A method and device for identifying interference signals based on a hybrid neural network

Abstract

The application discloses a kind of interference signal identification method and device based on hybrid neural network, it is related to communication technical field.The specific embodiment of the method includes: obtaining multiple signal samples and its real label;Multiple signal samples are converted into corresponding time-frequency image respectively;Based on multiple time-frequency images input hybrid neural network, obtain the predicted value of each signal sample;Wherein, hybrid neural network sequentially includes: input layer, first convolution block attention module, maximum pooling layer, hybrid hollow convolution layer, second convolution block attention module, average pooling layer, gated recurrent unit and fully connected layer;Based on the real label and predicted value of each signal sample, and preset loss function, calculate loss value;Based on loss value adjustment hybrid neural network's parameter;Based on the interference signal in the hybrid neural network of well-trained identification current signal.The embodiment can improve the identification accuracy of interference signal.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of communication technology, and in particular to a method and device for identifying interference signals based on a hybrid neural network. BACKGROUND

[0002] With the advent of the information age, various communication means have been widely applied to every corner of people's life. Among them, satellite communication has attracted wide attention in the field of wireless communication because of its wide communication range, large capacity and high reliability. However, the satellite communication system is exposed to the open environment for a long time, and various environmental noises, space interference and malicious human interference will seriously affect the quality of satellite communication. Therefore, in order to ensure that satellite communication can be carried out normally in a harsh environment, various anti-interference methods have emerged.

[0003] The premise of anti-interference is to accurately and timely classify and identify the interference signals in the received signals. As an important part of the satellite communication system anti-interference, interference signal identification can provide support for subsequent interference elimination and interference suppression methods, and improve the performance of the satellite communication system.

[0004] For different modal data, the existing recognition method usually uses different networks for recognition, for example, model 1 is used for processing time domain data, and model 2 is used for processing frequency domain data. However, this method only uses single modal data in the identification process, ignoring the features in other modal data, resulting in low recognition accuracy. SUMMARY

[0005] Therefore, the embodiments of the present application provide a method and device for identifying interference signals based on a hybrid neural network, which can improve the recognition accuracy of interference signals.

[0006] In a first aspect, the embodiments of the present application provide a method for identifying interference signals based on a hybrid neural network, comprising:

[0007] Obtaining a plurality of signal samples and their true labels; wherein the plurality of signal samples include: an interference-free sample obtained by superimposing a noise signal and a useful signal, and interference samples corresponding to a plurality of interference categories respectively; any interference sample is obtained by superimposing a noise signal, a useful signal and an interference signal of any interference category;

[0008] Converting the plurality of signal samples into corresponding time-frequency images respectively;

[0009] inputting a plurality of time-frequency image into a hybrid neural network to obtain a prediction value of each of the signal samples; wherein the hybrid neural network comprises, in sequence, an input layer, a first convolution block attention module, a max-pooling layer, a hybrid dilated convolution layer, a second convolution block attention module, an average-pooling layer, a gated recurrent unit and a fully connected layer;

[0010] calculating a loss value based on the true label and the prediction value of each of the signal samples and a preset loss function;

[0011] adjusting parameters of the hybrid neural network based on the loss value;

[0012] identifying an interference signal in a current signal based on the trained hybrid neural network.

[0013] In a second aspect, an interference signal identification device based on a hybrid neural network is provided, comprising:

[0014] an obtaining module configured to obtain a plurality of signal samples and true labels thereof; wherein the plurality of signal samples comprises: an interference-free sample obtained by superimposing a noise signal and a useful signal, and interference samples corresponding to a plurality of interference categories respectively; any of the interference samples is obtained by superimposing a noise signal, a useful signal and an interference signal of any interference category;

[0015] a converting module configured to convert the plurality of signal samples into corresponding time-frequency images respectively;

[0016] a training module configured to input a plurality of time-frequency images into a hybrid neural network to obtain a prediction value of each of the signal samples; wherein the hybrid neural network comprises, in sequence, an input layer, a first convolution block attention module, a max-pooling layer, a hybrid dilated convolution layer, a second convolution block attention module, an average-pooling layer, a gated recurrent unit and a fully connected layer; a loss value is calculated based on the true label and the prediction value of each of the signal samples and a preset loss function; parameters of the hybrid neural network are adjusted based on the loss value;

[0017] a prediction module configured to identify an interference signal in a current signal based on the trained hybrid neural network.

[0018] In a third aspect, an electronic device is provided, comprising:

[0019] one or more processors;

[0020] a storage device configured to store one or more programs,

[0021] when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any of the above embodiments.

[0022] In a fourth aspect, an embodiment of the present application provides a computer readable medium, having stored thereon a computer program, which, when executed by a processor, implements the method according to any one of the above embodiments.

[0023] An embodiment of the above application has the following advantages or beneficial effects: based on the mixed neural network, the multi-modal data can be processed, in the recognition process, the features in different modal data are considered, and the recognition accuracy is improved. Based on the attention mechanism provided by the convolution block attention module, a plurality of categories of interference signals can be accurately recognized. The CLDNN (Convolutional Neural Network Long Short Term Memory Deep Neural Network) is improved, the LSTM is replaced by the GRU (Gate Recurrent Unit), the model structure is simplified while the performance is maintained, and then the training speed is improved; the convolutional neural network is replaced by the mixed dilated convolution layer, the feature map receptive field is expanded in the case of reducing the amount of calculation, the correlation of the data in a larger space range is captured, and then the recognition accuracy is improved. The CBAM (Convolutional Block Attention Module) is introduced before and after the mixed dilated convolution layer to realize feature optimization selection, the effective feature information is strengthened and the invalid feature information is weakened through the channel attention mechanism and the spatial attention mechanism contained in the CBAM, and then the classification performance and the generalization ability of the model are improved.

[0024] Further effects of the above-described non-conventional optional mode will be described in the following in conjunction with the specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0025] The accompanying drawings are used to better understand the present application and do not constitute an improper limitation on the present application. Among them:

[0026] Figure 1 is a flowchart of a mixed neural network-based interference signal recognition method provided by an embodiment of the present application;

[0027] Figure 2 is a schematic diagram of a mixed neural network provided by an embodiment of the present application;

[0028] Figure 3 is a schematic diagram of a CBAM provided by an embodiment of the present application;

[0029] Figure 4 is a schematic diagram of a mixed neural network-based interference signal recognition device provided by an embodiment of the present application;

[0030] Figure 5 is a structural schematic diagram of a computer system of a terminal device or a server suitable for implementing embodiments of the present application. DETAILED DESCRIPTION

[0031] Exemplary embodiments of the present application are described below with reference to the accompanying drawings, which include various details of the embodiments of the present application to assist in understanding, and should be considered as merely exemplary. Thus, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of known functions and constructions are omitted in the following description for clarity and conciseness.

[0032] As Figure 1 shown, the embodiments of the present application provide a method for identifying interference signals based on a hybrid neural network, comprising:

[0033] Step 101: Obtain a plurality of signal samples and their true labels.

[0034] Among the plurality of signal samples, there are: non-interference samples obtained by superimposing noise signals and useful signals, and interference samples corresponding to a plurality of interference categories respectively; any interference sample is obtained by superimposing noise signals, useful signals and interference signals of any interference category.

[0035] The signal samples can be generated by simulation software, for example, by MATLAB software. One interference sample corresponds to one interference category, and one interference category corresponds to a plurality of interference samples with different jamming ratios. The labels of the signal samples include: non-interference and interference category. For example, the true label of the non-interference sample is non-interference, and the true label of the interference sample is its corresponding interference category.

[0036] Step 102: Convert the plurality of signal samples into corresponding time-frequency images respectively.

[0037] Step 103: Input the plurality of time-frequency images into a hybrid neural network to obtain the predicted values of each signal sample.

[0038] As Figure 2 shown, the hybrid neural network sequentially includes: an input layer, a first convolutional block attention module, a maximum pooling layer, a hybrid dilated convolutional layer, a second convolutional block attention module, an average pooling layer, a gated recurrent unit and a fully connected layer.

[0039] Step 104: Based on the true labels and the predicted values of each signal sample, and a preset loss function, calculate the loss value.

[0040] Step 105: Adjust the parameters of the hybrid neural network based on the loss value.

[0041] Step 106: identifying the interference signal in the current signal based on the trained hybrid neural network.

[0042] Based on the hybrid neural network, the multi-modal data can be processed, and in the identification process, the features in different modal data are considered to improve the identification accuracy. Based on the attention mechanism provided by the convolution block attention module, a variety of interference signals can be accurately identified. The CLDNN is improved by replacing the LSTM with GRU, which can simplify the model structure while maintaining the performance, thereby improving the training speed; the convolutional neural network is replaced by a hybrid dilated convolution layer, which can expand the feature map receptive field while reducing the amount of calculation, capture the correlation of data in a larger spatial range, and thus improve the identification accuracy. CBAM is introduced before and after the hybrid dilated convolution layer to realize feature optimization selection, and the effective feature information is strengthened and the invalid feature information is weakened through the channel attention mechanism and spatial attention mechanism contained in CBAM, thereby improving the classification performance and generalization ability of the model.

[0043] In an embodiment of the present application, taking satellite communication as an example, the useful signal is a BPSK (Binary Phase Shift Keying) modulated signal, the noise signal is additive white Gaussian noise, and the signal parameters are set as follows: the signal-to-noise ratio is 5 dB, the jamming-to-signal ratio is [-30 dB, 10 dB] with an interval of 4 dB, the carrier frequency is 140 KHz, and the sampling frequency is 800 KHz.

[0044] The interference categories include single-tone interference, multi-tone interference, linear sweep interference, pulse interference, comb spectrum interference, narrowband noise interference, and wideband noise interference.

[0045] Among them, the frequency points of the single-tone interference are randomly generated within the bandwidth, the frequency set of the multi-tone interference is randomly generated within the bandwidth, the initial frequency of the linear sweep interference is 120 KHz, the sweep rate is 8 MHz, the pulse interference is a rectangular pulse with a duty cycle of 2%, the total bandwidth of the comb spectrum interference accounts for more than 1 / 2, the bandwidth of the narrowband noise interference accounts for 1 / 10, and the bandwidth of the wideband noise interference accounts for more than 2 / 3.

[0046] The number of signal samples of each interference category under each jamming-to-signal ratio is 250, and the signal samples are divided into a training set, a validation set, and a test set according to a ratio of 6:2:2.

[0047] In an embodiment of the present application, the plurality of signal samples are respectively converted into corresponding time-frequency images, including: converting the plurality of signal samples into corresponding time-frequency images based on short-time Fourier transform.

[0048] In the embodiment of the present application, the resolution of the time-frequency image is 3x224x224. The embodiment of the present application converts the signal data into a two-dimensional time-frequency image through short-time Fourier transform, concentrates the advantages of time domain features and frequency domain features, and further improves the recognition accuracy. In addition, the short-time Fourier transform can also be replaced by wavelet transform.

[0049] In an embodiment of the present application, the input layer comprises three second feature extraction units.

[0050] Any second feature extraction unit comprises a two-dimensional convolution layer, a two-dimensional batch normalization layer and a rectified linear unit.

[0051] The input layer uses the second feature extraction unit to extract features from the input time-frequency image. The second feature extraction unit is composed of three basic units, which are a two-dimensional convolution layer (Conv2d), a two-dimensional batch normalization layer (BatchNorm2d) and a rectified linear unit (ReLU). The convolution kernel size of the two-dimensional convolution layer is 7x7, and the number of channels is 16. The resolution of the time-frequency image is converted from 3x224x224 to 16x112x112.

[0052] In an embodiment of the present application, the first convolution block attention module comprises a first channel attention module and a first spatial attention module.

[0053] The second convolution block attention module comprises a second channel attention module and a second spatial attention module.

[0054] The first convolution block attention module and the second channel attention module are two CBAMs (Convolutional Block Attention Modules) with the same structure, which are arranged before and after the mixed dilated convolution layer, respectively.

[0055] As shown in Figure 3 , the convolution block attention module is composed of a channel attention module and a spatial attention module in series. The input of the channel attention module is F∈R C◇H◇W , and the output is a one-dimensional channel attention feature map M C ∈R C◇1◇1 , which is input into the spatial attention module to obtain a two-dimensional spatial attention feature map M S ∈R 1◇H◇W . The whole attention process can be summarized as:

[0056]

[0057]

[0058] Wherein, F is used to represent the feature map of the input channel attention module, R is used to represent the real domain, C is used to represent the channel number, H is used to represent the height of the feature map, W is used to represent the width of the feature map, F' is used to represent the feature map after the channel attention mechanism processing, and F'' is used to represent the feature map after the spatial attention mechanism processing.

[0059] Formula (1) means that the feature map after the channel attention mechanism processing = channel attention feature map * original feature map.

[0060] Formula (2) means that the feature map after the spatial attention mechanism processing = spatial attention feature map * feature map after the channel attention mechanism processing.

[0061] The channel attention module focuses on the relationship between different channels of the time-frequency image, gives different weights to each channel, and strengthens the attention to important channels, thereby helping to improve the representation ability of the model.

[0062] In the channel attention module, the feature map F is extracted through two parallel branches of maximum pooling and average pooling to obtain global features, and two spatial feature maps are generated: and The channel number is compressed to 1 / s of the original channel number through a multi-perception machine (MLP), and then expanded back to the original channel number, wherein s represents the original channel number, and preferably s = 4. Then, the two output feature maps are added element by element, and the feature information is activated by a Sigmoid function to obtain the output result. The above calculation process is shown in formula (3).

[0063] Formula (3):

[0064]

[0065] Wherein, σ is the Sigmoid function, and W0 and W1 are the weight values of the two fully connected layers of the multi-perception machine.

[0066] The spatial attention module focuses on the spatial structure inside the time-frequency image, and gives different weights to each pixel position, so that the model can focus on important areas in the time-frequency image. In the spatial attention module, the feature map F' output by the channel attention module is taken as the input, and maximum pooling and average pooling are used to aggregate the channel information of the feature map to obtain two 2D features: and Then, they are connected and convolved by a standard convolution layer and a Sigmoid function to generate a spatial attention map, and the process is shown in formula (4).

[0067]

[0068] Wherein, f7◇7 This represents a convolution operation with a kernel size of 7×7. Finally, M... S The new feature is obtained by multiplying (F′) element-wise with the feature map F′, and the local information of the new feature is richer.

[0069] The output of CBAM is fed into a max pooling layer with a filter size of 3×3 and a stride of 1, and the output feature map has a size of 32×56×56. The feature map is then fed into a hybrid dilated convolutional layer.

[0070] In one embodiment of the present invention, the hybrid dilated convolutional layer includes: 4 layers of first feature extraction units;

[0071] The first feature extraction unit in any layer includes: a two-dimensional convolutional layer, a two-dimensional batch normalization layer, and a parameter-corrected linear unit;

[0072] The number of channels in the four 2D convolutional layers are 32, 64, 32 and 16, respectively, the convolutional kernels are all 3×3, and the dilation rates are 1, 3, 7 and 1, respectively.

[0073] Dilated convolution introduces a new parameter called "dilation rate" to the convolutional layer, which defines the number of intervals between points in the convolution kernel. A dilated convolution with a dilation rate of r introduces r-1 zeros between the weights of a normal convolution kernel, which is equivalent to expanding a k×k convolution kernel to k+(k-1)×(r-1) without increasing any parameters or computational cost.

[0074] When designing hybrid dilated convolutions, the following requirements should be met: ① The dilation rates of the stacked convolutions cannot have a common divisor greater than 1; ② The dilation rate should be designed as a zigzag structure, such as [1, 2, 5, 1, 2, 5]; ③ Assuming there are n convolutional layers, the kernel size is k×k, and the dilation rate is [r1, ..., r...] i ,...,r n The goal of hybrid dilated convolution is to allow the receptive field of a series of convolution operations to cover a square region without any holes or missing edges. The maximum distance between two non-zero values ​​is defined as Equation (5).

[0075] M i =max[M i+1 -2r i M i+1 -2(M i+1 -r i ), r i (5)

[0076] 1≤i≤n-1; M n =r n The design goal is to make M² ≤ k. Where r i M is used to characterize the expansion rate of layer i. iThe maximum expansion rate is used to characterize the i layer, n is used to characterize the total number of mixed hollow convolution layers, and k is used to characterize the convolution kernel size.

[0077] The embodiment of the application uses the superposition of mixed hollow convolution with different expansion rates to expand the receptive field. The output result of the mixed hollow convolution layer is sent to the CBAM, and then passes through the 4*4 average pooling layer to reduce the parameter quantity. Finally, the extracted feature information is converted into the GRU.

[0078] The GRU includes an update gate and a reset gate. The update gate is used to control the degree of state information at the previous moment being brought into the current state, and the reset gate is used to control how much information of the previous state is written. The size of the hidden layer is 128. Through the GRU, the feature data is converted from 16*4*4 to 128 and is sent to the fully connected layer.

[0079] The fully connected layer maps the outputs of multiple neurons to probabilities in [0, 1] through a softmax activation function to realize multi-label classification. The loss function is a cross-entropy loss function.

[0080] In the training process, the loss value calculated according to the loss function is used to adjust the learning rate adaptively using the adaptive optimization algorithm Adam optimizer. The initial learning rate is 0.0001, and the total iteration is 40 times. After the training is stable, the validation set and the test set are input into the trained mixed neural network to test the interference recognition accuracy.

[0081] As shown in Figure 4 , the embodiment of the application provides an interference signal recognition device based on a mixed neural network, which comprises:

[0082] The acquisition module 401 is configured to acquire a plurality of signal samples and their true labels. The plurality of signal samples include: an interference-free sample obtained by superimposing a noise signal and a useful signal, and interference samples corresponding to a plurality of interference categories respectively. Any interference sample is obtained by superimposing a noise signal, a useful signal and an interference signal of any interference category.

[0083] The conversion module 402 is configured to convert the plurality of signal samples into corresponding time-frequency images respectively.

[0084] The training module 403 is configured to input the mixed neural network based on the plurality of time-frequency images to obtain the predicted value of each signal sample. The mixed neural network comprises, in sequence: an input layer, a first convolution block attention module, a maximum pooling layer, a mixed hollow convolution layer, a second convolution block attention module, an average pooling layer, a gated recurrent unit and a fully connected layer. Based on the true label and the predicted value of each signal sample and the preset loss function, the loss value is calculated. The parameters of the mixed neural network are adjusted based on the loss value.

[0085] The prediction module 404 is configured to identify the interference signal in the current signal based on the trained hybrid neural network.

[0086] Reference will now be made to the following description Figure 5 which shows a schematic diagram of a computer system 500 suitable for implementing the terminal device of an embodiment of the present application. Figure 5 The terminal device shown is merely an example and should not impose any limitation on the functions and scope of use of the embodiments of the present application.

[0087] As Figure 5 shown, the computer system 500 includes a central processing unit (CPU) 501 which can perform various appropriate actions and processes in accordance with a program stored in a read only memory (ROM) 502 or a program loaded from a storage section 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0088] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as necessary. A removable media 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 510 as necessary, so that a computer program read therefrom is installed into the storage section 508 as necessary.

[0089] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program in accordance with the embodiments disclosed herein. For example, the embodiments disclosed herein include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 509 and / or installed from the removable media 511. When the computer program is executed by the central processing unit (CPU) 501, the above-described functions defined in the system of the present application are performed.

[0090] It should be noted that the computer-readable medium shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0091] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code containing one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order than that shown in the drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0092] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. The described modules can also be arranged in a processor, for example, a processor can be described as including a sending module, an obtaining module, a determining module and a first processing module. In some cases, the names of these modules do not constitute a limitation on the modules themselves, for example, the sending module can also be described as "a module that sends a picture obtaining request to a connected server".

[0093] The specific embodiments described above do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can occur depending on design requirements and other factors. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application shall fall within the scope of the protection of the present application.

Claims

1. A method for identifying an interference signal based on a hybrid neural network, characterized in that, The method comprises the following steps: obtaining a plurality of signal samples and their true labels; wherein the plurality of signal samples include: an undisturbed sample obtained by superimposing a noise signal and a useful signal, and a plurality of interference samples corresponding to a plurality of interference categories respectively; any of the interference samples is obtained by superimposing a noise signal, a useful signal and an interference signal of any interference category; converting the plurality of signal samples into corresponding time-frequency images respectively; inputting a plurality of time-frequency images into a hybrid neural network to obtain the predicted values of each of the signal samples; wherein the hybrid neural network comprises, in sequence: an input layer, a first convolutional block attention module, a maximum pooling layer, a hybrid dilated convolution layer, a second convolutional block attention module, an average pooling layer, a gated recurrent unit and a fully connected layer; the input layer comprises: three second feature extraction units; wherein any of the second feature extraction units comprises: a two-dimensional convolution layer, a two-dimensional batch normalization layer and a rectified linear unit; the hybrid dilated convolution layer comprises: four first feature extraction units; wherein any of the first feature extraction units comprises: a two-dimensional convolution layer, a two-dimensional batch normalization layer and a parameter rectified linear unit; the channel numbers of the four two-dimensional convolution layers are 32, 64, 32 and 16 respectively, the convolution kernels are all 3x3, and the dilation rates are 1, 3, 7 and 1 respectively; the first convolutional block attention module comprises: a first channel attention module and a first spatial attention module; the second convolutional block attention module comprises: a second channel attention module and a second spatial attention module, and the first channel attention module and the second channel attention module are two convolutional block attention modules CBAM with the same structure; the gated recurrent unit comprises an update gate and a reset gate, the update gate is used to control the degree of state information at the previous moment being brought into the current state, and the reset gate is used to control how much information of the previous state is written, and the size of the hidden layer thereof is 128; calculating a loss value based on the true labels and the predicted values of each of the signal samples and a preset loss function; adjusting the parameters of the hybrid neural network based on the loss value; identifying the interference signal in the current signal based on the trained hybrid neural network, wherein converting the plurality of signal samples into corresponding time-frequency images respectively comprises: converting the plurality of signal samples into corresponding time-frequency images respectively based on short-time Fourier transform.

2. The method of claim 1, wherein the interference categories include: single-tone interference, multi-tone interference, linear sweep interference, pulse interference, comb spectrum interference, narrowband noise signal interference and wideband noise signal interference.

3. The method of claim 1, wherein the useful signal is a BPSK modulated signal, the noise signal is additive white Gaussian noise, and the signal-to-interference ratio is [-30dB, 10dB].

4. A hybrid neural network-based interference signal recognition apparatus, characterized by, ​ An acquisition module configured to acquire a plurality of signal samples and their true labels; wherein the plurality of signal samples include: an interference-free sample obtained by superimposing a noise signal and a useful signal, and interference samples corresponding to a plurality of interference categories respectively; any of the interference samples is obtained by superimposing a noise signal, a useful signal and an interference signal of any interference category; A conversion module configured to convert the plurality of signal samples into corresponding time-frequency images respectively; A training module configured to input a plurality of time-frequency images into a hybrid neural network to obtain predicted values of each of the signal samples; wherein the hybrid neural network sequentially includes: an input layer, a first convolutional block attention module, a maximum pooling layer, a hybrid dilated convolution layer, a second convolutional block attention module, an average pooling layer, a gated recurrent unit and a fully connected layer; the input layer includes: 3 layers of second feature extraction units; wherein any of the second feature extraction units includes: a two-dimensional convolution layer, a two-dimensional batch normalization layer and a rectified linear unit; the hybrid dilated convolution layer includes: 4 layers of first feature extraction units; wherein any of the first feature extraction units includes: a two-dimensional convolution layer, a two-dimensional batch normalization layer and a parameter rectified linear unit; the channel numbers of the 4 layers of two-dimensional convolution layers are 32, 64, 32 and 16 respectively, the convolution kernels are all 3×3, and the dilation rates are 1, 3, 7 and 1 respectively; the first convolutional block attention module includes: a first channel attention module and a first spatial attention module; the second convolutional block attention module includes: a second channel attention module and a second spatial attention module, the first channel attention module and the second channel attention module are two convolutional block attention modules CBAM with the same structure; the gated recurrent unit includes an update gate and a reset gate, the update gate is used to control the degree of state information at the previous moment being brought into the current state, the reset gate is used to control how much information of the previous state is written, and the size of the hidden layer thereof is 128; based on the true labels and the predicted values of each of the signal samples, and a preset loss function, a loss value is calculated; based on the loss value, parameters of the hybrid neural network are adjusted; A prediction module configured to identify interference signals in a current signal based on the trained hybrid neural network. The conversion module is further configured to convert the plurality of signal samples into corresponding time-frequency images based on short-time Fourier transform.

5. An electronic device, comprising: One or more processors; A storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-3. The program is executed by the processor to implement the method of any one of claims 1-3.

6. A computer readable medium having stored thereon a computer program, characterized in that, ​

Images