Small sample electromagnetic signal identification method, device, medium and system based on improved twin network

Abstract

The invention discloses a small sample electromagnetic signal identification method, device, medium and system based on an improved twin network, and belongs to the field of intelligent signal processing, and the method comprises the steps: carrying out the fast Fourier transform of a received broadband electromagnetic signal, and estimating the center frequency of the electromagnetic signal; performing digital down-conversion by taking the center frequency as a carrier frequency so as to obtain an IQ signal; intercepting a leading-edge fixed-length signal, and carrying out space mapping by adopting a convolutional neural network to obtain a feature vector; and judging whether the two electromagnetic signals belong to the same radiation source or not by taking the cosine similarity between the feature vectors as a cost function. The method does not depend on electromagnetic parameters used by a radiation source, so that a more robust small sample identification result can be obtained.

Description

Technical Field

[0001] The present invention relates to the field of intelligent signal processing, and more specifically, to a small sample electromagnetic signal recognition method, device, medium and system based on an improved twin network. Background Art

[0002] Traditional electromagnetic signal recognition is mainly based on manually extracted features, but due to the low parameter dimension, it leads to serious parameter overlap and poor adaptability, especially under the condition of small sample size, the recognition performance is further reduced. Artificial intelligence recognition methods represented by deep learning require a large number of data samples, and the recognition effect is very limited under small sample conditions. For small sample recognition processing, the traditional twin network uses Euclidean distance as the cost function, and there is a problem of threshold setting for the similarity measurement of vectors. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a small sample electromagnetic signal recognition method, device, medium and system based on an improved twin network, which is independent of the electromagnetic parameters used by the radiation source, thereby obtaining a more robust small sample recognition result.

[0004] The object of the present invention is achieved through the following solutions:

[0005] A small sample electromagnetic signal recognition method based on an improved twin network comprises the following steps:

[0006] Firstly, a fast Fourier transform is performed on the received broadband electromagnetic signal and the center frequency of the electromagnetic signal is estimated;

[0007] Secondly, digital down-conversion is performed using the center frequency as the carrier frequency to obtain the IQ signal;

[0008] Thirdly, the fixed-length signal at the front is intercepted, and the convolutional neural network is used for spatial mapping to obtain the feature vector;

[0009] Finally, the cosine similarity between the feature vectors is used as the cost function to determine whether the two electromagnetic signals belong to the same radiation source.

[0010] Furthermore, the method of performing a fast Fourier transform on the received broadband electromagnetic signal and estimating the center frequency of the electromagnetic signal specifically comprises the following sub-steps:

[0011] S1, assuming that the received broadband electromagnetic signal is s(n)(, where N is the length of the sampling point and the sampling rate is Fs, the received broadband electromagnetic signal s(n)( is processed by fast Fourier transform to obtain Spec(n * ) and Spec(n * ) takes the amplitude to get |Spec(n *)|; take|Spec(n * )|The serial number corresponding to the first maximum value is assumed to be Idx, then the estimated center frequency fc of the signal is

[0012] Furthermore, the digital down-conversion using the center frequency as the carrier frequency to obtain the IQ signal specifically includes the following sub-steps:

[0013] S2, with the estimated intermediate frequency fc, is digitally down-converted to obtain s1(n), and the processing formula is:

[0014]

[0015] Among them, j is an imaginary unit, and after processing, s1(n) is a complex signal.

[0016] S3, filters s1(n) and performs M-fold decimation to obtain in Indicates rounding down.

[0017] Furthermore, the fixed-length signal of the intercepted front is spatially mapped using a convolutional neural network to obtain a feature vector, which specifically includes the following sub-steps:

[0018] S4, the in-phase component and orthogonal component of the complex signal s2(m) are combined into a two-dimensional vector sigVec, the size of which is And perform a cut of length K1, the cut two-dimensional vector is sigVec * , whose vector size is 2*K1.

[0019] S5, for sigVec * Use convolutional neural network for spatial mapping, and get the feature vector FeatureVec, let the feature be expressed as [x1,x2,…,x K ], where K is the length of the feature vector;

[0020] S6, the known small sample electromagnetic signal is processed according to steps S1 to S5, and the feature obtained by processing the pth (p = 1, 2, ..., P) electromagnetic signal is FeatureVec p =[x 1,p ,x 2,p ,…,x K,p ](p=1,2,…,P), where P is the number of known small sample electromagnetic signals. The electromagnetic signal to be identified is also processed according to steps 1 to 5 to obtain the feature FeatureVec ID =[x 1,ID ,x 2,ID ,…,x K,ID].

[0021] Furthermore, the method of using the cosine similarity between the feature vectors as a cost function to determine whether the two electromagnetic signals belong to the same radiation source specifically includes the following sub-steps:

[0022] S7, the feature FeatureVec of the electromagnetic signal to be identified ID The cost function is estimated based on the cosine similarity with the known small sample electromagnetic signal features, and the calculation formula is:

[0023]

[0024] in[·] T Represents a transpose operation;

[0025] S8, with cost function cost p The category corresponding to the electromagnetic signal sample when it is the minimum is used as the category of the electromagnetic signal to be identified.

[0026] A small sample electromagnetic signal recognition device based on an improved twin network includes: a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is loaded by the processor, any of the methods described above is executed.

[0027] A computer-readable storage medium stores a computer program, wherein the computer program executes any of the above methods when loaded by a processor.

[0028] A small sample electromagnetic signal recognition system based on an improved twin network includes the small sample electromagnetic signal recognition device based on the improved twin network as described above.

[0029] The beneficial effects of the present invention include:

[0030] The present invention provides a small sample electromagnetic signal recognition method based on an improved twin network. First, the received broadband electromagnetic signal is subjected to a fast Fourier transform and the center frequency of the electromagnetic signal is estimated; secondly, the center frequency is used as the carrier frequency for digital down-conversion to obtain an IQ signal; thirdly, the fixed-length signal at the leading edge is intercepted, and a convolutional neural network is used for spatial mapping to obtain a feature vector; finally, the cosine similarity between the feature vectors is used as a cost function to determine whether the two electromagnetic signals belong to the same radiation source.

[0031] The method of the present invention does not rely on the electromagnetic parameters used by the radiation source, so that a more robust small sample recognition result can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0033] Figure 1 The figure is a flowchart of the steps of the method according to the embodiment of the present invention. DETAILED DESCRIPTION

[0034] All features disclosed in all embodiments in this specification, or steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or expanded or replaced in any manner.

[0035] In view of the current situation in the background, the present invention provides a small sample electromagnetic signal recognition method based on an improved twin network. First, the received broadband electromagnetic signal is subjected to a fast Fourier transform and the center frequency of the electromagnetic signal is estimated; secondly, the center frequency is used as the carrier frequency for digital down-conversion to obtain an IQ signal. Again, the fixed-length signal at the leading edge is intercepted, and a convolutional neural network is used for spatial mapping to obtain a feature vector. Finally, the cosine similarity between the feature vectors is used as the cost function to determine whether the two electromagnetic signals belong to the same radiation source. The method of the present invention does not depend on the electromagnetic parameters used by the radiation source, so that a more robust small sample recognition result can be obtained. In a further implementation manner, if Figure 1 As shown, the specific steps are as follows:

[0036] Assume that the received broadband electromagnetic signal is s(n) (n=1,…,N), where N is the length of the sampling points and the sampling rate is Fs.

[0037] Step 1: Perform fast Fourier transform on the received broadband electromagnetic signal s(n) (n=1,…,N) to obtain Spec(n * )(n * =1,…,N), and for Spec(n * ) takes the amplitude to get |Spec(n * )|. Take |Spec(n * )|The serial number corresponding to the first maximum value is assumed to be Idx, then the estimated center frequency fc of the signal is

[0038]

[0039] Step 2: Perform digital down-conversion processing with the estimated intermediate frequency fc to obtain s1(n). The processing formula is:

[0040]

[0041] Among them, j is an imaginary unit, and after processing, s1(n) is a complex signal.

[0042] Step 3: Filter s1(n) and perform M-fold decimation to obtain in Indicates rounding down.

[0043] Step 4: Combine the in-phase and orthogonal components of the complex signal s2(m) into a two-dimensional vector sigVec, the size of which is And perform a cut of length K1, the cut two-dimensional vector is sigVec * , whose vector size is 2*K1.

[0044] Step 5: sigVec * Use convolutional neural network for spatial mapping, and get the feature vector FeatureVec, let the feature be expressed as [x1,x2,…,x K ], where K is the length of the feature vector.

[0045] Step 6: Process the known small sample electromagnetic signal according to steps 1 to 5. The feature obtained by processing the pth (p = 1, 2, ..., P) electromagnetic signal is FeatureVec p =[x 1,p ,x 2,p ,…,x K,p ](p=1,2,…,P), where P is the number of known small sample electromagnetic signals. The electromagnetic signal to be identified is also processed according to steps 1 to 5 to obtain the feature FeatureVec ID =[x 1,ID ,x 2,ID ,…,x K,ID ].

[0046] Step 7: FeatureVec of the electromagnetic signal to be identified ID The cost function is estimated based on the cosine similarity with the known small sample electromagnetic signal features. The calculation formula is:

[0047]

[0048] in[·] T Represents a transpose operation.

[0049] Step 8: Use cost function cost p The category corresponding to the electromagnetic signal sample when it is the minimum is used as the category of the electromagnetic signal to be identified.

[0050] In other embodiments of the present invention, the following steps are performed: Assume that the received broadband electromagnetic signal is s(n) (n=1, ..., N), where N is the length of the sampling point and the sampling rate is Fs=600 MHz.

[0051] Step 1: Perform fast Fourier transform on the received broadband electromagnetic signal s(n) (n=1,…,N) to obtain Spec(n * )(n * =1,…,N), and for Spec(n * ) takes the amplitude to get |Spec(n * )|. Take |Spec(n * )|The serial number corresponding to the first maximum value is assumed to be Idx, then the estimated center frequency fc of the signal is

[0052]

[0053] Step 2: Perform digital down-conversion processing with the estimated intermediate frequency fc to obtain s1(n). The processing formula is:

[0054]

[0055] Where j is an imaginary unit, and after processing, s1(n) is a complex signal.

[0056] Step 3: Filter s1(n) and perform M=4 decimation to obtain in Indicates rounding down.

[0057] Step 4: Combine the in-phase and orthogonal components of the complex signal s2(m) into a two-dimensional vector sigVec, the size of which is And perform a cut with a length of K1=256, and the cut two-dimensional vector is sigVec * , whose vector size is 2*K1.

[0058] Step 5: sigVec * Use convolutional neural network for spatial mapping, and get the feature vector FeatureVec, let the feature be expressed as [x1,x2,…,x K ], where K = 32 is the length of the feature vector.

[0059] Step 6: Process the known small sample electromagnetic signal according to steps 1 to 5. The feature obtained by processing the pth (p = 1, 2, ..., P) electromagnetic signal is FeatureVec p =[x 1,p ,x 2,p ,…,x K,p](p=1,2,…,P), where P is the number of known small sample electromagnetic signals. The electromagnetic signal to be identified is also processed according to steps 1 to 5 to obtain the feature FeatureVec ID =[x 1,ID ,x 2,ID ,…,x K,ID ].

[0060] Step 7: FeatureVec of the electromagnetic signal to be identified ID The cost function is estimated based on the cosine similarity with the known small sample electromagnetic signal features. The calculation formula is:

[0061]

[0062] in[·] T Represents a transpose operation.

[0063] Step 8: Use cost function cost p The category corresponding to the electromagnetic signal sample when it is the minimum is used as the category of the electromagnetic signal to be identified.

[0064] The units involved in the embodiments of the present invention may be implemented by software or hardware, and the units described may also be arranged in a processor. The names of these units do not, in some cases, limit the units themselves.

[0065] According to one aspect of an embodiment of the present invention, a computer program product or a computer program is provided, the computer program product or the computer program includes a computer instruction, and the computer instruction is stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in the above various optional implementations.

[0066] As another aspect, an embodiment of the present invention further provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment.

Claims

1. A small sample electromagnetic signal recognition method based on an improved twin network, characterized in that: The following steps are involved: Firstly, a fast Fourier transform is performed on the received broadband electromagnetic signal and the center frequency of the electromagnetic signal is estimated; Secondly, digital down-conversion is performed using the center frequency as the carrier frequency to obtain the IQ signal; Thirdly, the fixed-length signal at the front is intercepted, and the convolutional neural network is used for spatial mapping to obtain the feature vector; Finally, the cosine similarity between the feature vectors is used as the cost function to determine whether the two electromagnetic signals belong to the same radiation source.

2. The small sample electromagnetic signal recognition method based on the improved twin network according to claim 1 is characterized in that: The method of performing fast Fourier transform on the received broadband electromagnetic signal and estimating the center frequency of the electromagnetic signal specifically comprises the following sub-steps: S1, assuming that the received broadband electromagnetic signal is s(n)(, where N is the length of the sampling point and the sampling rate is Fs, the received broadband electromagnetic signal s(n)( is processed by fast Fourier transform to obtain Spec(n * ) and Spec(n * ) takes the amplitude to get |Spec(n * )|; take|Spec(n * )|The serial number corresponding to the first maximum value is assumed to be Idx, then the estimated center frequency fc of the signal is 3. The small sample electromagnetic signal recognition method based on the improved twin network according to claim 2 is characterized in that: The method of performing digital down-conversion using the center frequency as the carrier frequency to obtain the IQ signal specifically includes the following sub-steps: S2, with the estimated intermediate frequency fc, is digitally down-converted to obtain s1(n), and the processing formula is: Among them, j is an imaginary unit, and after processing, s1(n) is a complex signal. S3, filters s1(n) and performs M-fold decimation to obtain in Indicates rounding down.

4. The small sample electromagnetic signal recognition method based on the improved twin network according to claim 3 is characterized in that: The fixed-length signal of the intercepted front is spatially mapped using a convolutional neural network to obtain a feature vector, which specifically includes the following sub-steps: S4, the in-phase component and orthogonal component of the complex signal s2(m) are combined into a two-dimensional vector sigVec, the size of which is And perform a cut of length K1, the cut two-dimensional vector is sigVec * , whose vector size is 2*K1. S5, for sigVec * Convolutional neural network is used for spatial mapping, and the feature vector is obtained as FeatureVec, which is represented as [x1, x2,…, x K ], where K is the length of the feature vector; S6, the known small sample electromagnetic signal is processed according to steps S1 to S5, and the feature obtained by processing the pth (p = 1, 2, ..., P) electromagnetic signal is FeatureVec p =[x 1,p ,x 2,p ,…,x K,p ](p=1,2,…,P), where P is the number of known small sample electromagnetic signals. The electromagnetic signal to be identified is also processed according to steps 1 to 5 to obtain the feature FeatureVec ID =[x 1,ID ,x 2,ID ,…,x K,ID ].

5. The small sample electromagnetic signal recognition method based on the improved twin network according to claim 4 is characterized in that: The method of using the cosine similarity between the feature vectors as a cost function to determine whether two electromagnetic signals belong to the same radiation source specifically includes the following sub-steps: S7, the feature FeatureVec of the electromagnetic signal to be identified ID The cost function is estimated based on the cosine similarity with the known small sample electromagnetic signal features, and the calculation formula is: in[·] T Represents a transpose operation; S8, with cost function cost p The category corresponding to the electromagnetic signal sample when it is the minimum is used as the category of the electromagnetic signal to be identified.

6. A small sample electromagnetic signal recognition device based on an improved twin network, characterized in that: include: A processor and a memory, wherein a computer program is stored in the memory, and when the computer program is loaded by the processor, the method according to any one of claims 1 to 5 is executed.

7. A computer-readable storage medium, characterized in that: A computer program is stored in a readable storage medium, and when the computer program is loaded by a processor, the method according to any one of claims 1 to 5 is executed.

8. A small sample electromagnetic signal recognition system based on an improved twin network, characterized in that: It includes the small sample electromagnetic signal recognition device based on the improved twin network as described in claim 6.

Images