A radio frequency modulation fuze jamming signal recognition method based on convolutional neural network and attention mechanism

Abstract

The present application relates to a kind of radio frequency fuze jamming signal identification method based on convolutional neural network and attention mechanism, comprising: the preprocessed training data set is input to radio frequency fuze jamming signal identification model and carries out offline training, and optimization radio frequency fuze jamming signal identification network model;Simulation generates test data set, and the identification performance of radio frequency fuze jamming signal identification model after offline training is verified using test data set;The radio frequency fuze jamming signal to be identified is input after the data preprocessing, and the radio frequency fuze jamming signal identification model after verification is input to signal identification.The identification accuracy of network reaches more than 98% when the present application is in the ratio of-16dB.

Description

Technical Field

[0001] This invention relates to a method for identifying interference in radio linear frequency modulated fuses based on convolutional neural networks and attention mechanisms, belonging to the field of radio fuse technology. Background Technology

[0002] A fuze is a device that uses environmental, target, or platform information to ensure the safety of ammunition handling and ballistic trajectory, detonating the ammunition according to a predetermined strategy. As the core control component of an ammunition system, the fuze directly determines the efficient destructive power of weapons and equipment, and is figuratively referred to as the brain of the ammunition. Countries have a profound understanding of the status and role of fuzes, and its importance has now been elevated to the level of system-on-systems confrontation, highlighting the significance of fuze research. Fuzes face complex electromagnetic environments during storage, handling, launch, and flight trajectory, including both natural and man-made interference. Fuze jammers, in particular, are intentional acts of interference, and their methods are constantly evolving and becoming more sophisticated, posing an increasingly serious threat to radio fuzes. This necessitates stronger interference identification and anti-interference capabilities for radio fuzes. Therefore, the problem of interference identification for radio fuzes has become even more crucial. Traditional methods for identifying interference signals typically extract the time-domain features of the signal, which not only requires extensive analysis and processing but also often results in low accuracy in complex environments. In recent years, deep learning has been widely applied in many fields with significant results. In image processing, convolutional neural networks (CNNs) have garnered significant attention and yielded remarkable results due to their powerful feature learning and data processing capabilities. Conversely, attention mechanisms have also become a research hotspot due to their robust resource optimization and allocation capabilities. Therefore, jointly designing and applying CNNs and attention mechanisms to radio frequency modulation (FM) fuze systems holds promise for achieving better interference signal identification.

[0003] Dai Jian et al. (see Dai Jian, Yan Qi, Yan Xiaopeng et al. Pulse Doppler Fuze Interference and Target Signal Recognition Based on Fuzzy C-Mean Incremental Update [J]. Acta Ordnance et al., 2018, 39(09): 1711-1718.) proposed a fuzzy c-means (FCM) clustering algorithm with adaptive incremental update function. Based on the analysis of the fuze range gated output signal, the algorithm uses the time-domain and frequency-domain entropy characteristics of the signal to classify and identify interference and target signals with the help of the FCM algorithm. In view of the continuous deterioration of the signal-to-noise ratio, the improved incremental update algorithm realizes the adaptive update and adjustment of the fuze FCM classification model, so that the fuze can achieve an accuracy of 96.43% in identifying interference signals under -15.0dB conditions. However, this method requires manual feature extraction and a lot of professional knowledge.

[0004] With the development of information-based weaponry, the role and status of radio fuses have become more prominent, and the consequences of fuse failure have become more serious. Therefore, effectively identifying fuse interference signals is crucial for the development of radio frequency modulation fuses. Summary of the Invention

[0005] To address the interference identification problem of radio linear frequency modulation fuses, this invention proposes an interference identification method based on convolutional neural networks and attention mechanisms. When the interference-to-signal ratio is -16dB, the network's identification accuracy reaches over 98%.

[0006] Terminology Explanation:

[0007] 1. Radio Frequency Modulation (FM) Fuze Interference Identification System: This system consists of a target detector, a jammer, and a signal processing module. The FM fuze signal is transmitted through the target detector. When the fuze signal reaches the target, it is reflected to form an echo signal. This echo signal is interfered with by the jammer and external noise. The echo signal, the interference signal, and the noise together form the received signal, which is received by the target detector. The signal processing module then identifies the type of interference signal.

[0008] 2. Channel AM Model: Compressed Excitation Network; The Compressed Excitation Network is a channel attention model that represents the importance of different channel feature maps by assigning corresponding weights to each channel.

[0009] 3. Squeeze operation: A compression operation in the compression-excited network. It compresses the feature map into a feature vector by performing global average pooling on the feature map.

[0010] 4. Excitation operation: The excitation operation in the compressed excitation network learns the weights of each channel through fully connected layers and non-linear activation functions to capture the relationships between channels.

[0011] The technical solution of the present invention is as follows:

[0012] A method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks (CNN) and attention mechanisms (AM)

[0013] The method for identifying radio frequency modulation (FM) fuze interference signals is implemented through a radio frequency modulation (FM) fuze interference identification system, which includes a target detector, a jammer, and a signal processing module.

[0014] The target detector includes a transmitter and a receiver; the jammer generates several typical jamming signals; the signal processing module refers to the jamming identifier based on CNN and AM; including:

[0015] The training dataset is generated through simulation, and the data is preprocessed.

[0016] The preprocessed training dataset is input into the radio frequency modulation fuze interference signal recognition model for offline training to optimize the radio frequency modulation fuze interference signal recognition network model; a test dataset is generated by simulation, and the recognition performance of the radio frequency modulation fuze interference signal recognition model after offline training is verified using the test dataset;

[0017] After the radio frequency modulation fuze interference signal to be identified is preprocessed, it is input into the verified radio frequency modulation fuze interference signal identification model for signal identification to obtain the type of interference signal.

[0018] Further preferably, the jammer generates six typical interference signals, including: sinusoidal amplitude modulation interference signal, sinusoidal frequency modulation interference signal, noise frequency modulation interference signal, noise amplitude modulation interference signal, linear frequency modulation sweep interference signal, and logarithmic frequency modulation sweep interference signal.

[0019] According to a preferred embodiment of the present invention, a training dataset is generated through simulation, and data preprocessing is performed; including:

[0020] In a radio frequency modulation (FM) fuze interference identification system, the transmitted signal of a linear frequency modulation (LFM) fuze is represented as u. on (t)=U on cos(2πf0t+πμt 2 ), where U on f0 is the amplitude of the transmitted signal, μ is the carrier frequency of the transmitted signal of the linear frequency modulation fuze, t is the frequency modulation slope, and t refers to time.

[0021] After the fuze's transmitted signal reaches the target, it is reflected by the target to form a fuze echo signal, denoted as u. c (t)=U c cos[2πf0(t-τ)+πμ(t-τ) 2 ], where Uc is the amplitude of the echo signal, τ=2r / c is the transmission delay between the fuze and the target, r is the distance between the fuze and the target, and c is the speed of light;

[0022] After the fuze echo signal is interfered with by the jammer and external noise, the signal received by the receiver is expressed as u(t) = u c (t)+j(t)+n(t), where j(t) is the interference signal emitted by the jammer and n(t) is additive white Gaussian noise;

[0023] A training dataset is generated based on a radio frequency modulation fuze system. The generated training data is preprocessed using short-time Fourier transform to obtain the corresponding time-frequency domain signal and generate a time-frequency image.

[0024] According to a preferred embodiment of the present invention, the radio frequency modulation fuze interference signal identification model includes CNN and AM;

[0025] A CNN consists of one input layer, three convolutional layers, three pooling layers, one fully connected layer, and one output layer.

[0026] One-hot encoding is performed on the types of interference signals to generate corresponding labels. The labels and time-frequency images are then paired and input into the radio frequency modulation fuze interference signal identification model for supervised training.

[0027] The convolutional layer uses convolutional kernels to extract and map features from the input data (labels and time-frequency images), and the output is represented as follows:

[0028]

[0029] Among them, c i,j The convolution output is represented by X(im,jn), the input matrix of the convolutional layer is represented by W(m,n), the convolution kernel is represented by b1, the bias vector of the convolutional layer is represented by i and j, the row index and column index of the input data matrix are represented by i and j, respectively, and the kernel size is represented by m and n.

[0030] Set the activation function for all three convolutional layers to ReLU;

[0031] Pooling layers are used for downsampling and dimensionality reduction; a pooling layer is added after each convolutional layer.

[0032] After the last two pooling layers, a channel AM model excitation network (SENet) is added. The channel AM model focuses on the key information of the input, reducing attention to other information and filtering out irrelevant information. In the SENet excitation network, firstly, the spatial features of the channels are encoded into a global feature through the Squeeze operation; then, the relationship between the channels is learned through the Excitation operation to obtain the weights of different channels; finally, the weights of different channels are multiplied by the original feature through the Scale operation to obtain the final feature.

[0033] The feature vector is obtained by global average pooling, and then the feature vector obtained by global average pooling is input into the fully connected layer. Finally, the predicted output of the radio frequency modulation fuze interference signal identification model is obtained through the output layer.

[0034] A further preferred expression for the ReLU function is: f(x) = max(0,x).

[0035] A further preferred embodiment encodes the spatial features on the channel into a global feature using the Squeeze operation, represented as:

[0036]

[0037] Where H and D represent the height and width of the input data, respectively, g c This represents the features extracted through the convolution operation;

[0038] A further optimized approach involves learning the relationships between channels through excitation operations to obtain the weights of different channels, expressed as follows:

[0039] s=σ(W2 ReLU(W1z))

[0040] Where σ(.) represents the Sigmoid activation function, and W1 and W2 represent the weights of the first and second fully connected layers, respectively. More preferably, the output of the fully connected layer is expressed as:

[0041] o=f(W f X f +b2)

[0042] Among them, W f X represents the weight matrix of the fully connected layer. f b1 represents the input matrix of the fully connected layer, b2 represents the bias vector of the fully connected layer, and f(.) represents the activation function.

[0043] Further preferred, the activation function of the output layer is the Softmax function, with the expression:

[0044]

[0045] Where K is the number of categories, x r This represents the input to the r-th neuron.

[0046] According to a preferred embodiment of the present invention, the preprocessed training dataset is input into the radio frequency modulation fuze interference signal identification model for offline training to optimize the radio frequency modulation fuze interference signal identification network model, including:

[0047] The loss function is:

[0048]

[0049] Among them, y v Indicates the real label, p v This represents the predicted output of the radio frequency modulation fuze interference signal identification model;

[0050] The Adam optimizer was used to accelerate network convergence. The initial learning rate was set to 0.001, the batch size to 128, and the epoch to 50. After the parameters were set, the radio frequency modulation fuze interference signal identification model was trained under supervision to optimize the parameters and thus achieve the classification of interference signals.

[0051] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a method for identifying radio frequency modulation fuze interference signals based on a convolutional neural network and an attention mechanism.

[0052] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for identifying radio frequency modulation fuze interference signals based on a convolutional neural network and an attention mechanism.

[0053] The beneficial effects of this invention are as follows:

[0054] 1. Compared with traditional interference signal identification methods, this invention does not require manual feature extraction or a large amount of professional knowledge. Instead, it uses neural networks to adaptively extract features and perform interference identification, which greatly saves human resources.

[0055] 2. The radio linear frequency modulation fuze interference identification method based on CNN and AM proposed in this invention can achieve a network identification accuracy of over 98% when the interference-to-signal ratio is -16.0dB, realizing high identification accuracy under low interference-to-signal ratio conditions. Attached Figure Description

[0056] Figure 1 This is a schematic block diagram of the radio frequency modulation fuze interference identification system of the present invention.

[0057] Figure 2 This is a schematic diagram showing the overall recognition accuracy of six types of interference signals under different interference-to-signal ratios.

[0058] Figure 3 This is a schematic diagram of the network structure of a radio frequency modulation fuze interference signal identification model. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings and embodiments, but is not limited thereto.

[0060] Example 1

[0061] A method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks (CNN) and attention mechanisms (AM)

[0062] This method for identifying radio frequency modulation (FM) fuze interference signals is implemented through a radio frequency modulation (FM) fuze interference identification system, such as... Figure 1 As shown, the radio frequency modulation fuse jamming identification system includes a target detector, a jammer, and a signal processing module;

[0063] The target detector includes a transmitter and a receiver; the jammer generates several typical jamming signals; the signal processing module refers to the jamming identifier based on CNN and AM.

[0064] Meanwhile, it is assumed that the received noise is additive white Gaussian noise, and internal system effects are ignored, including:

[0065] The training dataset is generated through simulation, and the data is preprocessed.

[0066] The preprocessed training dataset is input into the radio frequency modulation fuze interference signal recognition model for offline training to optimize the radio frequency modulation fuze interference signal recognition network model; a test dataset is generated by simulation, and the recognition performance of the radio frequency modulation fuze interference signal recognition model after offline training is verified using the test dataset;

[0067] After the radio frequency modulation fuze interference signal to be identified is preprocessed, it is input into the verified radio frequency modulation fuze interference signal identification model for signal identification to obtain the type of interference signal.

[0068] Example 2

[0069] The difference between the radio frequency modulation fuze interference signal identification method based on convolutional neural network (CNN) and attention mechanism (AM) described in Example 1 is as follows:

[0070] The jammer generates six typical interference signals, including: sinusoidal amplitude modulation interference signal, sinusoidal frequency modulation interference signal, noise frequency modulation interference signal, noise amplitude modulation interference signal, linear frequency modulation sweep interference signal, and logarithmic frequency modulation sweep interference signal.

[0071] The training dataset is generated through simulation, and data preprocessing is performed, including:

[0072] In a radio frequency modulation (FM) fuze interference identification system, the transmitted signal of a linear frequency modulation (LFM) fuze is represented as u. on (t)=U on cos(2πf0t+πμt 2 ), where U on f0 is the amplitude of the transmitted signal, μ is the carrier frequency of the transmitted signal of the linear frequency modulation fuze, t is the frequency modulation slope, and t refers to time.

[0073] After the fuze's transmitted signal reaches the target, it is reflected by the target to form a fuze echo signal, denoted as u. c(t)=U c cos[2πf0(t-τ)+πμ(t-τ) 2 ], where Uc is the amplitude of the echo signal, τ=2r / c is the transmission delay between the fuze and the target, r is the distance between the fuze and the target, and c is the speed of light; through the parameters in the fuze echo signal, information such as the distance between the fuze and the target can be obtained.

[0074] After the fuze echo signal is interfered with by the jammer and external noise, the signal received by the receiver is expressed as u(t) = u c (t)+j(t)+n(t), where j(t) is the interference signal emitted by the jammer and n(t) is additive white Gaussian noise;

[0075] A training dataset was generated based on a radio frequency modulation (FM) fuze system. The training data in the dataset were generated from simulation data using MATLAB, with an interference-to-signal ratio (ISR) ranging from -10.0 dB to 10.0 dB and a step size of 2.0 dB. 5.5 × 10⁻⁶ ppm were generated for each type of interference signal. 3 A total of 3.3 × 10 samples were generated. 4 The dataset consists of 80% training samples and 20% validation samples. The generated training data is preprocessed using a short-time Fourier transform to obtain the corresponding time-frequency domain signal, and a time-frequency image is generated. The sampling frequency is set to 50MHz, and a Hamming window is used as the window function.

[0076] like Figure 3 As shown, the radio frequency modulation fuze interference signal identification model includes CNN and AM;

[0077] A CNN consists of one input layer, three convolutional layers, three pooling layers, one fully connected layer, and one output layer.

[0078] One-hot encoding is performed on the types of interference signals to generate corresponding labels. The labels and time-frequency images are then paired and input into the radio frequency modulation fuze interference signal identification model for supervised training.

[0079] The convolutional layer uses convolutional kernels to extract and map features from the input data (labels and time-frequency images), and the output is represented as follows:

[0080]

[0081] Among them, c i,j The convolution output is represented by X(im,jn), the input matrix of the convolutional layer is represented by W(m,n), the convolution kernel is represented by b1, the bias vector of the convolutional layer is represented by i and j, the row index and column index of the input data matrix are represented by i and j, respectively, and the kernel size is represented by m and n.

[0082] In the three convolutional layers, the kernel size of the first convolutional layer is 7×7, which increases the receptive field and better extracts the features of the input image; the kernel size of the second and third convolutional layers is 3×3, and the number of kernels in the three convolutional layers are 32, 16 and 64 respectively;

[0083] Set the activation function for all three convolutional layers to ReLU;

[0084] Pooling layers perform downsampling and dimensionality reduction, and a pooling layer is added after each convolutional layer; the pooling window of each pooling layer is 2×2, and the stride is 2.

[0085] After the last two pooling layers, a channel-based AM (Advanced Activation Model) excitation network (SENet) is added. This channel-based AM model focuses on key information in the input, reducing attention to other information and filtering out irrelevant details, thus improving the efficiency and accuracy of task processing. In the SENet excitation network, firstly, a Squeeze operation encodes the spatial features of each channel into a global feature; then, an Excitation operation learns the relationships between channels to obtain the weights of different channels; finally, a Scale operation multiplies the obtained weights of different channels with the original feature to obtain the final feature.

[0086] The feature vector is obtained by global average pooling, and then the feature vector obtained by global average pooling is input into the fully connected layer. Finally, the predicted output of the radio frequency modulation fuze interference signal identification model is obtained through the output layer.

[0087] The ReLU function is expressed as: f(x) = max(0,x).

[0088] The Squeeze operation encodes the spatial features of a channel into a global feature, represented as follows:

[0089]

[0090] Where H and D represent the height and width of the input data, respectively, g c This represents the features extracted through the convolution operation;

[0091] The relationships between channels are learned through the excitation operation, and the weights of different channels are obtained, represented as follows:

[0092] s=σ(W2 Re LU(W1z))

[0093] Where σ(.) represents the Sigmoid activation function, and W1 and W2 represent the weights of the first and second fully connected layers, respectively.

[0094] The output of the fully connected layer is represented as:

[0095] o=f(W f X f +b2)

[0096] Among them, W f X represents the weight matrix of the fully connected layer. f b1 represents the input matrix of the fully connected layer, b2 represents the bias vector of the fully connected layer, and f(.) represents the activation function.

[0097] The number of neurons in the fully connected layer was set to 128, and the activation function was ReLU. In order to avoid overfitting, a Dropout layer was added, which randomly removed some neurons in each iteration with a dropout rate of 0.5.

[0098] The output layer has 6 neurons, corresponding to the types of interference signals, and the activation function is the Softmax function, expressed as:

[0099]

[0100] Where K is the number of categories, x r This represents the input to the r-th neuron.

[0101] The preprocessed training dataset is input into the radio frequency modulation (FM) fuze jamming signal recognition model for offline training, optimizing the FM fuze jamming signal recognition network model, including:

[0102] The loss function is:

[0103]

[0104] Among them, y v Indicates the real label, p v This represents the predicted output of the radio frequency modulation fuze interference signal identification model;

[0105] The Adam optimizer was used to accelerate network convergence. The initial learning rate was set to 0.001, the batch size to 128, and the epoch to 50. After the parameters were set, the radio frequency modulation fuze interference signal identification model was trained under supervision to optimize the parameters and thus achieve the classification of interference signals.

[0106] A test dataset was generated through simulation, and this dataset was used to verify the recognition performance of the radio frequency modulation (FM) fuze interference signal recognition model after offline training. This included online testing after offline network training. To test the network's robustness, the interference-to-signal ratio (ISR) range of the test data was set to -20.0 dB to 10.0 dB, with a step size of 2.0 dB. 1.6 × 10⁻⁶ samples were generated for each signal class.3 A total of 9.6 × 10 samples were generated. 3 A number of test samples are used to form the test data. The test data under different interference-to-signal ratios are input into the trained network to obtain the predicted interference signal category, and then compared with the true label to obtain the network's recognition accuracy. Figure 2 This is a schematic diagram showing the overall recognition accuracy of six types of interference signals under different interference-to-signal ratios. The test results show that the radio linear frequency modulation fuze interference recognition method based on CNN and AM proposed in this invention can achieve a recognition accuracy of over 98% when the interference-to-signal ratio is -16.0dB.

[0107] Example 3

[0108] A computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the radio frequency modulation fuze interference signal identification method based on convolutional neural networks and attention mechanisms described in Embodiment 1 or 2.

[0109] Example 4

[0110] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the radio frequency modulation fuze interference signal identification method based on convolutional neural networks and attention mechanisms as described in Embodiment 1 or 2.

Claims

1. A method for identifying radio frequency modulation (FM) fuse interference signals based on convolutional neural networks and attention mechanisms, characterized in that, The method for identifying radio frequency modulation (FM) fuze interference signals is implemented through a radio frequency modulation (FM) fuze interference identification system, which includes a target detector, a jammer, and a signal processing module. The target detector includes a transmitter and a receiver; the jammer generates several typical jamming signals; The signal processing module refers to an interference detector based on CNN and AM; it includes: The training dataset is generated through simulation, and the data is preprocessed. The preprocessed training dataset is input into the radio frequency modulation fuze interference signal recognition model for offline training to optimize the radio frequency modulation fuze interference signal recognition network model; a test dataset is generated by simulation, and the recognition performance of the radio frequency modulation fuze interference signal recognition model after offline training is verified using the test dataset; After the radio frequency modulation fuze interference signal to be identified is preprocessed, it is input into the verified radio frequency modulation fuze interference signal identification model for signal identification to obtain the type of interference signal; The jammer generates six typical interference signals, including: sinusoidal amplitude modulation interference signal, sinusoidal frequency modulation interference signal, noise frequency modulation interference signal, noise amplitude modulation interference signal, linear frequency modulation sweep interference signal, and logarithmic frequency modulation sweep interference signal; Radio frequency modulation fuze interference signal identification models include CNN and AM; A CNN consists of one input layer, three convolutional layers, three pooling layers, one fully connected layer, and one output layer. One-hot encoding is performed on the types of interference signals to generate corresponding labels. The labels and time-frequency images are then paired and input into the radio frequency modulation fuze interference signal identification model for supervised training. The convolutional layer uses convolutional kernels to extract and map features from the input data (labels and time-frequency images), and the output is represented as follows: Among them, c i,j The convolution output is represented by X(im, jn), the input matrix of the convolutional layer is represented by W(m, n), the convolutional kernel is represented by b1, the bias vector of the convolutional layer is represented by i and j, the row index and column index of the input data matrix are represented by i and j, respectively, and the kernel size is represented by m and n. Set the activation function for all three convolutional layers to ReLU; Pooling layers are used for downsampling and dimensionality reduction; a pooling layer is added after each convolutional layer. After the last two pooling layers, a channel AM model is added to compress the activation network. The channel AM model focuses on the key information of the input, reducing attention to other information and filtering out irrelevant information. In the compressed activation network, firstly, the spatial features of the channels are encoded into a global feature through the Squeeze operation; then, the relationship between the channels is learned through the Excitation operation to obtain the weights of different channels; finally, the weights of different channels are multiplied by the original feature through the Scale operation to obtain the final feature. ; The feature vector is obtained by global average pooling, and then the feature vector obtained by global average pooling is input into the fully connected layer. The predicted output of the radio frequency modulation fuze interference signal identification model is then obtained through the output layer. The expression for the ReLU function is: ; The Squeeze operation encodes the spatial features of a channel into a global feature, represented as follows: Where H and D represent the height and width of the input data, respectively, g c This represents the features extracted through convolution operations.

2. The method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks and attention mechanisms according to claim 1, characterized in that, The training dataset is generated through simulation, and data preprocessing is performed, including: In a radio frequency modulation (FM) fuze interference identification system, the transmitted signal of a linear frequency modulation (LFM) fuze is represented as: , among which, U on f0 is the amplitude of the transmitted signal, f0 is the carrier frequency of the transmitted signal of the linear frequency modulated fuze, and μ is the frequency modulation slope. It refers to time; After the fuze's transmitted signal reaches the target, it is reflected by the target to form a fuze echo signal, which is represented as... , among which, U c The amplitude of the echo signal. , where is the transmission delay between the fuze and the target, r is the distance between the fuze and the target, and c is the speed of light; After the fuze echo signal is interfered with by the jammer and external noise, the signal received by the receiver is expressed as u(t) = u c (t)+j(t)+n(t), where j(t) is the interference signal emitted by the jammer and n(t) is additive white Gaussian noise; A training dataset is generated based on a radio frequency modulation fuze system. The generated training data is preprocessed using short-time Fourier transform to obtain the corresponding time-frequency domain signal and generate a time-frequency image.

3. The method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks and attention mechanisms according to claim 1, characterized in that, The relationships between channels are learned through the excitation operation, and the weights of different channels are obtained, represented as follows: in, This represents the Sigmoid activation function, and W1 and W2 represent the weights of the first and second fully connected layers, respectively.

4. The method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks and attention mechanisms according to claim 1, characterized in that, The output of the fully connected layer is represented as: in, X represents the weight matrix of the fully connected layer. f b1 represents the input matrix of the fully connected layer, b2 represents the bias vector of the fully connected layer, and f(.) represents the activation function.

5. The method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks and attention mechanisms according to claim 1, characterized in that, The activation function of the output layer is the Softmax function, expressed as: Where K is the number of categories, x r This represents the input to the r-th neuron.

6. A method for identifying radio frequency modulation fuze interference signals based on convolutional neural networks and attention mechanisms according to any one of claims 1-5, characterized in that, The preprocessed training dataset is input into the radio frequency modulation (FM) fuze jamming signal recognition model for offline training, optimizing the FM fuze jamming signal recognition network model, including: The loss function is: Among them, y v Indicates the real label, p v This represents the predicted output of the radio frequency modulation fuze interference signal identification model; The Adam optimizer was used to accelerate network convergence. The initial learning rate was set to 0.001, the batch size to 128, and the epoch to 50. After the parameters were set, the radio frequency modulation fuze interference signal identification model was trained under supervision to optimize the parameters and thus achieve the classification of interference signals.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the radio frequency modulation fuze interference signal identification method based on convolutional neural networks and attention mechanisms as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the radio frequency modulation fuze interference signal identification method based on convolutional neural networks and attention mechanisms as described in any one of claims 1-6.

Images