A Radar Active Composite Jamming Recognition Method Based on an Efficient Multimodal Caching Model

Abstract

The present invention discloses a method for identifying radar active composite interference based on an efficient multi-modal caching model, belonging to the field of radar active interference identification, including: obtaining simulated radar active interference time-domain data, splitting the simulated radar active interference time-domain data into a training set and a test set, and dividing the training set into a training support set and a training query set; constructing an efficient multi-modal caching model based on a visual caching module, a text caching module, and a residual connection module; training the efficient multi-modal caching model based on the training support set and the training query set, testing the trained model through the test set to obtain an interference identification model; and identifying radar active composite interference based on the interference identification model. The present invention can efficiently utilize and deeply mine limited interference samples, thereby improving the accuracy of radar active composite interference identification.

Description

Technical Field

[0001] The present invention belongs to the field of radar active interference recognition, and particularly relates to a method for radar active composite interference recognition based on an efficient multi-modal caching model. Background Art

[0002] An anti-ship missile refers to a missile launched from a naval ship, ashore, or an aircraft for attacking surface warships. In modern warfare, with the rapid development of shipborne jamming equipment, the survival and operation of anti-ship missiles are facing increasingly severe threats. Therefore, it has become particularly urgent to improve the anti-jamming ability of the seeker radar of anti-ship missiles. As the primary link in radar anti-jamming technology, interference type recognition has received extensive attention and research in recent years.

[0003] In order to accurately distinguish radar interference types, a variety of traditional interference recognition methods have emerged. These methods can be roughly classified into two categories: likelihood-based methods and multi-domain feature extraction methods. Likelihood-based methods focus on deeply modeling echo signals and using statistical means such as generalized likelihood ratio tests to distinguish and identify different types of interference. However, these methods highly rely on prior knowledge and usually require manually setting complex threshold parameters, which to a certain extent limits their wide application in complex and changeable battlefield environments. Feature extraction-based methods, on the other hand, emphasize extracting features from radar signals in multiple domains (such as time domain, frequency domain, spatial domain, etc.) and combining machine learning algorithms or other advanced classifiers to achieve automatic recognition of interference signals. Although this method theoretically has higher flexibility and adaptability, manual feature extraction is not only time-consuming and laborious, increasing the recognition cost, but also the quality of the extracted features directly determines the accuracy and robustness of the recognition results. Therefore, how to automatically and efficiently extract high-quality features has become the key challenge in improving radar interference recognition performance. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for radar active composite interference recognition based on an efficient multi-modal caching model, including:

[0005] Obtain the time-domain data of simulated radar active interference, split the time-domain data of simulated radar active interference into a training set and a test set, and divide the training set into a training support set and a training query set;

[0006] Construct an efficient multi-modal caching model based on a visual caching module, a text caching module, and a residual connection module;

[0007] Train the efficient multi-modal caching model based on the training support set and the training query set, and test the trained model through the test set to obtain an interference recognition model;

[0008] Identify radar active compound jamming based on the jamming identification model.

[0009] Preferably, the radar active jamming time-domain data includes: noise product jamming, intelligent noise jamming, spectrum dispersion jamming, dense false target jamming, intermittent sampling and forwarding jamming, and frequency modulation jamming.

[0010] Preferably, the operation process of the efficient multi-modal cache model includes:

[0011] Construct key-value pairs based on the time-frequency image features and label one-hot encoding of the training support set jamming extracted by the visual cache module; wherein, the visual cache module includes a visual encoder, a visual feature optimizer, a one-hot encoder, and a visual cache. The visual feature optimizer consists of two fully connected layers and a dropout layer. The output dimension of the first fully connected layer is 512, and the output dimension of the second fully connected layer is 36. The visual encoder is composed of the visual encoding part of the pre-trained CLIP.

[0012] Extract text features constructed based on label prior information by the text cache module; wherein the text cache module includes a text encoder, a text feature optimizer, and a text cache. The text encoder is composed of the text encoding part of the pre-trained CLIP. The text feature optimizer consists of a fully connected layer and a dropout layer, and the output dimension of the fully connected layer is 36.

[0013] Extract the time-frequency features of the training query set jamming based on the visual encoder and the visual feature optimizer, and obtain the visual prediction features based on the visual cache to retrieve the time-frequency features of the training query set jamming and the visual affinity of the training support set.

[0014] Obtain text prediction features based on the text cache to retrieve the time-frequency features of the training query set jamming and the text affinity of the training support set.

[0015] Fuse the visual prediction features and the text prediction features based on the residual module to obtain the jamming identification result.

[0016] Preferably, the process of constructing key-value pairs based on the time-frequency image features and label one-hot encoding of the training support set jamming extracted by the visual cache module includes:

[0017] Convert the label of the training support set into one-hot encoding based on the one-hot encoder:

[0018] Convert the time-domain signal of the training support set into a time-frequency diagram based on the short-time Fourier transform;

[0019] Extract time-frequency features from the time-frequency diagram based on the visual encoder and the visual feature optimizer;

[0020] Store the time-frequency feature as the value of the visual cache and store the one-hot encoding as the key of the visual cache in the visual cache.

[0021] Preferably, the process of extracting text features based on label prior information by the text cache module includes:

[0022] Convert the training support set label into a simple text prompt, convert the simple text prompt into a text vector and input it into a text encoder to obtain text features;

[0023] Optimize the text features by the text feature optimizer to obtain an optimized text vector;

[0024] Form a text feature matrix from the text vectors related to each category in the optimized text vector and store it in the text cache.

[0025] Preferably, the expression for obtaining the visual prediction feature is:

[0026]

[0027] where is the visual prediction feature, β represents a hyperparameter for adjusting the sharpness of similarity, represents calculating the cosine similarity between the feature f query of a training query set sample and any sample feature in the training support set, and L train is the one-hot encoding matrix.

[0028] Preferably, the expression for obtaining the text prediction feature is:

[0029]

[0030] where is the text prediction feature, is the text feature matrix.

[0031] Preferably, the process of training the efficient multi-modal cache model based on the training support set and the training query set further includes:

[0032] Calculate the cross-entropy loss using the true labels of the training query set samples, freeze the parameters of the visual encoder and the text encoder in the model in combination with the efficient transfer learning strategy, and use the backpropagation algorithm to iteratively optimize the parameters of the visual feature optimizer and the text feature optimizer in the model until the loss value no longer decreases, then stop the iterative optimization and output the interference recognition model.

[0033] Preferably, the expression for obtaining the interference recognition result is:

[0034]

[0035] Among them, f test represents the visual time-frequency feature obtained after the interference of the test set is processed by the visual encoder and the visual feature optimizer, and S final is the interference recognition result.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] The present invention uses the optimized efficient multi-modal cache model to extract the interference features of the training set and store them in the visual cache and the text cache respectively. The test set interference time-frequency diagram is input into the visual encoder and the visual feature optimizer in the visual cache module to extract the interference time-frequency features of the training query set, and the prediction results are obtained by using the visual cache and the text cache respectively. Finally, the interference recognition result is obtained in the residual connection module. The present invention can efficiently utilize and deeply excavate limited interference samples, thereby improving the accuracy of radar active composite interference recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0039] Figure 1 is a schematic flow chart of the method for radar active composite interference recognition based on an efficient multi-modal cache model according to an embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of the efficient multi-modal cache model according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0042] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0043] Embodiment 1

[0044] As Figure 1-2 shown, in this embodiment, a method for radar active composite interference recognition based on an efficient multi-modal cache model is provided, including:

[0045] Simulate the time-domain data of active interference of a radar, and split the samples into a training set and a test set. Further split the training set into a training support set and a training query set for each epoch;

[0046] Construct an efficient multi-modal caching model;

[0047] Use the visual caching module to extract the time-frequency image features and one-hot encoded labels of the interference in the training support set, and construct key-value pairs to store in the visual cache;

[0048] Use the text caching module to extract the text features constructed based on the prior information of the labels and store them in the text cache;

[0049] Use the visual encoder and visual feature optimizer in the visual caching module to extract the time-frequency features of the interference in the training query set. At the same time, use the visual cache to retrieve the visual affinity between the time-frequency features of the interference in the training query set and the visual features of the training support set to obtain visual prediction features;

[0050] Use the text cache to retrieve the text affinity between the time-frequency features of the interference in the training query set and the text features of the training support set to obtain text prediction features;

[0051] Use the residual connection module to fuse the prediction results of the two modalities;

[0052] Use the true labels of the samples in the training query set to calculate the cross-entropy loss. Combine the efficient transfer learning strategy to freeze the parameters of the visual encoder and text encoder in the model, and only optimize the parameters of the visual feature optimizer and text feature optimizer in the model;

[0053] Use the optimized efficient multi-modal caching model to extract the interference features of the training set and store them in the visual cache and text cache respectively. Input the time-frequency map of the interference in the test set into the visual encoder and visual feature optimizer in the visual caching module to extract the time-frequency features of the interference in the training query set, and use the visual cache and text cache to obtain prediction features respectively. Finally, obtain the interference recognition result in the residual connection module;

[0054] In this embodiment, by constructing an efficient multi-modal caching model, the information in limited interference samples can be fully explored and utilized. Specifically, the model designs a visual caching module and a text caching module. The visual caching module is used to extract the time-frequency image features and one-hot encoded labels of the interference to form key-value pairs. By adopting an efficient key-value affinity calculation mechanism, the model effectively utilizes and deeply explores the prior information of the interference, thereby obtaining discriminative visual prediction features. At the same time, the text description constructed based on the interference prior information is input into the text caching module to extract text features and obtain text prediction features. In the designed residual connection module, the prediction features of the two modalities are fused to make full use of the complementarity between different modalities to improve the sensitivity and accuracy of the model in identifying similar interference types. During the model optimization process, an efficient transfer learning strategy is combined to optimize only some parameters of the model to improve its optimization efficiency.

[0055] In this embodiment, the time-domain data of simulated radar active interference is generated, and the samples are split into a training set and a test set. Each epoch further splits the training set into a training support set and a training query set, specifically including:

[0056] Use MATLAB to simulate and generate the time-domain data of six common single types and six efficient combined types of radar active interference. The single interference types include noise product jamming (NPJ), smart noise jamming (SNJ), smeared spectrum (SMSP), dense false target jamming (DFTJ), intermittent sampling repeater jamming (ISRJ), and modulated intermittent sampling repeater jamming (MISRJ). In addition, the six combined interferences are formed by combining the aforementioned single interferences in pairs, specifically the combination of DFTJ and ISRJ, the combination of DFTJ and MISRJ, the combination of MISRJ and ISRJ, the combination of MISRJ and SMSP, the combination of SMSP and DFTJ, and the combination of SMSP and ISRJ. Each interference type contains 500 simulated samples. Randomly select 3 samples among them to form the training set, and the remaining samples are classified into the test set. In each training cycle (epoch), the samples in the training set are randomly assigned again, with some as the training support set and the rest as the training query set.

[0057] In this embodiment, an efficient multi-modal caching model is constructed, referring to Figure 2, specifically including: The efficient multi-modal caching model mainly includes a visual caching module, a text caching module, and a residual connection module; among them, the visual caching module mainly includes a visual encoder, a visual feature optimizer, a one-hot encoder, and a visual cache. The visual encoder is taken from the visual encoding part of the pre-trained CLIP, and CLIP is pre-trained on two hundred million image-text pairs. The output end of the visual encoder is connected to the input end of the visual feature optimizer. The visual feature optimizer consists of two fully connected layers and a dropout layer, where the output dimension of the first fully connected layer is 512, and the output dimension of the second fully connected layer is 36. The one-hot encoder encodes according to the labels of the training support set samples. The output ends of the visual feature optimizer and the one-hot encoder are connected to the input end of the visual cache, and in the visual cache, the interference time-frequency features output by the visual feature optimizer and the label one-hot encoding are stored as key-value pairs for querying the categories of the training query set samples in the visual modality. The text caching module mainly includes three parts: a text encoder, a text feature optimizer, and a text cache. The text encoder is taken from the text encoding part of the pre-trained CLIP, and CLIP is pre-trained on two hundred million image-text pairs. The output end of the text encoder is connected to the input end of the text feature optimizer. The text feature optimizer consists of a fully connected layer and a dropout layer, where the output dimension of the fully connected layer is 36. The output end of the text feature optimizer is connected to the input end of the text cache, and the text features output by the text feature optimizer are stored in the text cache for querying the categories of the training query set samples in the text modality. The residual connection module consists of a feature fuser and a softmax. The input end of the feature fuser is connected to the output ends of the visual cache and the text cache. The output of the feature fuser undergoes a softmax operation to obtain the interference recognition result.

[0058] In a further embodiment, the visual caching module is used to extract the time-frequency image features and label one-hot encoding of the training support set interference and construct key-value pairs to be stored in the visual cache, specifically including:

[0059] In the visual caching module, the one-hot encoder is used to convert the training support set labels into one-hot encoding:

[0060] L train = OneHot(L N ). (1)

[0061] Among them, OneHot(L N ) represents the one-hot encoding operation. L N is the true label of the samples in the training support set.

[0062] The short-time Fourier transform is used to convert the time-domain signal of the training support set into a time-frequency diagram, and the visual encoder and the visual feature optimizer are used to extract the time-frequency features:

[0063] F train = VFEVFO(S K ). (2)

[0064] Wherein, VFEVFO(S K ) represents the combined operation of the visual encoder and the visual feature optimizer. S K is the two-dimensional time-frequency original image of the samples in the training support set.

[0065] Based on the cache model paradigm, the interference time-frequency feature F train is regarded as the value of the visual cache, and the label one-hot encoding L train is regarded as the key of the visual cache and stored in the visual cache.

[0066] In a further embodiment, a text cache module is configured to extract text features constructed based on label prior information and store them in the text cache, specifically including:

[0067] In the text cache module, the training support set labels are converted into simple texts in the form of "A photo of a {class}". For example, for MISRJ interference, the text prompt constructed in this embodiment is "A photo of a {MISRJ}". Subsequently, the text prompt is converted into a text vector through tokenize and input into the frozen CLIP pre-trained text encoder for text feature extraction. Immediately afterwards, the designed text feature adaptation head is used to further optimize the text feature representation:

[0068] f t = TETFO(T N ) (3)

[0069] Wherein, TETFO(T N ) represents the combined operation of the text encoder and the text feature adaptation head. T N is the text vector corresponding to the text prompt.

[0070] The text vectors related to each category in the text vector are combined into a text feature matrix and stored in the text cache.

[0071] In a further embodiment, the visual encoder and the visual feature optimizer in the visual cache module are used to extract the interference time-frequency features of the training query set. At the same time, the visual cache is used to retrieve the visual affinity between the interference time-frequency features of the training query set and the training support set to obtain visual prediction features, specifically including:

[0072] The short-time Fourier transform is used to obtain the time-frequency diagram of the samples in the training query set, and the visual encoder in the visual cache module is used to initially extract features, and the visual feature optimizer is used to further extract features f query. Utilize the interfering visual time-frequency feature f query Perform efficient key-value affinity calculation in the visual cache to obtain the visual prediction feature of the sample:

[0073]

[0074] Among them, β represents the hyperparameter for adjusting the sharpness of similarity, denotes calculating the cosine similarity between the feature f of the training query set sample query and the feature of any sample in the training support set. Through the exponential function exp, the value range of the similarity is limited to 0-1. This affinity describes the closeness of the training query sample and the sample in the training support set in the visual feature space. After calculating the affinity, in this embodiment, by multiplying the affinity with the one-hot encoding matrix L train of the label, the visual prediction feature of the training query set sample is obtained

[0075] In a further embodiment, utilize the text cache to retrieve the text affinity between the interfering time-frequency feature of the training query set and the training support set to obtain the text prediction feature, specifically including:

[0076] Utilize the interfering visual time-frequency feature f query to multiply with the text feature matrix in the text cache to obtain the text prediction feature of the training query set sample

[0077]

[0078] Preferably, utilize the residual connection module to fuse the two-modal prediction features, specifically including:

[0079] In the residual connection module, through the residual connection, jointly retrieve the query set interference prediction features obtained under the two-modal caches, and then through the softmax layer, the probability that the sample belongs to different interference categories can be obtained:

[0080]

[0081] Among them, α represents the residual ratio, which is used to balance the visual prediction feature and the text prediction feature

[0082] In a further embodiment, utilize the true label of the training query set sample to calculate the cross-entropy loss, and combine the efficient transfer learning strategy to freeze the parameters of the visual encoder and the text encoder in the model, and only optimize the parameters of the visual feature optimizer and the text feature optimizer in the model, specifically including:

[0083] Utilize the true label and the predicted label of the training query set sample to calculate the cross-entropy loss function:

[0084]

[0085] Among them, M represents the number of training query set samples, and y m represents the true label of the sample. The model combines an efficient transfer learning strategy to freeze the parameters of the visual encoder and the text encoder, and uses the reverse gradient propagation algorithm to iteratively optimize the parameters of the visual feature optimizer and the text feature optimizer in the model. The iterative optimization is stopped until the loss value no longer decreases, that is, the training of the proposed model is completed.

[0086] The AdamW optimizer is used in the optimization process, the learning rate is set to 0.001, and the weight decay coefficient is set to 0.0001.

[0087] In a further embodiment, the interference features of the training set are extracted by the optimized efficient multi-modal cache model and stored in the visual cache and the text cache respectively. The interference time-frequency map of the test set is input into the visual encoder and the visual feature optimizer in the visual cache module to extract the interference time-frequency features of the training query set, and the prediction features are obtained by using the visual cache and the text cache respectively. Finally, the interference recognition result is obtained in the residual connection module, specifically including:

[0088] Use the visual cache module in the optimized efficient multi-modal cache model to extract the interference video features of all samples in the training set as the key F train and store it in the visual cache, and perform one-hot encoding on all sample labels to form a one-hot encoding matrix as the value L train and store it in the visual cache. Use the text cache module to extract the text features of different categories and store them in the text cache.

[0089] Use the short-time Fourier transform to obtain the interference time-frequency map of the test set, and obtain the interference feature f of the test machine through the visual encoder and the visual feature optimizer in the visual cache module test , and then obtain the interference prediction features of different modalities in the visual cache and the text cache respectively, and connect them in the residual connection module to obtain the interference recognition result:

[0090]

[0091] Among them, f test represents the visual time-frequency feature obtained after the interference of the test set is processed by the visual encoder and the visual feature optimizer, and the meanings of the other letters are the same as those above.

[0092] The effect of this embodiment can be further illustrated by the following simulation, as shown in Tables 1 and 2.

[0093] 1. Simulation conditions

[0094] Table 1

[0095] CPU AMD Ryzen 7 4800H Memory 32 GB Graphics Card Model NVIDIA GeForce GTX 4070 Ti Graphics Card VRAM 12 GB

[0096] Table 2

[0097] Operating System Windows 10 64-bit Compiler PyCharm 2021.3.1 CUDA Version 11.2.0

[0098] 2. Simulation Method

[0099] (1) The radar interference classification methods based on convolutional neural network (CNN) in the prior art include deep fusion model (DFCNN), Siamese-CNN (S-CNN), Model-Agnostic Meta-Learning (MAML), JR-TFSAD, and CL-TD; (2) The method described in this embodiment, namely, the radar active composite interference recognition based on an efficient multi-modal caching model.

[0100] 3. Simulation Content and Results

[0101] The simulation experiment selects six single types of interference, namely, noise multiplication interference, intelligent noise interference, spectrum diffusion interference, dense false target interference, intermittent sampling and forwarding interference, and frequency modulation interference, as well as six efficient composite interferences, including the combination of dense false target interference and intermittent sampling and forwarding interference, the combination of dense false target interference and frequency modulation interference, the combination of frequency modulation interference and intermittent sampling and forwarding interference, the combination of frequency modulation interference and spectrum diffusion interference, the combination of spectrum diffusion interference and dense false target interference, and the combination of spectrum diffusion interference and intermittent sampling and forwarding interference. The time-frequency diagrams of the twelve interferences are obtained through short-time Fourier transform.

[0102] In Simulation Experiment 1, DFCNN, S-CNN, MAML, JR-TFSAD, and CL-TD are used to identify and classify the radar active deception interference at a fixed signal-to-noise ratio and sample size, as shown in Table 3. The active interference recognition method based on small-sample learning and multi-structure feature fusion in this embodiment is used to identify and classify the radar active deception interference signal at a fixed signal-to-noise ratio and sample size, as shown in Table 3.

[0103] Table 3

[0104]

[0105]

[0106] To verify the superiority of this embodiment under the condition of small samples, when there are 3 training samples for each type of interference, the recognition results of different models for 12 types of interference are provided, as shown in Table 1. It can be seen from Table 1 that in terms of the accuracy, precision, F1, and kappa of interference recognition, this embodiment is superior to other comparison models. Specifically, the recognition accuracy of this embodiment is as high as 92%, which is significantly better than methods such as DFCNN that rely heavily on the number of training samples. In addition, compared with the existing few recognition algorithms (such as S-CNN, MAML, JR-TFSAD, and CL-TD), the recognition accuracy of this embodiment has increased by more than 10%. More specifically, this embodiment achieves the highest recognition accuracy on 9 types of interference signals and maintains a recognition accuracy of more than 90% on 7 types of interference. Especially when facing DFTJ+ISRJ interference, the recognition accuracy of EMMCM is as high as 99.38%, while the recognition accuracy of the comparison model is less than 60%. In addition, the standard deviation of the interference recognition results of EMMCM is much lower than that of the comparison model, indicating that this embodiment has excellent stability. In short, the results in Table 3 show that this embodiment is superior to other comparison models in terms of the accuracy and stability of interference recognition.

[0107] Table 4

[0108]

[0109] Simulation Experiment 2. To quantitatively evaluate the optimization efficiency of this embodiment, the training set with 2, 3, and 30 labeled interference samples per class is used to train this embodiment for 100 epochs. At the same time, a few-sample interference recognition models such as S-CNN, MAML, JR-TFSAD, and CL-TD are used as comparison models. The training times of different models under different sample sizes are shown in Table 4. It can be seen from Table 4 that compared with the comparison models, this embodiment has the highest optimization efficiency under different sample sizes. Especially when there are only 2 training samples for each type of interference, this embodiment only needs 15.41 seconds to train for 100 epochs, which is much less than the few-sample recognition model S-CNN with relatively high optimization efficiency. This shows the effectiveness of this embodiment in combining an efficient transfer learning strategy by freezing the parameters of the visual encoder and text encoder in the model and only optimizing the parameters of the visual feature optimizer and text feature optimizer. This makes the proposed method have the highest optimization efficiency.

[0110] The above is only a preferred specific embodiment of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in this application should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A radar active composite interference recognition method based on an efficient multi-modal cache model, characterized in that Including: Obtain the time-domain data of the simulated radar active interference, split the time-domain data of the simulated radar active interference into a training set and a test set, and divide the training set into a training support set and a training query set; Construct an efficient multi-modal caching model based on a visual caching module, a text caching module, and a residual connection module; Train the efficient multi-modal caching model based on the training support set and the training query set, and test the trained model with the test set to obtain an interference recognition model; Identify the radar active composite interference based on the interference recognition model; The operation process of the efficient multi-modal caching model includes: Extract the time-frequency image features and one-hot encoded labels of the training support set interference based on the visual caching module to construct key-value pairs; wherein, the visual caching module includes a visual encoder, a visual feature optimizer, a one-hot encoder, and a visual cache, the visual feature optimizer consists of two fully connected layers and a dropout layer, the output dimension of the first fully connected layer is 512, and the output dimension of the second fully connected layer is 36; the visual encoder is based on the visual encoding part of the pre-trained CLIP; Extract text features constructed based on label prior information based on the text caching module; wherein the text caching module includes a text encoder, a text feature optimizer, and a text cache; the text encoder is based on the text encoding part of the pre-trained CLIP, the text feature optimizer consists of a fully connected layer and a dropout layer, and the output dimension of the fully connected layer is 36; Extract the time-frequency features of the training query set interference based on the visual encoder and the visual feature optimizer, and obtain visual prediction features based on the visual cache to retrieve the visual affinity between the time-frequency features of the training query set and the training support set; Obtain text prediction features based on the text cache to retrieve the text affinity between the time-frequency features of the training query set and the training support set; Fuse the visual prediction features and the text prediction features based on the residual module to obtain an interference recognition result; The process of extracting the time-frequency image features and one-hot encoded labels of the training support set interference based on the visual caching module to construct key-value pairs includes: Convert the labels of the training support set into one-hot encoding based on the one-hot encoder: Convert the time-domain signal of the training support set into a time-frequency diagram based on the short-time Fourier transform; Extract time-frequency features from the time-frequency diagram based on the visual encoder and the visual feature optimizer; Store the time-frequency features as the value of the visual cache and the one-hot encoding as the key of the visual cache into the visual cache.

2. The method according to claim 1, characterized in that, The time-domain data of the radar active interference includes: noise product interference, intelligent noise interference, spectrum spreading interference, dense false target interference, intermittent sampling and forwarding interference, and frequency modulation interference.

3. The method according to claim 1, wherein The process of extracting text features constructed based on label prior information based on the text caching module includes: Convert the training support set labels into simple text prompts, convert the simple text prompts into text vectors and input them into the text encoder to obtain text features; Optimize the text features based on the text feature optimizer to obtain an optimized text vector; Form a text feature matrix from the text vectors related to each category in the optimized text vector and store it in the text cache.

4. The method according to claim 1, wherein The expression for obtaining the visual prediction feature is: Among them, is the visual prediction feature, and β represents the hyperparameter that adjusts the sharpness of similarity. represents calculating the cosine similarity between the feature f query of the training query set sample and the feature of any sample in the training support set. L train is the one-hot encoding matrix.

5. The method according to claim 1, characterized in that, The expression for obtaining the text prediction feature is: Among them, is the text prediction feature, is the text feature matrix.

6. The method according to claim 1, characterized in that The process of training the efficient multi-modal cache model based on the training support set and the training query set further includes: Calculate the cross-entropy loss using the true labels of the training query set samples, freeze the parameters of the visual encoder and the text encoder in the model in combination with the efficient transfer learning strategy, and use the backpropagation algorithm to iteratively optimize the parameters of the visual feature optimizer and the text feature optimizer in the model until the loss value no longer decreases, then stop the iterative optimization and output the interference recognition model.

7. The method according to claim 1, characterized in that, The expression for obtaining the interference recognition result is: Among them, f test represents the visual time-frequency feature obtained after the test set interference is processed by the visual encoder and the visual feature optimizer, and S final is the interference recognition result.

Images