Radar jamming recognition method based on context-aware block feature fusion network

Abstract

This invention provides a radar interference identification method based on a feature fusion network of context-aware blocks, comprising: acquiring time-domain radar interference data to be classified; performing data conversion processing on the time-domain radar interference data to be classified to obtain time-frequency domain radar interference data to be classified; performing data block segmentation processing on the time-frequency domain radar interference data to be classified to obtain multiple radar time-frequency data blocks; inputting the radar time-frequency data blocks into a SAM encoder for processing to obtain radar coding information with position attributes; performing context feature concatenation processing on the radar coding information with corresponding position attributes to obtain context concatenation features; inputting the time-domain radar interference data to be classified into a time-domain feature extraction network for time-domain feature extraction to obtain time-domain radar features; performing feature fusion processing on the context concatenation features and time-domain radar features to obtain fused features; and performing classification processing on the fused features based on a feature classification fusion network to obtain radar interference classification results.

Description

Technical Field

[0001] This invention relates to the field of radar electronic countermeasures technology, and specifically to a radar interference identification method based on a feature fusion network of context-aware blocks. Background Technology

[0002] Radar systems are an indispensable and crucial component of modern electronic warfare. Through various functions such as target detection and tracking, threat warning, jamming and counter-jamming, electronic intelligence gathering, and navigation and control, they provide critical support for battlefield situational awareness, tactical decision-making, and combat operations. With the continuous development of electronic warfare, the electromagnetic warfare scenarios faced by radars are becoming increasingly harsh and complex, especially with various complex jamming signals that seriously threaten the decision-making capabilities of radar systems. Therefore, taking anti-jamming measures in radar electronic warfare is necessary. However, different jamming signals require different suppression methods; thus, identifying and classifying received jamming signals before implementing anti-jamming measures is of great significance.

[0003] Traditional interference classification methods primarily rely on signal processing techniques and expert experience, classifying interference signals by analyzing their spectral and temporal characteristics. However, with the increasing variety of interference signals and the growing complexity of the electromagnetic environment, traditional methods face challenges such as difficulty in feature extraction and limited generalization ability of classifiers. Furthermore, traditional methods often require preprocessing and feature extraction of large amounts of data, which not only increases computational complexity but also reduces the real-time performance and accuracy of classification. Therefore, researchers have begun exploring the use of deep learning techniques to improve the accuracy and efficiency of interference signal classification. Deep learning is a machine learning technique that simulates the neural networks of the human brain. By constructing multi-layered neural network models, it automatically extracts and learns features from data, thereby enabling the modeling and solving of complex problems. In the field of radar interference signal classification, the introduction of deep learning technology has provided new possibilities for improving classification accuracy and efficiency. Deep learning-based radar interference classification algorithms generally employ deep learning models such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Long Short-Term Memory Networks (LSTM). By training these network models on interference signal data, the models can automatically learn the feature representations of interference signals, thereby achieving the classification of different interference types. However, in real-world radar electronic warfare environments, jamming signals are often mixed with noise. While existing deep learning-based jamming classification methods can intelligently extract jamming features, in complex electromagnetic environments with low signal-to-noise ratios and multiple types of interference, the features of jamming signals are difficult to accurately extract and identify, making it difficult to maintain high classification accuracy. This not only reduces the effectiveness of jamming signal classification but may also lead to the adoption of incorrect anti-jamming measures. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention provides a radar interference identification method based on a feature fusion network of context-aware blocks.

[0005] The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] In a first aspect, the present invention provides a radar interference identification method based on a feature fusion network of context-aware blocks, comprising:

[0007] Acquire time-domain radar interference data to be classified;

[0008] The time-domain radar interference data to be classified is transformed to obtain the time-frequency domain radar interference data to be classified.

[0009] The radar interference data in the time and frequency domain to be classified is divided into multiple radar time and frequency data blocks X according to the preset data blocks. tf The size of the time-frequency domain radar interference data to be classified is 1024×1024, the size of the preset data block P is 16×16, and the size of multiple radar time-frequency data blocks is 64×64; X tf Represented as

[0010] Multiple radar time-frequency data blocks are input into the SAM encoder for encoding processing to obtain radar encoded information with position attributes; the size of the radar encoded information is 256×64×64.

[0011] The radar coding information corresponding to the location attributes is processed by concatenating context features to obtain context concatenated features;

[0012] The time-domain radar interference data to be classified is input into the time-domain feature extraction network of the interference signal to extract time-domain features and obtain time-domain radar features;

[0013] The context splicing features and the temporal radar features are fused to obtain the fused features.

[0014] The fused features are classified and processed using a feature classification fusion network to obtain radar interference classification results.

[0015] Optionally, the time-domain radar interference data to be classified is subjected to data transformation processing to obtain the time-frequency domain radar interference data to be classified, including:

[0016] Short-time Fourier transform is performed on the time-domain radar interference data to be classified to obtain the time-frequency domain radar interference data to be classified.

[0017] Optionally, the radar interference data in the time-frequency domain to be classified is divided into data blocks to obtain multiple radar time-frequency data blocks, including:

[0018] The radar interference data in the time and frequency domain to be classified is traversed and segmented according to the preset data blocks to obtain multiple radar time and frequency data blocks.

[0019] Optionally, the interference signal time-domain feature extraction network includes: multiple interference signal time-domain feature extraction subnetworks;

[0020] Each interference signal temporal feature extraction subnetwork is configured with a first batch normalization layer, a first convolutional layer, a first linear rectified layer, and a first dropout layer connected in sequence.

[0021] The kernel size of the first convolutional layer is 3×3, and the stride is 1×1.

[0022] The dropout rate for the first Dropout layer is set to 0.5.

[0023] Optionally, when the interference signal time-domain feature extraction network contains two interference signal time-domain feature extraction sub-networks, the processing result of the interference signal time-domain feature extraction network is expressed as follows:

[0024] G t =σ(W 2 ·BN(σ(W 1 ·BN(X t ))));

[0025] Among them, G t The result of the interference signal temporal feature extraction network is represented by σ(·), where σ(·) is the activation function, BN(·) is the feature normalization function, and W is the feature normalization function. 1 W represents the convolutional layer weights of subnetwork 1 for extracting temporal features of interference signals. 2 X represents the convolutional layer weights of subnetwork 2 for extracting temporal features of interference signals. t This represents the time-domain radar interference data to be classified.

[0026] Optionally, the feature classification fusion network includes a fully connected layer and multiple pre-defined structural layers;

[0027] The pre-defined structure includes: a second batch normalization layer, a second convolutional layer, and a second Dropout layer;

[0028] The kernel size of the second convolutional layer is 3×3, and the stride is 1×1.

[0029] The dropout rate for the second Dropout layer is set to 0.5.

[0030] Optionally, when the feature classification fusion network includes three preset structural layers, the output of the feature classification fusion network is represented as follows:

[0031] G out =FC[σ(W5 ·σ(W 4 ·(σ(W 3 ·G))))];

[0032] Among them, G out The output of the feature classification fusion network is represented by FC(·), where FC(·) represents a fully connected layer, σ(·) is the activation function, G represents the fused features, and W represents the fused features. 3 W represents the convolutional layer weights of the preset structure layer 1. 4 W represents the convolutional layer weights of the preset structure layer 2. 5 This represents the convolutional layer weights of the preset structure layer 3.

[0033] Optionally, the training process of the interference signal temporal feature extraction network and the feature classification fusion network includes:

[0034] Acquire time-domain radar interference samples;

[0035] The time-domain radar interference samples are converted into time-frequency domain radar interference samples.

[0036] The time-frequency domain radar interference samples are segmented into data blocks according to the preset data blocks to obtain multiple radar time-frequency sample blocks.

[0037] Multiple radar time-frequency sample blocks are input into the SAM encoder for encoding processing to obtain radar coded sample information with position attributes;

[0038] The radar-coded sample information corresponding to the location attributes is concatenated with context features to obtain multiple context fusion sample features;

[0039] The time-domain radar interference sample is input into the initial interference signal time-domain feature extraction network to extract time-domain features, thus obtaining the time-domain radar sample features;

[0040] Feature fusion processing is performed on multiple context fusion sample features and temporal radar sample features to obtain fused sample features;

[0041] The initial feature classification fusion network is used to classify the features of the fused samples to obtain the classification results of the radar interference samples.

[0042] Based on the cross-entropy loss function and the classification results of radar interference samples, it is determined whether the initial interference signal temporal feature extraction network and the initial feature classification fusion network have reached the preset convergence condition.

[0043] The initial interference signal temporal feature extraction network that has reached the convergence condition is used as the interference signal temporal feature extraction network, and the initial feature classification and fusion network that has reached the convergence condition is used as the feature classification and fusion network.

[0044] In a second aspect, the present invention provides a radar interference identification device based on a feature fusion network of context-aware blocks. The radar interference identification device based on a feature fusion network of context-aware blocks includes: an acquisition unit, a conversion unit, a segmentation unit, an encoding unit, a feature splicing unit, a feature extraction unit, a fusion unit, and a classification unit.

[0045] The acquisition unit is used to: acquire time-domain radar interference data to be classified;

[0046] The conversion unit is used to: perform data conversion processing on the time-domain radar interference data to be classified, and obtain the time-frequency domain radar interference data to be classified;

[0047] The segmentation unit is used to: segment the radar interference data to be classified in the time-frequency domain according to the preset data blocks to obtain multiple radar time-frequency data blocks;

[0048] The encoding unit is used to: input multiple radar time-frequency data blocks into the SAM encoder for encoding processing to obtain radar coded information with position attributes;

[0049] The feature splicing unit is used to: perform context feature splicing processing on the radar coding information corresponding to the location attribute to obtain context splicing features;

[0050] The feature extraction unit is used to: input the time-domain radar interference data to be classified into the interference signal time-domain feature extraction network to extract time-domain features and obtain time-domain radar features;

[0051] The fusion unit is used to perform feature fusion processing on context splicing features and temporal radar features to obtain fused features;

[0052] The classification unit is used to classify the fused features based on the feature classification fusion network to obtain the radar interference classification result.

[0053] Thirdly, the present invention provides a radar interference identification device based on a feature fusion network of context-aware blocks, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the radar interference identification device based on a feature fusion network of context-aware blocks is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the radar interference identification method based on a feature fusion network of context-aware blocks as described in the first aspect above.

[0054] This invention provides a radar interference identification method based on a feature fusion network of context-aware blocks. The method includes: acquiring time-domain radar interference data to be classified; performing data transformation processing on the time-domain radar interference data to be classified to obtain time-frequency domain radar interference data to be classified; performing data block segmentation processing on the time-frequency domain radar interference data to be classified to obtain multiple radar time-frequency data blocks; inputting the multiple radar time-frequency data blocks into a SAM encoder for encoding processing to obtain radar encoding information with location attributes; performing context feature concatenation processing on the radar encoding information with corresponding location attributes to obtain context concatenation features; inputting the time-domain radar interference data to be classified into a time-domain feature extraction network for time-domain feature extraction to obtain time-domain radar features; performing feature fusion processing on the context concatenation features and time-domain radar features to obtain fused features; and performing classification processing on the fused features based on a feature classification fusion network to obtain radar interference classification results. In this invention, by segmenting the time-frequency domain radar interference data into blocks and performing SAM encoding and context splicing on multiple radar time-frequency data blocks, robust radar coding information can be effectively extracted from the global data blocks in the time-frequency domain. Furthermore, classification is performed using fused features obtained by fusing context splicing features and time-domain radar features. Since the radar interference features represented by the fused features are more accurate, the accuracy of the identification and classification method in complex electromagnetic interference environments is greatly improved.

[0055] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0056] Figure 1 A flowchart illustrating a radar interference identification method based on a feature fusion network of context-aware blocks, provided in an embodiment of the present invention;

[0057] Figure 2 This is an overall architecture diagram of a radar interference identification method based on a feature fusion network of context-aware blocks, provided in an embodiment of the present invention.

[0058] Figure 3 The training loss variation curve provided in the embodiments of the present invention;

[0059] Figure 4 Accuracy curves provided for embodiments of the present invention;

[0060] Figure 5 A simulation flowchart provided for embodiments of the present invention;

[0061] Figure 6 A schematic diagram of the structure of a radar interference identification device based on a feature fusion network of context-aware blocks provided in an embodiment of the present invention;

[0062] Figure 7 This is a schematic diagram of the structure of a radar interference identification device based on a feature fusion network of context-aware blocks, provided in an embodiment of the present invention. Detailed Implementation

[0063] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0064] To improve the accuracy of interference classification in complex electromagnetic interference environments, this invention provides a radar interference identification method based on a feature fusion network of context-aware blocks. Figure 1 This is a flowchart illustrating a radar interference identification method based on a feature fusion network of context-aware blocks, as provided in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0065] S101. Acquire the time-domain radar interference data to be classified.

[0066] S102. Perform data conversion processing on the time-domain radar interference data to be classified to obtain the time-frequency domain radar interference data to be classified.

[0067] Optionally, S102 may specifically include:

[0068] Short-time Fourier transform is performed on the time-domain radar interference data to be classified to obtain the time-frequency domain radar interference data to be classified.

[0069] S103. The radar interference data to be classified in the time and frequency domain is segmented into multiple radar time and frequency data blocks X according to the preset data blocks. tf .

[0070] The radar interference data to be classified in the time-frequency domain is 1024×1024 in size, the preset data block P is 16×16 in size, and the multiple radar time-frequency data blocks are 64×64 in size; X tf Represented as

[0071]

[0072] Optionally, S103 may specifically include:

[0073] The radar interference data in the time-frequency domain to be classified is traversed and segmented according to the preset data blocks to obtain multiple radar time-frequency data blocks.

[0074] S104. Input multiple radar time-frequency data blocks into the SAM encoder for encoding processing to obtain radar coded information with position attributes.

[0075] The size of the radar coded information is 256×64×64.

[0076] S105. Perform context feature concatenation processing on the radar coding information corresponding to the location attributes to obtain context concatenation features.

[0077] It should be noted that in this embodiment, a global feature extraction strategy is adopted. By segmenting the data, the receptive field is expanded to all radar time-frequency data blocks and contextual feature splicing is performed, thereby realizing interaction with global information. By using an attention mechanism to perform weighted fusion of radar time-frequency data block information from the global scope, the accuracy of time-frequency map features can be effectively enhanced.

[0078] The specific process is as follows: ① First, when the radar interference data X in the time-frequency domain to be classified is... tf The dimensions are 1024×1024×3, and X... tf The data is divided into multiple radar time-frequency data blocks. Each radar time-frequency data block P can be 16×16×3 in size, resulting in a total of 64×64 radar time-frequency data blocks. If one radar time-frequency data block is represented as P... * The cropped data can then be represented as follows:

[0079]

[0080] ②Then, when it is P * During feature extraction, all 4096 data blocks were input into a SAM encoder with fixed parameters, and the SAM encoder network parameters were selected from the "vit_b" parameter file.

[0081] Output features after processing by the SAM encoder Its dimensions are 256×64×64. Features Each element in the array corresponds to a radar time-frequency data block at its respective location.

[0082] S106. Input the time-domain radar interference data to be classified into the interference signal time-domain feature extraction network to extract time-domain features and obtain time-domain radar features.

[0083] Optionally, the interference signal time-domain feature extraction network includes: multiple interference signal time-domain feature extraction subnetworks;

[0084] Each interference signal temporal feature extraction subnetwork is configured with a first batch normalization layer, a first convolutional layer, a first linear rectified layer, and a first dropout layer connected in sequence.

[0085] The kernel size of the first convolutional layer is 3×3, and the stride is 1×1; the dropout rate of the first Dropout layer is set to 0.5.

[0086] Specifically, the parameters of the interference signal time-domain feature extraction network are as follows:

[0087] Table 1. Parameters of the interference signal time-domain feature extraction network

[0088]

[0089] Optionally, when the interference signal time-domain feature extraction network contains two interference signal time-domain feature extraction sub-networks, the processing result of the interference signal time-domain feature extraction network is expressed as follows:

[0090] G t =σ(W 2 ·BN(σ(W 1 ·BN(X t ))));

[0091] Among them, G t The result of the interference signal temporal feature extraction network is represented by σ(·), where σ(·) is the activation function, BN(·) is the feature normalization function, and W is the feature normalization function. 1 W represents the convolutional layer weights of subnetwork 1 for extracting temporal features of interference signals. 2 X represents the convolutional layer weights of subnetwork 2 for extracting temporal features of interference signals. t This represents the time-domain radar interference data to be classified.

[0092] S107. Perform feature fusion processing on the context splicing features and the time-domain radar features to obtain fused features.

[0093] This invention provides a radar interference identification method based on a feature fusion network of context-aware blocks. The method includes: acquiring time-domain radar interference data to be classified; performing data conversion processing on the time-domain radar interference data to be classified to obtain time-frequency domain radar interference data to be classified; performing data block segmentation processing on the time-frequency domain radar interference data to be classified to obtain multiple radar time-frequency data blocks; inputting the multiple radar time-frequency data blocks into a SAM encoder for encoding processing to obtain radar encoding information with location attributes; performing context feature concatenation processing on the radar encoding information with corresponding location attributes to obtain context concatenation features; inputting the time-domain radar interference data to be classified into a time-domain interference signal feature extraction network for time-domain feature extraction to obtain time-domain radar features; performing feature fusion processing on the context concatenation features and time-domain radar features to obtain fused features; and performing classification processing on the fused features based on a feature classification fusion network to obtain radar interference classification results. In this embodiment of the invention, by segmenting the time-frequency domain radar interference data into blocks and performing SAM encoding and context splicing on multiple radar time-frequency data blocks, robust radar coding information can be effectively extracted from the global data blocks in the time and frequency domain. Furthermore, classification is performed using fused features obtained by fusing context splicing features and time-domain radar features. Since the radar interference features represented by the fused features are more accurate, the accuracy of the identification and classification method in complex electromagnetic interference environments is greatly improved.

[0094] Figure 2 This is an overall architecture diagram of a radar interference identification method based on a context-aware block feature fusion network provided in an embodiment of the present invention, as shown below. Figure 2 The radar interference data to be classified in the time domain is first subjected to a short-time Fourier transform. The radar interference data to be classified in the time-frequency domain, which includes the real part, imaginary part, and modulus, is divided into data blocks and then input into a global context-aware block feature fusion sub-network (SAM encoder performs encoding processing, and then the encoding processing result is processed by context-based feature splicing according to position attributes) to obtain context splicing features. The radar interference data to be classified in the time domain is then processed by the interference signal time-domain feature extraction network (the interference signal time-domain feature extraction network contains multiple interference signal time-domain feature extraction sub-networks, which are represented as CBR Blocks) to obtain time-domain radar features. The fused features obtained by fusing the time-domain radar features and the context splicing features are input into the feature fusion classification network for classification and recognition to obtain the classification result.

[0095] Optionally, in this embodiment of the invention, a feature splicing method is used to perform feature fusion processing on context splicing features and temporal radar features.

[0096] Optionally, the feature classification fusion network includes a fully connected layer and multiple preset structure layers; the preset structure layers include: a second batch normalization layer, a second convolutional layer, and a second dropout layer; the kernel size of the second convolutional layer is 3×3, and the stride is 1×1; the dropout rate of the second dropout layer is set to 0.5.

[0097] Specifically, the parameters of the feature classification fusion network are as follows:

[0098] Table 2 Parameters of the Feature Classification Fusion Network

[0099]

[0100] Optionally, when the feature classification fusion network includes three preset structural layers, the output of the feature classification fusion network is represented as follows:

[0101] G out =FC[σ(W 5 ·σ(W 4 ·(σ(W 3 ·G))))];

[0102] Among them, G out The output of the feature classification fusion network is represented by FC(·), where FC(·) represents a fully connected layer, σ(·) is the activation function, G represents the fused features, and W represents the fused features. 3 W represents the convolutional layer weights of the preset structure layer 1. 4 W represents the convolutional layer weights of the preset structure layer 2. 5 This represents the convolutional layer weights of the preset structure layer 3.

[0103] S108. Based on the feature classification fusion network, the fused features are classified to obtain the radar interference classification result.

[0104] Optionally, the training process of the interference signal temporal feature extraction network and the feature classification fusion network includes:

[0105] Acquire time-domain radar interference samples;

[0106] The time-domain radar interference samples are converted into time-frequency domain radar interference samples.

[0107] The time-frequency domain radar interference samples are segmented into data blocks according to the preset data blocks to obtain multiple radar time-frequency sample blocks.

[0108] Multiple radar time-frequency sample blocks are input into the SAM encoder for encoding processing to obtain radar coded sample information with position attributes;

[0109] The radar-coded sample information corresponding to the location attributes is subjected to context feature concatenation processing to obtain multiple context fusion sample features;

[0110] The time-domain radar interference sample is input into the initial interference signal time-domain feature extraction network to extract time-domain features, thus obtaining the time-domain radar sample features;

[0111] Feature fusion processing is performed on multiple context fusion sample features and temporal radar sample features to obtain fused sample features;

[0112] The initial feature classification fusion network is used to classify the features of the fused samples to obtain the classification results of the radar interference samples.

[0113] Based on the cross-entropy loss function and the classification results of radar interference samples, it is determined whether the initial interference signal temporal feature extraction network and the initial feature classification fusion network have reached the preset convergence condition.

[0114] The initial interference signal temporal feature extraction network that has reached the convergence condition is used as the interference signal temporal feature extraction network, and the initial feature classification and fusion network that has reached the convergence condition is used as the feature classification and fusion network.

[0115] Optionally, in this embodiment of the invention, the cross-entropy loss function L is specifically expressed as follows:

[0116]

[0117] Among them, O c (y i ) indicates label y i one-hot encoding format, Indicates sample x i The predicted probability of belonging to the c-th category, where C represents the total number of categories.

[0118] Invention Effect Analysis

[0119] The radar interference identification method based on the SAM encoder algorithm proposed in this invention demonstrates a significant performance improvement in the classification of radar interference signals. Specific effects are as follows:

[0120] (1) Significantly Improved Classification Accuracy: A feature fusion method based on global context radar time-frequency data patches. This algorithm can extract features of each data patch globally and use an attention mechanism to weightedly fuse global information, thereby enhancing the accuracy of time-frequency map features. Experimental results show that the algorithm can achieve an overall classification accuracy of 98.69% on datasets with a noise-to-interference ratio of -10 to 10 dB. The classification accuracy of each category is [1, 0.98312236 0.94581281, 1, 0.98253275, 1, 0.97727273, 1, 0.99029126]. The corresponding confusion matrix is ​​shown below:

[0121]

[0122] (2) Enhanced classification robustness: The method of this invention makes full use of the feature information in the time domain and frequency domain. By integrating the feature extraction methods of one-dimensional time domain and two-dimensional frequency domain, the stability and robustness of the classification results are ensured, with a KAPPA coefficient of 98.52%. Under different electromagnetic interference environments, it can effectively cope with various complex interference signals and maintain high classification performance.

[0123] In summary, the radar jamming identification method based on the SAM encoder algorithm proposed in this invention effectively solves the problems of classification accuracy and robustness of existing classification algorithms under complex electromagnetic environments and low signal-to-noise ratio conditions through innovative feature extraction and fusion methods. This significantly improves the effectiveness of radar jamming classification and provides strong support for the application of radar systems in modern electronic warfare. The training loss variation curve of the method provided in the embodiments of this invention is shown below. Figure 3 As shown in Figure 4, the training accuracy variation curve is as follows. Specifically, Figure 3 Figure (a) shows the loss curve for training iterations from 0 to 40k. Figure 3 Figure (b) shows the loss curve for training iterations of 0-5k. Figure 3 Figure (c) shows the loss curve for training iterations from 0 to 300+. Figure 3 Overall, it can be seen that the method of the present invention can achieve rapid convergence under various training iterations. Figure 4 The accuracy curves show that the method of the present invention has excellent recognition accuracy.

[0124] The specific simulation process also provided in this embodiment of the invention is as follows:

[0125] (1) Simulation conditions

[0126] In this embodiment of the invention, the feature fusion algorithm based on global context-aware blocks (SAM encoder processing part) is implemented in Python using the PyTorch framework. The initial learning rate is set to 0.001, and the cross-entropy loss function and Adam optimizer are employed. Training is conducted for 50 epochs. The main hardware specifications for implementing the method of this invention include: 1) CPU: Xeon Gold8163; 2) GPU: GTX-4090; 3) RAM: 128G. The input patch size is set to 16×16.

[0127] (2) Simulation process

[0128] Step 1: Obtain the raw time-domain data of radar interference echoes.

[0129] The specific implementation for obtaining the raw time-domain data of radar interference echoes is as follows:

[0130] First, an echo signal model is established, consisting of interference signals and Gaussian white noise. Specifically, it can be represented as:

[0131] s(t) = j(t) + n(t)

[0132] Where t is the time series, s(t) is the received echo signal, j(t) is the time-domain radar interference signal, and n(t) is Gaussian white noise.

[0133] Step 2 transforms the time-domain signal obtained in Step 1 to the time-frequency domain.

[0134] First, select an appropriate window function w(t) (such as Hanning window, Hamming window, rectangular window, etc.) and determine the length N of the window function; then, move the window function w(t) sequentially, segmenting the signal by selecting a suitable step size Δt. In this embodiment of the invention, the window function length is 128 and the step size is 118.

[0135] Finally, a Fourier transform is performed on each windowed signal segment to calculate the spectrum of each segment. The resulting spectra are then arranged along the time axis to obtain the time-frequency domain data.

[0136] Step 3: Design the interference dataset

[0137] Obtaining realistic radar active jamming signals is difficult. Therefore, this invention simulates and generates nine types of active radar jamming signals, and adds Gaussian white noise to the jamming signals to make the simulated signals closer to the real signals. The nine types of active radar jamming signals are: noise amplitude modulation jamming + dense false target jamming, intermittent sampling and forwarding jamming, intermittent sampling and forwarding jamming + dense false target jamming, agile noise jamming, noise amplitude modulation jamming + intermittent sampling and forwarding jamming, noise amplitude modulation jamming, noise frequency modulation jamming, dense false target jamming, and frequency sweep jamming.

[0138] The simulation parameters for various interference signals are as follows: center frequency Fc = 20MHz, bandwidth B = 50MHz, sampling frequency Fs = 100MHz, and duration PRT = 20us. The dense false target interference signal randomly generates 3-6 false targets with a delay time of 1-3us; the intermittent sampling and forwarding interference signal is randomly sampled 1-2 times and forwarded 1-2 times.

[0139] The additional noise of the interfering signal is Gaussian white noise with an interference noise ratio (IRR) of -10 to 10 dB, uniformly distributed across 11 INR ratios. For each INR ratio, 100 random samples are generated for each interfering signal, resulting in a dataset of 9900 samples (9*200*11). 70% of these samples are used as the training set, 20% as the validation set, and 10% as the test set. The specific division of the dataset is shown in Table 3 below.

[0140] Table 3. Detailed partitioning of the training dataset

[0141]

[0142]

[0143] Based on the network construction method disclosed in the above embodiments, a SAM encoder, an interference signal temporal feature extraction network, and a feature classification fusion network are built. The specific simulation flowchart is shown below. Figure 5 As shown, firstly, time-domain radar interference data is acquired, then the time-domain radar interference data is transformed to time-frequency threshold using the STEF algorithm. An interference dataset is designed based on the time-domain data and the time-frequency threshold data. Then, the model of this invention is built, and the model of this invention is trained based on the designed interference dataset to obtain a trained recognition network. Finally, the trained recognition network is used for classification, recognition, and simulation.

[0144] The method provided in this embodiment of the invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc., and this embodiment of the invention does not limit the application to such devices.

[0145] Based on the same inventive concept, embodiments of the present invention also provide a radar interference identification device based on a feature fusion network of context-aware blocks. Figure 6 This is a schematic diagram of a radar interference identification device based on a feature fusion network of context-aware blocks, provided as an embodiment of the present invention. Figure 6 As shown, it includes: an acquisition unit 601, a conversion unit 602, a segmentation unit 603, an encoding unit 604, a feature splicing unit 605, a feature extraction unit 606, a fusion unit 607, and a classification unit 608; the acquisition unit 601 is used to: acquire the time-domain radar interference data to be classified;

[0146] The conversion unit 602 is used to: perform data conversion processing on the time-domain radar interference data to be classified, and obtain the time-frequency domain radar interference data to be classified;

[0147] The cutting unit 603 is used to: perform data block cutting processing on the radar interference data to be classified in the time and frequency domain according to the preset data blocks to obtain multiple radar time and frequency data blocks;

[0148] The encoding unit 604 is used to: input multiple radar time-frequency data blocks into the SAM encoder for encoding processing to obtain radar encoded information with position attributes;

[0149] The feature splicing unit 605 is used to: perform context feature splicing processing on the radar coding information corresponding to the location attribute to obtain context splicing features;

[0150] The feature extraction unit 606 is used to: input the time-domain radar interference data to be classified into the interference signal time-domain feature extraction network to extract time-domain features and obtain time-domain radar features;

[0151] The fusion unit 607 is used to: perform feature fusion processing on context splicing features and temporal radar features to obtain fused features;

[0152] The classification unit 608 is used to classify the fused features based on the feature classification fusion network to obtain the radar interference classification result.

[0153] Figure 7 This invention provides a schematic diagram of a radar interference identification device based on a context-aware block feature fusion network, comprising a processor 710, a storage medium 720, and a bus 730. The storage medium 720 stores machine-readable instructions executable by the processor 710. When the radar interference identification device based on the context-aware block feature fusion network is running, the processor 710 communicates with the storage medium 720 via the bus 730, and the processor 710 executes the machine-readable instructions to perform the steps of the above-described method embodiment. Specific implementation methods and technical effects are similar and will not be described in detail here.

[0154] The storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the storage medium may also be at least one storage device located remotely from the aforementioned processor.

[0155] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0156] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0157] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0158] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0159] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A radar interference identification method based on a feature fusion network of context-aware blocks, characterized in that, include: Acquire time-domain radar interference data to be classified; The time-domain radar interference data to be classified is subjected to data conversion processing to obtain the time-frequency domain radar interference data to be classified. The radar interference data to be classified in the time-frequency domain is segmented into multiple radar time-frequency data blocks X according to a preset data block structure. tf The size of the radar interference data to be classified in the time-frequency domain is 1024×1024, the size of the preset data block P is 16×16, and the size of the multiple radar time-frequency data blocks is 64×64. tf Represented as The multiple radar time-frequency data blocks are input into the SAM encoder for encoding processing to obtain radar encoded information with position attributes; the size of the radar encoded information is 256×64×64. The radar coding information corresponding to the location attribute is subjected to context feature concatenation processing to obtain context concatenation features; The time-domain radar interference data to be classified is input into the time-domain feature extraction network of the interference signal to extract time-domain features and obtain time-domain radar features. The context splicing features and the temporal radar features are fused to obtain fused features; The fused features are classified using a feature classification fusion network to obtain radar interference classification results.

2. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 1, characterized in that, The step of performing data conversion processing on the time-domain radar interference data to be classified to obtain time-frequency domain radar interference data to be classified includes: The time-domain radar interference data to be classified is processed by short-time Fourier transform to obtain the time-frequency domain radar interference data to be classified.

3. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 1, characterized in that, The process involves dividing the radar interference data to be classified into data blocks according to a preset data block structure to obtain multiple radar time-frequency data blocks, including: The radar interference data to be classified in the time and frequency domain is traversed and segmented according to the preset data blocks to obtain the multiple radar time and frequency data blocks.

4. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 1, characterized in that, The interference signal time-domain feature extraction network includes: multiple interference signal time-domain feature extraction sub-networks; Each of the interference signal temporal feature extraction subnetworks is provided with a first batch normalization layer, a first convolutional layer, a first linear rectified layer, and a first dropout layer connected in sequence. The kernel size of the first convolutional layer is 3×3, and the stride is 1×1; The drop rate of the first Dropout layer is set to 0.

5.

5. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 4, characterized in that, When the interference signal time-domain feature extraction network contains two interference signal time-domain feature extraction sub-networks, the processing result of the interference signal time-domain feature extraction network is expressed as follows: G t =σ(W 2 ·BN(σ(W 1 ·BN(X t )))); Among them, G t The result of the interference signal temporal feature extraction network is represented by σ(·), where σ(·) is the activation function, BN(·) is the feature normalization function, and W is the feature normalization function. 1 W represents the convolutional layer weights of subnetwork 1 for extracting temporal features of interference signals. 2 X represents the convolutional layer weights of subnetwork 2 for extracting temporal features of interference signals. t This represents the time-domain radar interference data to be classified.

6. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 1, characterized in that, The feature classification fusion network includes a fully connected layer and multiple pre-defined structural layers; The preset structure layer includes: a second batch normalization layer, a second convolutional layer, and a second Dropout layer; The kernel size of the second convolutional layer is 3×3, and the stride is 1×1; The drop rate of the second Dropout layer is set to 0.

5.

7. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 6, characterized in that, When the feature classification fusion network includes three preset structural layers, the output result of the feature classification fusion network is expressed as follows: G out =FC[σ(W 5 ·σ(W 4 ·(σ(W 3 ·G))))]; Among them, G out The output of the feature classification fusion network is represented by FC(·), where FC(·) represents a fully connected layer, σ(·) is the activation function, G represents the fused features, and W represents the fused features. 3 W represents the convolutional layer weights of the preset structure layer 1. 4 W represents the convolutional layer weights of the preset structure layer 2. 5 This represents the convolutional layer weights of the preset structure layer 3.

8. The radar interference identification method based on a feature fusion network of context-aware blocks according to claim 1, characterized in that, The training process of the interference signal temporal feature extraction network and the feature classification and fusion network includes: Acquire time-domain radar interference samples; The time-domain radar interference sample is processed by data conversion to obtain the time-frequency domain radar interference sample. The time-frequency domain radar interference samples are segmented into data blocks according to preset data blocks to obtain multiple radar time-frequency sample blocks. The multiple radar time-frequency sample blocks are input into the SAM encoder for encoding processing to obtain radar coded sample information with position attributes. The radar-coded sample information corresponding to the location attributes is subjected to context feature concatenation processing to obtain multiple context fusion sample features; The time-domain radar interference sample is input into the initial interference signal time-domain feature extraction network to extract time-domain features, thus obtaining the time-domain radar sample features; The multiple context fusion sample features and the temporal radar sample features are subjected to feature fusion processing to obtain fused sample features; The fused sample features are classified based on the initial feature classification fusion network to obtain the radar interference sample classification result; Based on the cross-entropy loss function and the classification results of the radar interference samples, it is determined whether the initial interference signal temporal feature extraction network and the initial feature classification fusion network have reached the preset convergence condition. The initial interference signal temporal feature extraction network that has reached the convergence condition is used as the interference signal temporal feature extraction network, and the initial feature classification and fusion network that has reached the convergence condition is used as the feature classification and fusion network.

9. A radar interference identification device based on a feature fusion network of context-aware blocks, characterized in that, The radar interference identification device based on the feature fusion network of context-aware blocks includes: an acquisition unit, a conversion unit, a segmentation unit, an encoding unit, a feature splicing unit, a feature extraction unit, a fusion unit, and a classification unit; The acquisition unit is used to: acquire time-domain radar interference data to be classified; The conversion unit is used to: perform data conversion processing on the time-domain radar interference data to be classified to obtain time-frequency domain radar interference data to be classified. The cutting unit is used to: perform data block cutting processing on the radar interference data to be classified in the time-frequency domain according to the preset data blocks to obtain multiple radar time-frequency data blocks; The encoding unit is used to: input the plurality of radar time-frequency data blocks into the SAM encoder for encoding processing to obtain radar encoded information with position attributes; The feature splicing unit is used to: perform context feature splicing processing on the radar coding information corresponding to the location attribute to obtain context splicing features; The feature extraction unit is used to: input the time-domain radar interference data to be classified into the interference signal time-domain feature extraction network to extract time-domain features and obtain time-domain radar features; The fusion unit is used to: perform feature fusion processing on the context splicing features and the temporal radar features to obtain fused features; The classification unit is used to: classify the fused features based on the feature classification fusion network to obtain radar interference classification results.

10. A radar interference identification device based on a feature fusion network of context-aware blocks, characterized in that, include: The device includes a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the radar interference identification device based on the context-aware block feature fusion network is running, the processor communicates with the storage medium via the bus. The processor executes the machine-readable instructions to perform the steps of the radar interference identification method based on the context-aware block feature fusion network as described in any one of claims 1-8.

Images