Time frequency-image fusion distance joint azimuth SAR active interference identification method and device

Abstract

The invention relates to a time frequency-image fusion distance joint azimuth SAR active interference identification method and device. The method comprises the following steps: inputting radar interference echo data to be identified into a trained interference identification network; wherein the to-be-identified radar interference echo data comprises a to-be-identified interference data time-frequency spectrum and a to-be-identified SAR interference image; the interference identification network comprises a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected in parallel to the feature fusion network, and the feature fusion network is connected to the classification network; and obtaining an interference type identification result corresponding to the radar interference echo data to be identified. The method can effectively extract and fuse features from different modes to realize interference identification, and has high identification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of radar interference recognition, and in particular to a time-frequency-image fusion distance-azimuth SAR active interference recognition method and device. Background Art

[0002] In modern warfare, SAR (Synthetic Aperture Radar) is an important reconnaissance tool and an important research object in the field of electronic jamming. In order to disrupt SAR's reconnaissance of its own side, a series of jamming methods for SAR have been publicly reported in the literature, mainly including active jamming and passive jamming. Among them, active jamming signals have brought great difficulties to the normal and effective operation of SAR. In order to effectively deal with the technical challenges of SAR jamming, many scholars have proposed a series of SAR jamming suppression and anti-jamming technical methods. From the perspective of SAR reconnaissance and intelligence generation, the identification of SAR jamming types is the premise and basis for the use of SAR anti-jamming technology, which has extremely important research significance. The radar jamming identification methods proposed in the prior art mainly focus on the identification of SAR active jamming types, while there are few studies on the identification of active jamming types that take into account both SAR range and azimuth. Summary of the invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides a time-frequency-image fusion range-joint azimuth SAR active interference identification method and device.

[0004] According to a first aspect of an embodiment of the present invention, a time-frequency-image fusion range-joint azimuth SAR active interference identification method is provided, the method comprising:

[0005] Inputting the radar interference echo data to be identified into the trained interference identification network; wherein the radar interference echo data to be identified includes the time-frequency spectrum of the interference data to be identified and the SAR interference image to be identified; the interference identification network includes a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected to the feature fusion network in parallel, and the feature fusion network is connected to the classification network; the first feature extraction network includes a convolution module, and the second feature extraction network includes a plurality of alternately connected CBR layers and WavePool layers; the feature fusion network includes a cross attention module and a global average pooling layer connected in sequence;

[0006] Obtain an interference type identification result corresponding to the radar interference echo data to be identified.

[0007] Optionally, obtaining the interference type identification result corresponding to the radar interference echo data to be identified includes:

[0008] Processing the time-frequency spectrum of the interference data to be identified according to the first feature extraction network to obtain a first feature;

[0009] Processing the to-be-identified SAR interference image according to the second feature extraction network to obtain a second feature;

[0010] The first feature and the second feature are fused according to the feature fusion network and input into the classification network to obtain the interference type identification result corresponding to the radar interference echo data to be identified.

[0011] Optionally, the convolution module includes a BCL layer, a first CBL layer, a second CBL layer, a CB layer and a first CBR layer connected in sequence.

[0012] Optionally, the BCL layer includes a BN layer, a Conv layer and a ReLU6 activation function connected in sequence; the first CBL layer and the second CBL layer have the same structure, and both include a Conv layer, a BN layer and a ReLU6 activation function connected in sequence; the CB layer includes a Conv layer and a BN layer connected in sequence; the first CBR layer includes a Conv layer, a BN layer and a ReLU activation function connected in sequence.

[0013] Optionally, the second feature extraction network includes a second CBR layer, a first WavePool layer, a third CBR layer, a second WavePool layer and a fourth CBR layer connected in sequence.

[0014] Optionally, the second CBR layer, the third CBR layer and the fourth CBR layer have the same structure, and all include a Conv layer, a BN layer and a ReLU activation function connected in sequence.

[0015] Optionally, the training process of the interference identification network includes:

[0016] Constructing a training data set; wherein the training data set includes an interference data time-frequency spectrum data set and a SAR interference image data set;

[0017] Using the training data set to train the interference identification network, and adjusting the parameters of the interference identification network according to a pre-constructed loss function;

[0018] Continue to train the interference recognition network after parameter adjustment until the maximum number of iterations is reached, and obtain a trained interference loss network.

[0019] Optionally, the pre-constructed loss function is expressed as follows:

[0020]

[0021] Among them, y i represents the true label of the i-th sample in the training data set, P ic represents the probability that the interference recognition network predicts that the category of the i-th sample is c, N is the total number of samples, and K is the number of categories of the samples.

[0022] According to a second aspect of an embodiment of the present invention, a time-frequency-image fusion range-joint azimuth SAR active interference identification device is provided, the device comprising:

[0023] An identification module is used to input radar interference echo data to be identified into a trained interference identification network; wherein the radar interference echo data to be identified includes a time-frequency spectrum of interference data to be identified and a SAR interference image to be identified; the interference identification network includes a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected in parallel to the feature fusion network, and the feature fusion network is connected to the classification network; the first feature extraction network includes a convolution module, and the second feature extraction network includes a plurality of alternately connected CBR layers and WavePool layers; the feature fusion network includes a cross attention module and a global average pooling layer connected in sequence;

[0024] The result acquisition module is used to obtain the interference type identification result corresponding to the radar interference echo data to be identified.

[0025] The technical solution provided by the present invention may include the following beneficial effects:

[0026] Through the above technical scheme, the interference identification network constructed by the present invention can realize feature extraction and fusion of different modal data of radar interference, avoid the shortcomings of manually extracting radar interference data features, and realize effective extraction of radar interference data features; and the present invention extracts the features of two modalities of radar interference data, namely, time-frequency spectrum and SAR image, and uses a cross-attention module to perform complementary fusion of inter-modal features, giving different attention to feature maps of different modalities and positions, providing more information for subsequent identification, solving the problem that the existing method can only identify single-range SAR interference and has low generalization, so that the present invention has the advantages of simultaneously recognizing range and azimuth interference and high recognition accuracy, and adding multiple convolutional layers in the feature extraction network to enhance the characterization features of the neural network; in summary, the present invention can effectively extract and fuse features from different modalities and realize interference identification.

[0027] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention but do not constitute a limitation of the present invention. In the accompanying drawings:

[0029] Figure 1 The present invention is a flow chart of a method for identifying active interference of SAR in range and azimuth combined with time-frequency-image fusion according to an exemplary embodiment.

[0030] Figure 2 The figure is a schematic diagram showing the structure of an interference identification network according to an exemplary embodiment.

[0031] Figure 3 It is a block diagram of a time-frequency-image fusion range-azimuth SAR active interference identification device according to an exemplary embodiment. DETAILED DESCRIPTION

[0032] Figure 1 FIG. 1 is a flow chart of a method for identifying active interference of SAR in a range-azimuth manner by time-frequency-image fusion according to an exemplary embodiment. Figure 1 As shown, the method comprises the following steps:

[0033] S101, inputting the radar interference echo data to be identified into the trained interference identification network; wherein, the radar interference echo data to be identified includes the time-frequency spectrum of the interference data to be identified and the SAR interference image to be identified; the interference identification network includes a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected in parallel to the feature fusion network, and the feature fusion network is connected to the classification network; the first feature extraction network includes a convolution module, and the second feature extraction network includes a plurality of alternately connected CBR layers and WavePool layers; the feature fusion network includes a cross-attention module and a global average pooling layer connected in sequence.

[0034] Optionally, Figure 2 is a schematic diagram of a structure of an interference identification network according to an exemplary embodiment. Figure 2 As shown, the convolution module includes a BCL layer, a first CBL layer, a second CBL layer, a CB layer and a first CBR layer connected in sequence.

[0035] It can be understood that the BCL layer includes a BN layer, a Conv layer and a ReLU6 activation function connected in sequence; the first CBL layer and the second CBL layer have the same structure, both including a Conv layer, a BN layer and a ReLU6 activation function connected in sequence; the CB layer includes a Conv layer and a BN layer connected in sequence; the first CBR layer includes a Conv layer, a BN layer and a ReLU activation function connected in sequence.

[0036] It is worth mentioning that the BCL layer, the first CBL layer, and the second CBL layer use the ReLU6 activation function instead of the activation function used in conventional convolutional neural network feature extraction; limiting the output range makes the dynamic range of the network parameters smaller, which helps to lightweight the model; the first CBR layer uses 1×1 convolution to recombine the feature map output by the deep convolution to achieve feature enhancement effect and realize SAR active interference distance feature extraction.

[0037] Optionally, refer to Figure 2 , the second feature extraction network includes a second CBR layer, a first WavePool layer, a third CBR layer, a second WavePool layer and a fourth CBR layer connected in sequence.

[0038] It can be understood that the second CBR layer, the third CBR layer and the fourth CBR layer have the same structure, and all include a Conv layer, a BN layer and a ReLU activation function connected in sequence. The WavePool layer is used to replace the pooling layer in the conventional convolutional neural network. The wavelet transform principle is used to capture feature information at different frequencies through multi-scale analysis, which can be better used in complex scenes such as SAR images to improve the effectiveness of feature extraction; the fourth CBR layer also uses 1×1 convolution to recombine the feature map output by the deep convolution to achieve feature enhancement effect, so as to realize the range and azimuth feature extraction of SAR active interference.

[0039] It is worth mentioning that the feature fusion network includes a cross-attention module and a global average pooling layer. The cross-attention module significantly improves the model's representation ability and enhances the complementarity of information by dynamically allocating weights of different modal features. At the same time, it adaptively adjusts the importance of features, effectively reduces background noise interference, and reduces the loss of key information in the fusion process, improving the model's generalization ability when dealing with complex scenes; the classification network includes the first connection layer and the second fully connected layer.

[0040] In one embodiment, referring to Figure 2, using the basic structure of convolutional neural network, build interference recognition network, generate key vector K and value vector V according to the enhanced time-frequency graph feature vector f1 output after the first feature extraction network, generate query vector Q according to the feature vector f2 of SAR image output after the second feature extraction network, and construct the feature vector of cross attention module as A=MHA(Q,K,V). A represents the cross-modal attention between time-frequency branch and SAR image branch. After the calculation is completed, the global average pooling layer is used for downsampling, and then the first fully connected layer FC is added to receive the feature map from the previous layer and convert it into a one-dimensional feature vector, and then the converted one-dimensional feature vector is output to the second fully connected layer, which is directly connected to the output layer, and the number of neurons in the second fully connected layer corresponds to the number of categories of the classification task.

[0041] In one implementation, the parameters of each network layer in the interference identification network can refer to Table 1: Detailed parameters of each network layer are shown in Table 1:

[0042] Table 1

[0043]

[0044]

[0045] Optionally, the training process of the interference identification network includes:

[0046] Constructing a training data set; wherein the training data set includes an interference data time-frequency spectrum data set and a SAR interference image data set;

[0047] The interference recognition network is trained using the training data set, and the parameters of the interference recognition network are adjusted according to the pre-constructed loss function;

[0048] Continue to train the interference recognition network with adjusted parameters until the maximum number of iterations is reached to obtain a trained interference loss network.

[0049] Optionally, a pre-built loss function is expressed as follows:

[0050]

[0051] Among them, y i represents the true label of the i-th sample in the training dataset, P ic It represents the probability that the interference recognition network predicts that the category of the i-th sample is c, N is the total number of samples, and K is the number of sample categories.

[0052] It can be understood that the interference data time-frequency spectrum data set in the training data set can be constructed in the following way: set the short-time Fourier transform parameters: the window function is a Hamming window, the window length is 65, the step size is 1, and the number of Fourier transform points is 128; perform short-time Fourier transform on the simulated interference data to obtain the interference data time-frequency spectrum data set;

[0053] The SAR interference image dataset in the training dataset can be constructed in the following way: setting the SAR system parameters: carrier frequency 18 GHz, bandwidth 300 MHz, pulse width 1 us, pulse repetition frequency 2000 Hz; performing imaging processing on the simulated interference echo data to obtain the SAR interference image dataset.

[0054] After constructing the training data set, preprocessing is required. The maximum normalization method can be used to normalize the interference data time-frequency spectrum data set and the SAR interference image data set.

[0055] S102: Obtain interference type identification results corresponding to the radar interference echo data to be identified.

[0056] Optionally, obtaining an interference type identification result corresponding to the radar interference echo data to be identified includes:

[0057] Processing the frequency spectrum of the interference data to be identified according to the first feature extraction network to obtain a first feature;

[0058] The SAR interference image to be identified is processed according to the second feature extraction network to obtain a second feature;

[0059] The first feature and the second feature are fused according to the feature fusion network and input into the classification network to obtain the interference type identification result corresponding to the radar interference echo data to be identified.

[0060] It can be understood that after the first feature and the second feature are fused according to the feature fusion network, the interference type recognition result can be obtained by using the classification network and processed by the Softmax layer. Specifically, the radar interference echo data to be identified is input into the trained interference recognition network, and the network is forward propagated. After passing through two fully connected layers and the Softmax layer, the final interference type recognition result is obtained. The recognition result is shown in Table 2. The Softmax layer operation method is as follows:

[0061]

[0062] Among them, z j represents the output value of the jth category, j (j = 1, 2, ..., K) represents the number of categories, A (z j) represents the probability of the jth type of interference, and its value range is (0,1). The category with the largest probability is taken as the predicted interference category result.

[0063] Table 2

[0064]

[0065] Through the above technical scheme, the interference identification network constructed by the present invention can realize feature extraction and fusion of different modal data of radar interference. When extracting time-frequency features, the ReLU6 activation function is used instead of the activation function used in conventional convolutional neural network feature extraction; the output range is limited to reduce the dynamic range of network parameters, which is conducive to lightweight model; and finally, 1×1 convolution is added to recombine the feature map output by deep convolution to achieve feature enhancement effect, so as to realize SAR active interference range feature extraction; when extracting SAR image features, the WavePool layer is used instead of the pooling layer in the conventional convolutional neural network, and the wavelet transform principle is used through multi-scale analysis to capture feature information at different frequencies, which is suitable for complex scenes such as SAR images, improves the effectiveness of feature extraction, and realizes SAR active interference range and azimuth feature extraction. Finally, the cross-attention module is used to perform complementary fusion of inter-modal features. This module significantly improves the representation ability of the model and enhances the complementarity of information by dynamically allocating weights of different modal features. At the same time, it adaptively adjusts the importance of features, effectively reduces background noise interference, and reduces the loss of key information in the fusion process, improving the generalization ability of the model when processing complex scenes. The whole process avoids the shortcomings of manually extracting radar interference data features and realizes the effective extraction of radar interference data features. In addition, the present invention extracts the features of two modalities, namely, the time spectrum and SAR image of radar interference data, and uses the cross-attention module to perform complementary fusion of inter-modal features, giving different attention to feature maps of different modalities and positions, providing more information for subsequent recognition, solving the problem that the existing method can only identify single-range SAR interference and has low generalization, so that the present invention has the advantages of simultaneously recognizing range and azimuth interference and high recognition accuracy. In summary, the present invention can effectively extract and fuse features from different modalities and realize interference recognition.

[0066] Figure 3 FIG. 1 is a block diagram of a time-frequency-image fusion distance-azimuth SAR active interference identification device according to an exemplary embodiment. Figure 3 As shown, the device 300 includes:

[0067] The identification module 301 is used to input the radar interference echo data to be identified into the trained interference identification network; wherein the radar interference echo data to be identified includes the time-frequency spectrum of the interference data to be identified and the SAR interference image to be identified; the interference identification network includes a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected in parallel to the feature fusion network, and the feature fusion network is connected to the classification network; the first feature extraction network includes a convolution module, and the second feature extraction network includes a plurality of alternately connected CBR layers and WavePool layers; the feature fusion network includes a cross attention module and a global average pooling layer connected in sequence;

[0068] The result acquisition module 302 is used to obtain the interference type identification result corresponding to the radar interference echo data to be identified.

[0069] Through the above technical scheme, the interference identification network constructed by the present invention can realize feature extraction and fusion of different modal data of radar interference, avoid the shortcomings of manually extracting radar interference data features, and realize effective extraction of radar interference data features; and the present invention extracts the features of two modalities of radar interference data, namely, time-frequency spectrum and SAR image, and uses a cross-attention module to perform complementary fusion of inter-modal features, giving different attention to feature maps of different modalities and positions, providing more information for subsequent identification, solving the problem that the existing method can only identify single-range SAR interference and has low generalization, so that the present invention has the advantages of simultaneously recognizing range and azimuth interference and high recognition accuracy, and adding multiple convolutional layers in the feature extraction network to enhance the characterization features of the neural network; in summary, the present invention can effectively extract and fuse features from different modalities and realize interference identification.

[0070] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0071] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0072] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0073] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A time-frequency-image fusion distance-azimuth SAR active interference identification method, characterized in that: The method comprises: Inputting the radar interference echo data to be identified into the trained interference identification network; wherein the radar interference echo data to be identified includes the time-frequency spectrum of the interference data to be identified and the SAR interference image to be identified; the interference identification network includes a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected to the feature fusion network in parallel, and the feature fusion network is connected to the classification network; the first feature extraction network includes a convolution module, and the second feature extraction network includes a plurality of alternately connected CBR layers and WavePool layers; the feature fusion network includes a cross attention module and a global average pooling layer connected in sequence; Obtain an interference type identification result corresponding to the radar interference echo data to be identified.

2. The time-frequency-image fusion distance-joint azimuth SAR active interference identification method according to claim 1 is characterized in that: The obtaining of the interference type identification result corresponding to the radar interference echo data to be identified includes: Processing the time-frequency spectrum of the interference data to be identified according to the first feature extraction network to obtain a first feature; Processing the to-be-identified SAR interference image according to the second feature extraction network to obtain a second feature; The first feature and the second feature are fused according to the feature fusion network and input into the classification network to obtain the interference type identification result corresponding to the radar interference echo data to be identified.

3. The time-frequency-image fusion distance-joint azimuth SAR active interference identification method according to claim 1 is characterized in that: The convolution module includes a BCL layer, a first CBL layer, a second CBL layer, a CB layer and a first CBR layer which are connected in sequence.

4. The time-frequency-image fusion distance-joint azimuth SAR active interference identification method according to claim 3 is characterized in that: The BCL layer includes a BN layer, a Conv layer and a ReLU6 activation function connected in sequence; the first CBL layer and the second CBL layer have the same structure, both including a Conv layer, a BN layer and a ReLU6 activation function connected in sequence; the CB layer includes a Conv layer and a BN layer connected in sequence; the first CBR layer includes a Conv layer, a BN layer and a ReLU activation function connected in sequence.

5. The time-frequency-image fusion distance-joint azimuth SAR active interference identification method according to claim 1 is characterized in that: The second feature extraction network includes a second CBR layer, a first WavePool layer, a third CBR layer, a second WavePool layer and a fourth CBR layer which are connected in sequence.

6. The time-frequency-image fusion distance-joint azimuth SAR active interference identification method according to claim 5 is characterized in that: The second CBR layer, the third CBR layer and the fourth CBR layer have the same structure, and all include a Conv layer, a BN layer and a ReLU activation function connected in sequence.

7. The time-frequency-image fusion distance-azimuth SAR active interference identification method according to claim 1 is characterized in that: The training process of the interference identification network includes: Constructing a training data set; wherein the training data set includes an interference data time-frequency spectrum data set and a SAR interference image data set; Using the training data set to train the interference identification network, and adjusting the parameters of the interference identification network according to a pre-constructed loss function; Continue to train the interference recognition network after parameter adjustment until the maximum number of iterations is reached, and obtain a trained interference loss network.

8. The time-frequency-image fusion distance-joint azimuth SAR active interference identification method according to claim 7 is characterized in that: The pre-built loss function is expressed as follows: Among them, y i represents the true label of the i-th sample in the training data set, P ic represents the probability that the interference recognition network predicts that the category of the i-th sample is c, N is the total number of samples, and K is the number of categories of the samples.

9. A time-frequency-image fusion distance-azimuth SAR active interference identification device, characterized in that: The device comprises: An identification module is used to input radar interference echo data to be identified into a trained interference identification network; wherein the radar interference echo data to be identified includes a time-frequency spectrum of interference data to be identified and a SAR interference image to be identified; the interference identification network includes a first feature extraction network, a second feature extraction network, a feature fusion network and a classification network; the first feature extraction network and the second feature extraction network are connected in parallel to the feature fusion network, and the feature fusion network is connected to the classification network; the first feature extraction network includes a convolution module, and the second feature extraction network includes a plurality of alternately connected CBR layers and WavePool layers; the feature fusion network includes a cross attention module and a global average pooling layer connected in sequence; The result acquisition module is used to obtain the interference type identification result corresponding to the radar interference echo data to be identified.

Images