Interference signal identification method and device based on multi-dimensional feature fusion

Abstract

The invention discloses an interference signal identification method and device based on multi-dimensional feature fusion, and belongs to the technical field of wireless communication, and the method comprises the following steps: S1, carrying out the modeling of an interference signal, and obtaining the time sequence feature of the interference signal; s2, extracting a time domain characteristic parameter and a frequency domain characteristic parameter of the interference signal; s3, performing time-frequency characteristic conversion on the time sequence characteristics of the interference signals to obtain a two-dimensional time-frequency energy diagram; s4, constructing a first neural network model, inputting the two-dimensional time-frequency energy diagram obtained in the step S3 into the first neural network model, and extracting feature parameters of interference signals by using a full-connection layer of the network; and S5, fusing the feature parameters of the three dimensions in the steps S2 and S4, inputting the fused feature parameters into the neural network model 2, and identifying an interference signal. According to the invention, the interference signal identification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication, and more specifically, to a method and device for identifying interference signals based on multi-dimensional feature fusion. Background Art

[0002] With the rapid popularization and development of wireless communication technology, wireless communication has become an indispensable part of modern society. However, due to the shared and open characteristics of the wireless signal propagation medium, wireless communication systems are extremely vulnerable to malicious interference attacks. Interference signal identification has become the premise and foundation for the anti-interference of communication systems and plays a crucial role.

[0003] The interference received by a wireless communication system at the receiver is mainly suppression interference, which makes the useful information received by the other communication device blurred or completely masked by transmitting high-power noise or noise-like signals. Common interference types include sweep interference, multi-tone interference, narrowband noise interference, etc. Suppression interference technology has become the most widely used interference form because of its low implementation difficulty, simple technology, and low cost. Those skilled in the art urgently need to solve the technical problem of the accuracy of interference signal identification. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method and device for identifying interference signals based on multi-dimensional feature fusion, which improves the accuracy of interference signal identification.

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

[0006] A method for identifying interference signals based on multi-dimensional feature fusion includes the following steps:

[0007] S1, model the interference signal to obtain the timing characteristics of the interference signal;

[0008] S2, extract the time-domain feature parameters and frequency-domain feature parameters of the interference signal;

[0009] S3, perform time-frequency feature conversion on the timing characteristics of the interference signal to obtain a two-dimensional time-frequency energy diagram;

[0010] S4, construct a neural network model one, input the two-dimensional time-frequency energy diagram obtained in step S3 into the neural network model one, and use the fully connected layer of the network to extract the feature parameters of the interference signal;

[0011] S5, fuse the feature parameters in three dimensions in step S2 and step S4, and input the fused feature parameters into the neural network model two to identify the interference signal.

[0012] Further, in step S1, the modeling of the interference signal specifically includes modeling the interference signals in six communication interference fields, namely, single-tone interference signal, multi-tone interference signal, narrowband noise interference signal, wideband noise interference signal, comb-shaped spectrum interference signal, and frequency-sweeping interference signal.

[0013] Further, the single-tone interference signal is emitted at a certain frequency point and is a single-frequency continuous sine waveform. The time-domain expression is as follows:

[0014]

[0015] where P J is the power of the interference signal, f J is the frequency of the interference signal, and θ J is the phase of the interference signal and follows a uniform distribution.

[0016] Further, the multi-tone interference signal is emitted at multiple discrete frequency points. The time-domain expression is as follows:

[0017]

[0018] Further, for the narrowband noise interference, white noise is specifically passed through a narrowband filter to obtain a blocking signal targeting a certain narrowband range. The frequency-domain expression of the narrowband filter is:

[0019]

[0020] Further, the wideband noise interference signal is obtained by passing Gaussian white noise through a wideband filter to obtain a blocking noise code in each wideband range. The frequency-domain expression of the filter is:

[0021]

[0022] where P J is the interference power, and W I is the cut-off frequency of the low-pass filter.

[0023] Further, for the comb-shaped spectrum interference, white noise is specifically passed through multiple narrowband filters to obtain blocking signals targeting multiple narrowband ranges. The frequency-domain expression of the filter is:

[0024]

[0025] where i = 1, 2, …, Q, and Q represents the number of narrowband filters.

[0026] Further, the frequency-sweeping interference exhibits the characteristic of linear scanning with time within a partial spectrum. The time-domain expression is:

[0027]

[0028] Among them, β i is the sweep rate, ω i is the initial angular frequency, is the initial phase, and T is the sweep duration.

[0029] Furthermore, in step S2, before extracting the time-domain characteristic parameters and frequency-domain characteristic parameters of the interference signal, it includes the steps of: preprocessing the signal, including normalization and centering after normalization.

[0030] Furthermore, in step S2, extracting the time-domain characteristic parameters of the interference signal specifically includes the following sub-steps:

[0031] Extracting the time-domain moment skewness coefficient and R parameter, specifically calculated according to the following formula:

[0032] Time-domain moment skewness coefficient a 3 : Among them, μ is the mean of the time-domain signal envelope, σ is its standard deviation, and a 3 is used to describe the degree of deviation of the sample from the normal distribution;

[0033] R parameter: The R parameter is used to reflect the degree of change of the envelope of the sample x.

[0034] Furthermore, in step S2, extracting the frequency-domain characteristic parameters of the interference signal specifically includes the following sub-steps:

[0035] Extracting the carrier factor coefficient, average spectral flatness coefficient, frequency-domain moment skewness coefficient, R f parameter; specifically calculated according to the following formula:

[0036] Carrier factor coefficient C: The carrier factor coefficient describes the prominence of the signal spectral line; among them, X[λ 1 is the maximum value of the frequency-domain signal, X[λ 2 is the second-largest value of the frequency-domain signal;

[0037] Average spectral flatness coefficient Fse: Among them, N is the number of samples of the signal, N s is the number of samples for spectral analysis; Fse reflects whether there are obvious impulse signals locally in the signal; is the mean of the spectral amplitude X(n), and X 1 (n) is the smoothed filtering of the spectrum X(n);

[0038] R f parameter: Among them, μ is the mean of the spectral amplitude X, and σ is its standard deviation;

[0039] Frequency domain moment skewness coefficient b 3 :

[0040] Further, in step S3, the time-frequency feature conversion of the timing characteristics of the interference signal is performed to obtain a two-dimensional time-frequency energy diagram, which specifically includes the following sub-steps:

[0041] The timing characteristics of the interference signal are converted into a two-dimensional time-frequency energy diagram using the short-time Fourier transform. The expression for performing the short-time Fourier transform on the input signal x(t) is as follows:

[0042]

[0043] By adjusting the position τ of the window w(t - τ) on the time axis, the STFT provides a method to observe and analyze the frequency distribution of the signal in different time periods, thereby obtaining the time-frequency representation of the signal.

[0044] Further, in step S4, the two-dimensional time-frequency energy diagram obtained in step S3 is input into neural network model one, and the feature parameters of the interference signal are extracted using the fully connected layer of the network, which specifically includes the following sub-steps:

[0045] Step S41, randomly select a batch of time-frequency diagrams and their corresponding labels from the interference signal database;

[0046] Step S42, input the batch of time-frequency diagrams into neural network model one for classification and recognition, and use the activation vector output by the fully connected layer before the SoftMax layer as the feature vector of the time-frequency diagram. Neural network model one includes a CNN network model;

[0047] Step S43, input the extracted feature vector into the supervised classifier to obtain the classification result, and update the parameters of the supervised classifier according to the cross-entropy loss function between the classification result and the true label;

[0048] Step S44, loop through steps S41 - S43 until the network reaches a convergence state. At this time, the network learns the accurate feature representation of the known interference patterns.

[0049] Further, in step S5, the feature parameters of the three dimensions in steps S2 and S4 are fused, and the fused feature parameters are input into neural network model two to identify the interference signal, which specifically includes the following sub-steps:

[0050] S51, form the training data from the five groups of feature parameters, and the format is as follows:

[0051] tr_data = [a3_list, R_list, C_list, b3_list, mav_list, type_list];

[0052] Among them, the a3_list, R_list, C_list, b3_list, and mav_list in columns 1 - 5 are characteristic parameters, including time - domain characteristics, frequency - domain characteristics, and time - frequency domain characteristics. The type_list in the 6th column is a data label, indicating the nth type of interference signal;

[0053] S52, input the fused characteristic parameters into the second neural network model for signal type recognition; the second neural network model includes an LM neural network.

[0054] An interference signal recognition device based on multi - dimensional feature fusion includes a processor and a memory. A computer program is stored in the memory, and when the computer program is loaded by the processor, it executes the method described in any one of the above.

[0055] The beneficial effects of the present invention include:

[0056] The present invention integrates the amplitude fluctuation characteristics, distribution characteristics in the time domain of interference signals, the spectral distribution and amplitude - frequency distribution in the frequency domain, and the distribution characteristics of the time - frequency diagram. It not only separately considers the characteristics of interference signals in the time domain and frequency domain, but also combines their characteristics in the time - frequency domain, thereby improving the accuracy of the model in identifying interference signals in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0058] Figure 1 It is the time - frequency diagram of six communication interference signals in the embodiments of the present invention;

[0059] Figure 2 It is the step - flow block diagram of the method in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] All the features disclosed in all the embodiments in this specification, or all the steps in the methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or extended, replaced in any way.

[0061] The specific implementation process of the present invention is as follows:

[0062] In a preferred embodiment, as Figure 1 and Figure 2As shown, a method for identifying interference signals based on multi-dimensional feature fusion is specifically provided, including the following steps:

[0063] S1. Model the interference signal to obtain the timing characteristics of the interference signal;

[0064] S2. Extract the time-domain characteristic parameters and frequency-domain characteristic parameters of the interference signal;

[0065] S3. Perform time-frequency characteristic conversion on the timing characteristics of the interference signal to obtain a two-dimensional time-frequency energy diagram;

[0066] S4. Construct a neural network model one, input the two-dimensional time-frequency energy diagram obtained in step S3 into the neural network model one, and use the fully connected layer of the network to extract the characteristic parameters of the interference signal;

[0067] S5. Fuse the characteristic parameters of the three dimensions in step S2 and step S4, input the fused characteristic parameters into the neural network model two, and identify the interference signal.

[0068] In a further other embodiment, the present invention first models six common signals in the field of communication interference, including: single-tone interference, multi-tone interference, narrowband noise interference, broadband noise interference, comb-shaped spectrum interference, and swept-frequency interference, and obtains the timing characteristics of the interference signal. Specifically:

[0069] The single-tone interference signal is emitted at a certain frequency point and is a single-frequency continuous sine waveform. The time-domain expression is as follows:

[0070]

[0071] Among them, P J is the power of the interference signal, f J is the frequency of the interference signal, and θ J is the phase of the interference signal and follows a uniform distribution.

[0072] The multi-tone interference signal is emitted at multiple discrete frequency points, and the time-domain expression is as follows:

[0073]

[0074] Narrowband noise interference: Pass white noise through a narrowband filter to obtain a blocking signal aiming at a certain narrowband range. The frequency-domain expression of the narrowband filter is:

[0075]

[0076] Broadband noise interference is to pass Gaussian white noise through a broadband filter to obtain blocking noise in each broadband range. The frequency-domain expression of the filter is:

[0077]

[0078] Among them, P J is the interference power, and W I is the cut-off frequency of the low-pass filter.

[0079] Comb-like spectrum interference: Pass white noise through multiple narrow-band filters to obtain blocking signals targeting multiple narrow-band ranges. The frequency-domain expression of the filter is:

[0080]

[0081] Among them, i = 1, 2,..., Q, and Q represents the number of narrow-band filters.

[0082] Swept-frequency interference exhibits the characteristic of linear scanning with time within a partial spectrum. Its time-domain expression is:

[0083]

[0084] Among them, β i is the swept-frequency rate, ω i is the initial angular frequency, is the initial phase, and T is the swept-frequency duration.

[0085] In further other embodiments, before extracting the characteristic parameters of the interference signal, the signal is first preprocessed, including normalization and centering after normalization.

[0086] Normalization: Normalize the sample data to [-1, 1] using the median method:

[0087]

[0088] Centering k = 1, 2,..., N;

[0089] Calculation of time-domain characteristic parameters of the interference signal: including the time-domain moment skewness coefficient and the R parameter; the specific definitions are:

[0090] Time-domain moment skewness coefficient a 3 : Among them, μ is the mean of the envelope of the time-domain signal, and σ is its standard deviation. a 3 describes the degree of deviation of the sample from the normal distribution.

[0091] R parameter: The R parameter reflects the degree of change in the envelope of the sample x.

[0092] Calculation of frequency-domain characteristic parameters of the interference signal: including the carrier factor coefficient, the average spectrum flatness coefficient, the frequency-domain moment skewness coefficient, and R f parameter; the specific definitions are:

[0093] Carrier factor coefficient C: The carrier factor coefficient describes the prominence of the signal spectral lines. Among them, X[λ 1 is the maximum value of the frequency-domain signal, and X[λ 2 is the second-largest value of the frequency-domain signal.

[0094] Average spectral flatness coefficient Fse: Among them, N is the number of samples of the signal, and N s is the number of samples for spectral analysis. Fse reflects whether there are obvious impulse signals locally in the signal. is the mean value of the spectral amplitude X(n), and X 1 (n) is to perform smoothing filtering on the spectrum X(n).

[0095] R f Parameter: Among them, μ is the mean value of the spectral amplitude X, and σ is its standard deviation.

[0096] Frequency-domain moment skewness coefficient b 3 :

[0097] In further other embodiments, the Fourier transform is good at analyzing stationary signals with uniform and stable frequency characteristics. For mutations in signals, it is difficult for the Fourier transform to capture them in a timely manner.

[0098] After separately extracting the time-domain and frequency-domain characteristic parameters of the signal, the short-time Fourier transform (STFT) is used to convert the time-series characteristics of the interference signal into a two-dimensional time-frequency energy diagram. The short-time Fourier transform (STFT) is a method for analyzing how the frequency of a signal changes over time. It has been widely used in the analysis of non-stationary signals. The expression for performing the short-time Fourier transform on the input signal x(t) is as follows:

[0099]

[0100] By adjusting the position τ of the window w(t - τ) on the time axis, STFT provides a method to observe and analyze the frequency distribution of the signal in different time periods, thereby obtaining the time-frequency representation of the signal.

[0101] The obtained time-frequency diagram of the interference signal is input into a convolutional neural network (CNN) for image feature extraction. The CNN is a type of feedforward neural network that contains convolutional calculations and has a deep structure, and is one of the representative algorithms of deep learning. The time-frequency diagram feature extraction includes the following steps:

[0102] Step 1, randomly select a batch of time-frequency diagrams and their corresponding labels from the interference signal database. The interference signals and their corresponding labels are shown in Table 1;

[0103] Step 2: Input the batch of time-frequency diagrams into the CNN network model for classification and recognition. Take the activation vector output by the fully connected layer before the SoftMax layer as the feature vector of the time-frequency diagram. The network structure is shown in Table 2.

[0104] Step 3: Input the extracted feature vector into the supervised classifier to obtain the classification result, and update the parameters of the supervised classifier according to the cross-entropy loss function between the classification result and the true label.

[0105] Loop through Steps 1 - 3 until the network reaches a convergence state. At this time, the network learns the exact feature representation of the known interference patterns.

[0106] Table 1

[0107]

[0108] Table 2

[0109]

[0110] For the test sample signal, call the trained CNN network to extract its feature vector, and calculate the mean value of the feature vector to obtain the feature parameter mav of the time-frequency diagram of the signal.

[0111] In further other embodiments, fuse the feature parameters in the three dimensions of time domain, frequency domain, and time-frequency diagram. The specific fusion method is as follows: Compose the five groups of feature parameters into training data in the following format:

[0112] tr_data = [a3_list, R_list, C_list, b3_list, mav_list, type_list];

[0113] Among them, columns 1 - 5 are feature parameters, including time-domain features, frequency-domain features, and time-frequency domain features, and the 6th column is the data label, representing the nth type of interference signal.

[0114] Input the fused feature parameters into the LM neural network for signal type recognition. The LM neural network usually refers to specifying its training function as the Levenberg-Marquardt optimization algorithm when creating the PatternNet network.

[0115] PatternNet is a neural network algorithm for pattern recognition and classification problems, belonging to a feedforward network, consisting of an input layer, one or more hidden layers, and an output layer.

[0116] The Levenberg - Marquardt algorithm combines the advantages of the gradient descent method (steepest descent method) and the Gauss - Newton method, and balances the two by adjusting a damping factor, thereby accelerating the convergence speed and improving stability.

[0117] The parameter update rule of the LM algorithm is:

[0118]

[0119] where \(J\) is the Jacobian matrix, \(\nabla J\) is the gradient of the cost function, \(\lambda\) is the damping factor, and \(I\) is the identity matrix.

[0120] In the above - mentioned embodiments of the present invention, the interference signals adopt common wireless interference signals, such as single - tone interference, multi - tone interference, narrow - band noise interference, wide - band noise interference, comb - shaped spectrum interference, and swept - frequency interference. The time - frequency diagrams of various interference signals are generated by Matlab as the subsequent data set. The signal - to - noise ratio of the signals is 0 - 15 dB. Under this condition, simulation experiments are carried out and good performance is obtained.

[0121] In summary, in the above - mentioned embodiments of the present invention, first, the types of communication interference signal samples are disclosed, including single - tone interference, multi - tone interference, narrow - band noise interference, wide - band noise interference, comb - shaped spectrum interference, and swept - frequency interference, and the mathematical models of 6 typical jamming signals are disclosed. Secondly, the methods for extracting time - domain characteristic parameters and frequency - domain characteristic parameters are disclosed, and the fluctuation characteristics, correlation characteristics, and distribution characteristics of the interference signal data can be extracted quickly and simply. Then, further, the feature extraction of the interference signal time - frequency diagram by using a convolutional neural network is disclosed, and the basic structure of the convolutional neural network model is described. For 6 interference types, simulation experiments are carried out at a signal - to - noise ratio of 0 - 15 dB. The communication interference signal recognition based on multi - dimensional feature parameter fusion proposed by the present invention has achieved good performance in the final test and provides a new alternative solution in the field of communication interference signal recognition.

[0122] The units involved in the embodiments described in the present invention can be implemented in software or in hardware, and the described units can also be set in the processor. Among them, the names of these units do not constitute a limitation to the unit itself in some cases.

[0123] According to one aspect of the embodiments of the present invention, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer - readable storage medium. The processor of the computer device reads the computer instructions from the computer - readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the above various optional implementation manners.

[0124] 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 embodiments; or may exist alone without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiments.

Claims

1. A method for identifying interference signals based on multi-dimensional feature fusion, characterized in that: The following steps are involved: S1, model the interference signal and obtain the time series characteristics of the interference signal; S2, extracting time domain characteristic parameters and frequency domain characteristic parameters of the interference signal; S3, converting the time series characteristics of the interference signal into time-frequency characteristics to obtain a two-dimensional time-frequency energy map; S4, constructing a neural network model 1, inputting the two-dimensional time-frequency energy map obtained in step S3 into the neural network model 1, and using the fully connected layer of the network to extract characteristic parameters of the interference signal; S5, fusing the feature parameters of the three dimensions in step S2 and step S4, and inputting the fused feature parameters into the neural network model 2 to identify the interference signal.

2. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S1, the modeling of interference signals specifically includes modeling interference signals in six communication interference fields, namely, single-tone interference signals, multi-tone interference signals, narrowband noise interference signals, broadband noise interference signals, comb spectrum interference signals and swept frequency interference signals.

3. The interference signal identification method based on multi-dimensional feature fusion according to claim 2 is characterized in that: The single-tone interference signal is transmitted at a certain frequency point and is a single-frequency continuous sine waveform. The time domain expression is as follows: Among them, P J is the interference signal power, f J is the interference signal frequency, θ J is the interference signal phase and obeys uniform distribution.

4. The interference signal identification method based on multi-dimensional feature fusion according to claim 2 is characterized in that: The multi-tone interference signal is transmitted at multiple discrete frequency points, and the time domain expression is as follows:

5. The interference signal identification method based on multi-dimensional feature fusion according to claim 2 is characterized in that: The narrowband noise interference specifically passes white noise through a narrowband filter to obtain a blocking signal targeting a certain narrowband range. The frequency domain expression of the narrowband filter is:

6. The interference signal identification method based on multi-dimensional feature fusion according to claim 2 is characterized in that: The broadband noise interference signal is Gaussian white noise passing through a broadband filter to obtain a blocking noise code in each broadband range. The frequency domain expression of the filter is: Among them, P J is the interference power, W I is the cutoff frequency of the low-pass filter.

7. The interference signal identification method based on multi-dimensional feature fusion according to claim 2 is characterized in that: The comb spectrum interference specifically passes white noise through multiple narrowband filters to obtain blocking signals targeting multiple narrowband ranges. The filter frequency domain expression is: Wherein, i=1,2,…,Q, and Q represents the number of narrowband filters.

8. The interference signal identification method based on multi-dimensional feature fusion according to claim 2 is characterized in that: The swept frequency interference presents the characteristic of linear scanning over time in part of the spectrum, and its time domain expression is: Among them, β i is the sweep rate, ω i is the initial angular frequency, is the initial phase, and T is the sweep duration.

9. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S2, before extracting the time domain characteristic parameters and frequency domain characteristic parameters of the interference signal, the step of preprocessing the signal is included, including normalization and post-normalization centering processing.

10. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S2, extracting the time domain characteristic parameters of the interference signal specifically includes the following sub-steps: Extract the time domain moment skewness coefficient and R parameter, and calculate them according to the following formula: Time domain moment skewness coefficient a3: Among them, μ is the mean of the time domain signal envelope, σ is its standard deviation, and a3 is used to describe the degree to which the sample deviates from the normal distribution; R parameter: The R parameter is used to reflect the degree of change of the sample x envelope.

11. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S2, the frequency domain characteristic parameters of the interference signal are extracted, which specifically includes the following sub-steps: Extract carrier factor coefficient, average spectrum flatness coefficient, frequency domain moment skewness coefficient, R f Parameters; specifically calculated according to the following formula: Carrier factor coefficient C: The carrier factor coefficient describes the prominence of the signal spectrum line; where X[λ1] is the maximum value of the frequency domain signal, and X[λ2] is the second maximum value of the frequency domain signal; Average spectrum flatness coefficient Fse: Where N is the number of samples of the signal, N s is the number of samples for spectrum analysis; Fse reflects whether there is an obvious impulse signal in the local part of the signal; is the mean value of the spectrum amplitude X(n), and X1(n) is the smoothing filter of the spectrum X(n); R f parameter: Among them, μ is the mean of the spectrum amplitude X, and σ is its standard deviation; Frequency domain moment skewness coefficient b3:

12. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S3, the time-frequency feature conversion is performed on the time series feature of the interference signal to obtain a two-dimensional time-frequency energy graph, which specifically includes the following sub-steps: Use short-time Fourier transform to convert the time series characteristics of the interference signal into a two-dimensional time-frequency energy diagram. The expression of short-time Fourier transform of the input signal x(t) is as follows: By adjusting the position τ of the window w(t-τ) on the time axis, STFT provides a method to observe and analyze the frequency distribution of the signal in different time periods, thereby obtaining the time-frequency representation of the signal.

13. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S4, the two-dimensional time-frequency energy map obtained in step S3 is input into the neural network model 1, and the characteristic parameters of the interference signal are extracted using the fully connected layer of the network, which specifically includes the following sub-steps: Step S41, randomly selecting a batch of time-frequency graphs and their corresponding labels from the interference signal database; Step S42, inputting the batch of time-frequency graphs into a neural network model 1 for classification and recognition, and using the activation vector output by the fully connected layer before the SoftMax layer as the feature vector of the time-frequency graph, wherein the neural network model 1 includes a CNN network model; Step S43, inputting the extracted feature vector into a supervised classifier to obtain a classification result, and updating the supervised classifier parameters according to a cross entropy loss function between the classification result and the true label; Step S44, looping through steps S41 to S43 until the network reaches a convergence state, at which point the network learns the precise feature representation of the known interference pattern.

14. The interference signal identification method based on multi-dimensional feature fusion according to claim 1 is characterized in that: In step S5, the feature parameters of the three dimensions in step S2 and step S4 are fused, and the fused feature parameters are input into the neural network model 2 to identify the interference signal, which specifically includes the following sub-steps: S51, five sets of feature parameters are combined into training data in the following format: tr_data=[a3_list,R_list,C_list,b3_list,mav_list,type_list]; Among them, a3_list, R_list, C_list, b3_list, mav_list in columns 1-5 are feature parameters, including time domain features, frequency domain features, and time-frequency domain features. The sixth column type_list is a data label, indicating the nth type of interference signal; S52, inputting the fused feature parameters into a second neural network model for signal type recognition; the second neural network model includes a LM neural network.

15. An interference signal recognition device based on multi-dimensional feature fusion, characterized in that: The method comprises 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 14 is executed.

Images