the prior matrix into the fusion process and weights the data differently according to the contribution of each channel to the interference. This is superior to uniform weighting and pure data-driven methods, improving the recognition accuracy under low interference-to-noise ratio and enhancing the interpretability and robustness of the model.

[0020] 3. This invention extracts features along the radar signal processing link, is compatible with mainstream radar processing procedures, and can be directly used in the main and secondary radar collaborative nodes, with better engineering implementation and migration capabilities. Attached Figure Description

[0021] Figure 1 This is a flowchart of a composite interference identification method based on prior matrix embedding according to a specific embodiment of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0023] The main aspects of this invention are: acquiring multi-domain information of radar echo signals before and after pulse compression and after moving target detection; obtaining feature vectors through convolution processing and concatenation; calculating the discrimination probability of multiple channels to obtain interference classification results; and using a prior matrix to weightedly fuse the classification results to output multi-label composite interference identification results. Thus, by fully utilizing multi-transform domain information and taking into full account the prior discrimination information of various types of interference, all types of interference present in the current signal can be simultaneously determined in a single inference.

[0024] For details, see Figure 1 The flowchart illustrates a composite interference identification method based on prior matrix embedding according to a specific embodiment of the present invention, including the following steps: Multi-transform domain dataset construction step S110: Acquire radar interference echo signals containing multiple pulses over a specific time series. The radar interference echo signal is processed by pulse compression and moving target detection to obtain feature information in the time domain, frequency domain, time-frequency domain and after moving target detection, totaling seven types of transform domains, as a dataset.

[0025] In one specific embodiment Assuming radar interference echo signal The time series length is ,Include Multi-pulse interference echo signal; radar echo signal Time-domain features are used to extract the real and imaginary parts of the signal, resulting in... , , radar echo signal Frequency domain features, after Fourier transform, extract the real and imaginary parts of the signal to obtain... , Specifically, after Fourier transform, the frequency domain features are obtained. : , The frequency domain characteristics of the radar echo signal were obtained: .

[0026] radar echo signal The time-frequency domain features are analyzed, and after short Fourier transform, the real and imaginary parts of the signal are extracted to obtain... .

[0027] Specifically, short-time Fourier transform is performed to obtain time-frequency domain information. : , in For length is The time window function, For time indexing, For frequency indexing.

[0028] Time-frequency domain characteristics of radar echoes: .

[0029] The original radar echo signal is pulse-compressed to obtain the pulse-compressed signal. :

[0030] in To transmit linear frequency modulated signals The corresponding matched filter impulse response.

[0031] Radar echo signal after pulse compression Time-domain features are used to extract the real and imaginary parts of the signal, resulting in... , , Radar echo signal after pulse compression Frequency domain features, after Fourier transform, extract the real and imaginary parts of the signal to obtain... , Specifically, after Fourier transform, the frequency domain features are obtained. : , The frequency domain characteristics of the radar echo signal were obtained: .

[0032] Radar echo signal after pulse compression The time-frequency domain features are analyzed, and after short Fourier transform, the real and imaginary parts of the signal are extracted to obtain... .

[0033] Specifically, short-time Fourier transform is performed to obtain time-frequency domain information. : , in For length is The time window function, For time indexing, For frequency indexing.

[0034] Time-frequency domain characteristics of radar echoes: .

[0035] pulse compression signal Moving target detection processing is performed to obtain the range Doppler domain. The real and imaginary parts of the signal are extracted to obtain the range Doppler domain features. .

[0036] Specifically, for the radar echo signal after pulse compression The range Doppler domain output is obtained by performing moving target detection processing. : , For Doppler cell indexing, The number of points for FFT along the slow time dimension, i.e. the number of coherent accumulation pulses, corresponds to the radial velocity information of the target; For window functions, For distance cell index, This represents the number of fast-time sampling points after pulse compression, corresponding to the radial distance information of the target.

[0037] Obtain the range Doppler domain features

[0038] .

[0039] Multi-channel combined feature extraction step S120: The feature information of the seven transform domains is extracted by convolution of corresponding dimensions and unified into vectors of equal length; the combined features of the seven channels are obtained by combining seven-bit binary masks. .

[0040] Specifically, one-dimensional convolution Conv1d is used to extract one-dimensional features in the time and frequency domains, and two-dimensional convolution Conv2d is used to extract two-dimensional features in the time-frequency domain and the distance-Doppler domain. All single-domain features are then unified into equal-length feature vectors after global pooling.

[0041] In one specific embodiment First, convolutional feature extraction is performed on each transform domain separately to obtain single-domain feature vectors of the same dimension: , in For the convolutional feature extraction network corresponding to the transform domain, For the same feature dimension, and are selectable hyperparameters. Concatenate all seven single-domain features to obtain the base vector:

[0042] Then, a seven-bit binary mask is applied to each channel. Indicates whether to use the feature information carried by different transform domains. Indicates the first Should this combination be included in the first? The characteristics of the transform domain are obtained to obtain the first Combined characteristics of each channel: .

[0043] Binary mask combination Specifically: .

[0044] Multi-channel independent classification probability calculation step S130: The combined features of the seven channels are fed into independent classification branches, mapped through a fully connected layer and activated by a sigmoid function to obtain the pairs of all channels. The predicted probability of each type of interference is calculated, and the final output is a seven-channel independent probability vector of the interference category. Each channel's probability vector contains a pair of... The predicted probability of interference.

[0045] In one specific embodiment, the fully connected layer is mapped as follows: Features of each channel Logistic regression output is obtained through independent fully connected layer mapping. : , Number of interference categories; in and For the first Learnable parameters for a channel classifier.

[0046] In an optional embodiment, the solution is as follows: Let the signal sample set be... ,in The total number of samples, For the first The original received signal of each sample, Labels for interference categories, This represents the total number of interference categories. and Obtained through supervised training. The model optimizes all parameters by minimizing the following loss function. :

[0047] in The cross-entropy loss function is used, and the parameters are calculated using the gradient through backpropagation and iteratively updated using the Adam optimizer.

[0048] The Sigmoid activation is: Logistic regression output After Sigmoid activation, this channel is obtained for all... Predicted probability of interference: , For each type of interference, each channel independently outputs its own probability vector, resulting in a total of independent interference category probability vectors across all seven channels. Each of them .

[0049] In an optional embodiment, each Length is , including The probability of each type of interference.

[0050] Composite interference identification step S140: With prior matrix As fusion weights, for each type of interference, the predicted probabilities of all seven transform domain combined channels are weighted and fused in one go using a prior weight matrix to obtain the comprehensive confidence level of that type of interference. A threshold is then used to discriminate the interference result based on this comprehensive confidence level. The interference types are sequentially subjected to comprehensive confidence calculation and interference result discrimination to obtain the composite interference type identification result.

[0051] The prior matrix Regarding the first The interference class has a weight sum of 1.

[0052] The prior matrix is ​​constructed based on prior statistical analysis of multi-transformation domain features, or it is a uniform weight matrix.

[0053] In one specific embodiment The The term represents the first Transformation domain combination pairs of the first Prior discriminant contribution of interference-like phenomena. For each interference category ,by The The column serves as the fusion weight for the 7 channels, among which , For each type of interference The predicted probabilities of the 7 channels are then fused according to their weights: , All The overall prediction probability vector is obtained by summing the comprehensive prediction probabilities of various types of interference: , right Each component is determined by setting a decision threshold. Perform independent binary decision for each type of interference: , Finally, the composite interference multi-label recognition result vector is obtained. : , in Indicates the existence of the first Interference-like This indicates that the interference does not exist, enabling the identification of each type of interference in composite interference.

[0054] The present invention further discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described composite interference identification method based on prior matrix embedding.

[0055] The present invention also discloses a composite interference identification device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the composite interference identification method based on prior matrix embedding described above.

[0056] It should be noted that in step S110, the set Represented as a set of real numbers, this is a well-known method of use in mathematics. The parameters appearing in steps S130 and S140... This indicates the number of interference categories. Since they are different types, they do not conflict.

[0057] In summary, the present invention has the following advantages: 1. This invention adopts a multi-label classification architecture and multi-transform domain joint input, which can simultaneously identify multiple interferences in complex scenarios, solve the problem that traditional single-label networks can only output a single category and do not make full use of features, and the identification is more in line with the actual adversarial environment.

[0058] 2. This invention embeds the prior matrix into the fusion process and weights the data differently according to the contribution of each channel to the interference. This is superior to uniform weighting and pure data-driven methods, improving the recognition accuracy under low interference-to-noise ratio and enhancing the interpretability and robustness of the model.

[0059] 3. This invention extracts features along the radar signal processing link, is compatible with mainstream radar processing procedures, and can be directly used in the main and secondary radar collaborative nodes, with better engineering implementation and migration capabilities.

[0060] Obviously, those skilled in the art will understand that the various units or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device, or alternatively, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by the computing device. Alternatively, they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0061] The above description is a further detailed explanation of the present invention in conjunction with specific preferred embodiments. It should not be considered that the specific embodiments of the present invention are limited to this. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention as defined by the submitted claims.

Claims

1. A method for identifying composite interference based on prior matrix embedding, characterized in that, Includes the following steps: Multi-transform domain dataset construction step S110: Acquire radar interference echo signals containing multiple pulses over a specific time series. The radar interference echo signal is processed by pulse compression and moving target detection to obtain the feature information of seven types of transform domains before and after pulse compression, including time domain, frequency domain, time-frequency domain, and moving target detection, as a dataset. Multi-channel combined feature extraction step S120: The feature information of the seven transform domains is extracted by convolution of corresponding dimensions and unified into vectors of equal length; the combined features of the seven channels are obtained by combining seven-bit binary masks. ; Multi-channel independent classification probability calculation step S130: The combined features of the seven channels are fed into independent classification branches, mapped through a fully connected layer and activated by a sigmoid function to obtain the pairs of all channels. The predicted probability of each type of interference is calculated, and the final output is a seven-channel independent probability vector of the interference category. Each channel's probability vector contains a pair of... The predicted probability of interference; Composite interference identification step S140: With prior matrix As fusion weights, for each type of interference, the predicted probabilities of all seven transform domain combined channels are weighted and fused in one go using a prior weight matrix to obtain the comprehensive confidence level of that type of interference. A threshold is then used to discriminate the interference result based on this comprehensive confidence level. The interference types are sequentially subjected to comprehensive confidence calculation and interference result discrimination to obtain the composite interference type identification result.

2. The composite interference identification method according to claim 1, characterized in that: In step S110, Assuming radar interference echo signal The time series length is ,Include Multi-pulse interference echo signal; radar echo signal Time-domain features are used to extract the real and imaginary parts of the signal, resulting in... , radar echo signal Frequency domain features, after Fourier transform, extract the real and imaginary parts of the signal to obtain... , radar echo signal The time-frequency domain features are analyzed, and after short Fourier transform, the real and imaginary parts of the signal are extracted to obtain... , The original radar echo signal is pulse-compressed to obtain the pulse-compressed signal. : Radar echo signal after pulse compression Time-domain features are used to extract the real and imaginary parts of the signal, resulting in... , Radar echo signal after pulse compression Frequency domain features, after Fourier transform, extract the real and imaginary parts of the signal to obtain... , Radar echo signal after pulse compression The time-frequency domain features are analyzed, and after short Fourier transform, the real and imaginary parts of the signal are extracted to obtain... , pulse compression signal Moving target detection processing is performed to obtain the range Doppler domain. The real and imaginary parts of the signal are extracted to obtain the range Doppler domain features. .

3. The composite interference identification method according to claim 2, characterized in that: In step S120, One-dimensional features in the time and frequency domains are extracted using one-dimensional convolution Conv1d, while two-dimensional features in the time-frequency domain and the range-Doppler domain are extracted using two-dimensional convolution Conv2d. All single-domain features are then unified into equal-length feature vectors after global pooling.

4. The composite interference identification method according to claim 3, characterized in that: Step S120 is as follows: First, convolutional feature extraction is performed on each transform domain separately to obtain single-domain feature vectors of the same dimension: ， in For the convolutional feature extraction network corresponding to the transform domain, For the same feature dimension, all seven single-domain features are concatenated to obtain the basic vector: Apply a seven-bit binary mask to each channel Indicates whether to use the feature information carried by different transform domains. Indicates the first Should this combination be included in the first? The characteristics of the transform domain are obtained to obtain the first Combined characteristics of each channel: 。 5. The composite interference identification method according to claim 4, characterized in that: In step S130, The fully connected layer is mapped as follows: Features of each channel Logistic regression output is obtained through independent fully connected layer mapping. : , Number of interference categories; in and For the first Learnable parameters of a channel classifier; The Sigmoid activation is: Logistic regression output After Sigmoid activation, this channel is obtained for all... Predicted probability of interference: ， For each type of interference, each channel independently outputs its own probability vector, resulting in a total of independent interference category probability vectors across all seven channels. Each of them .

6. The composite interference identification method according to claim 5, characterized in that: Each Length is , including The probability of each type of interference.

7. The composite interference identification method according to claim 5, characterized in that: The prior matrix Regarding the first The interference type has a weight sum of 1. The prior matrix is ​​constructed based on the prior of multi-transform domain feature statistical analysis, or it is a uniform weight matrix. The prior matrix The The term represents the first Transformation domain combination pairs of the first Prior discriminant contribution of interference-like phenomena. For each interference category ,by The The column serves as the fusion weight for the 7 channels, among which .

8. The composite interference identification method according to claim 7, characterized in that: In step S140, For each type of interference The predicted probabilities of the 7 channels are then fused according to their weights: ， All The overall prediction probability vector is obtained by summing the comprehensive prediction probabilities of various types of interference: ， right Each component is determined by setting a decision threshold. Perform independent binary decision for each type of interference: ， Finally, the composite interference multi-label recognition result vector is obtained. : ， in Indicates the existence of the first Interference-like This indicates that the interference does not exist, enabling the identification of each type of interference in composite interference.

9. A computer-readable storage medium having a computer program stored thereon, When the computer program is executed by the processor, it implements the steps of the composite interference identification method based on prior matrix embedding as described in any one of claims 1-8.

10. A composite interference identification device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the composite interference identification method based on prior matrix embedding as described in any one of claims 1-8.

Images