Small sample radar signal intelligent identification method and electronic equipment

Abstract

The invention discloses a small sample radar signal intelligent identification method and electronic equipment, and the method comprises the steps: carrying out the preprocessing of a radar signal through CWD, and obtaining real sample data; inputting the real sample data into a DCGAN network, and expanding the real sample data to obtain corresponding generated sample data; and inputting the real sample data and the generated sample data into a VIT network to form a small sample radar signal intelligent identification network model. According to the method, the CWD is adopted to preprocess the data set of the radar signals, so that the method has relatively good energy aggregation while keeping a certain time-frequency resolution, and subsequent further processing is facilitated. A DCGAN network is combined with a VIT network to establish a small sample radar signal intelligent identification network model, a large amount of generated sample data with distribution closer to that of real sample data is generated to train the identification model, the fitting effect is more stable, and the identification accuracy of small sample radar signals is improved.

Description

Technical Field

[0001] The present invention relates to the field of digital signal processing, and in particular to a small sample radar signal intelligent recognition method and electronic equipment. Background Art

[0002] With the continuous development of modern signal processing technology, more and more radars are working in various complex ways, and more information categories are entering the information battlefield. This makes the problem of how to identify unknown radar signals with insufficient information samples increasingly prominent.

[0003] Early radar signal matching template recognition methods are generally based on the instantaneous changes in the signal time domain, so as to model and analyze certain statistical features, establish radar signal matching templates, and thus realize the recognition of radar signals. The above-mentioned radar signal matching template recognition method essentially requires a certain amount of labeled signals for manual analysis. However, when the sample data volume of radar signals is small and the information samples are insufficient, it is difficult to establish a multi-dimensional feature matching template suitable for such small sample radar signals through the above-mentioned radar signal matching template recognition method, and the recognition effect is poor, which has obvious limitations.

[0004] With the introduction of the concept of deep learning and the successful application of deep learning technology in the field of computer vision, people firmly believe that this technology can also be used to automatically extract the deep features of radar signals, thereby realizing the recognition of radar signals.

[0005] However, in some scenarios, radar signals are expensive to acquire and difficult to label, so it is still difficult to build a large and complex radar signal dataset. The current deep learning network for small sample radar signal datasets has poor robustness and is prone to overfitting, resulting in a low radar signal recognition rate. It cannot effectively solve the problem of how to identify unknown radar signals with small sample data and insufficient information samples, and cannot meet the recognition needs of small sample radar signals in today's informationized battlefield. Summary of the invention

[0006] The purpose of the present invention is to provide a small sample radar signal intelligent recognition method and electronic equipment, which is used to address the problem that the current deep learning network of small sample radar signal data sets has poor robustness and is prone to overfitting, resulting in a low radar signal recognition rate. By using CWD to preprocess the radar signal data set, while maintaining a certain time-frequency resolution, it also has better energy concentration. A small sample radar signal intelligent recognition network model is established by combining the DCGAN network with the VIT network, and a large amount of generated sample data with a distribution closer to the real sample data is generated to train the recognition model. The fitting effect is more stable, and the recognition accuracy of small sample radar signals is improved.

[0007] In order to achieve the above objectives, the present invention is implemented by the following technical solutions:

[0008] A first aspect of the present invention provides a small sample radar signal intelligent recognition method, comprising:

[0009] Preprocess the radar signal through CWD (Choi-Williams Distribution) to obtain real sample data;

[0010] Input the real sample data into the DCGAN (Deep Convolution Generative Adversarial Networks) network to expand the real sample data to obtain corresponding generated sample data;

[0011] The real sample data and the generated sample data are input into a VIT (Vision Transformer) network to form a small sample radar signal intelligent recognition network model.

[0012] Optionally, the preprocessing of the radar signal by CWD includes:

[0013] The radar signal is converted into a CWD time-frequency image by the following formula to obtain the real sample data:

[0014]

[0015] Among them, x * x(t) is the complex conjugate of the radar signal x(t); τ is the time delay; ω is the angular frequency; Ω is the integral variable; Jiang is the width parameter of the kernel; t is time; and f is frequency.

[0016] Optionally, the DCGAN network includes a discriminant network and a generator network;

[0017] The discriminant network is used to identify the real sample data; and the generative network is used to generate the generated sample data.

[0018] Optionally, inputting the real sample data into the DCGAN network to expand the real sample data comprises the following steps:

[0019] S21, scaling the real sample data and generating multiple Gaussian random noise vectors;

[0020] S22, setting parameters of the DCGAN network;

[0021] S23, randomly extracting part of the real sample data, and selecting part of the Gaussian random noise vector as the input of the generating network;

[0022] S24, alternately training the discriminant network and the generative network;

[0023] S25. Repeat steps S21 to S22 to complete the training of the DCGAN network.

[0024] Optionally, the parameters of the DCGAN network include: a maximum iteration step; a learning rate of the discriminant network; and a learning rate of the generator network.

[0025] Optionally, the real sample data of the randomly selected part includes:

[0026] The same amount of the real sample data is randomly selected from the real sample data of each category as the input of the generation network.

[0027] Optionally, the step of inputting the real sample data and the generated sample data into the VIT network to form the small sample radar signal intelligent recognition network model comprises the following steps:

[0028] S31, setting the parameters of the VIT network, and performing scaling processing on the real sample data and the generated sample data;

[0029] S32, dividing the real sample data and the generated sample data after scaling processing into a plurality of image blocks, and converting each of the image blocks into a vector sequence of a fixed length;

[0030] S33, processing the position information of the elements in the vector sequence;

[0031] S34, further processing the vector sequence to encode and learn image features;

[0032] S35, mapping the encoded image features to category labels to generate recognition results;

[0033] S36, repeating steps S31 to S34 to form the small sample radar signal intelligent recognition network model.

[0034] Optionally, the parameters of the VIT network include: maximum iteration step size; learning rate of the VIT network; hidden layer dimension; number of encoder layers; number of heads of the multi-head attention mechanism.

[0035] Optionally, the small sample radar signal intelligent identification method further includes:

[0036] The radar signal to be identified is preprocessed by CWD, and then identified by the small sample radar signal intelligent identification network model.

[0037] A second aspect of the present invention provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the small sample radar signal intelligent recognition method described in any one of the first aspects is implemented.

[0038] The present invention has at least one of the following technical effects:

[0039] By using CWD to preprocess the radar signal data set to obtain real sample data, it can be ensured that the real sample data has good energy concentration while maintaining a certain time-frequency resolution, thus facilitating subsequent further processing;

[0040] By expanding the radar signal dataset through the DCGAN network, the hierarchical features in the image can be captured, and the deep features of the real samples can be better learned without labels, so that the generated sample data generated based on the real samples have a distribution closer to the real sample data, ensuring the recognition accuracy of the small sample radar signal intelligent recognition network model established by the generated sample data and the real sample data;

[0041] By using the time-frequency image set of real sample data and generated sample data to train the VIT network, complex image features can be effectively learned and encoded, thereby forming a small sample radar signal intelligent recognition network model and realizing the recognition of complex small sample radar signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of a flow chart of a small sample radar signal intelligent recognition method provided by an embodiment of the present invention;

[0043] Figure 2 The CWD time-frequency diagram is obtained by preprocessing different types of radar signals;

[0044] Figure 3 This is a schematic diagram of the structure of the decision network and the generation network of the DCGAN network;

[0045] Figure 4 It is a schematic diagram of the structure of the VIT network;

[0046] Figure 5 The simulation experiment results of small sample radar signal recognition. DETAILED DESCRIPTION

[0047] The following is a further detailed description of a small sample radar signal intelligent identification method and electronic device proposed by the present invention in combination with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the accompanying drawings are in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purposes, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention, so they have no technical substantive significance. Any structural modification, change in proportional relationship or adjustment of size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention.

[0048] like Figure 1 As shown, the first aspect of this embodiment provides a small sample radar signal intelligent recognition method, including the following steps: step S1, preprocessing the radar signal through CWD to obtain real sample data; step S2, inputting the real sample data into the DCGAN network, expanding the real sample data, and obtaining corresponding generated sample data; step S3, inputting the real sample data and the generated sample data into the VIT network to form a small sample radar signal intelligent recognition network model; step S4, preprocessing the radar signal to be identified through CWD; step S5, identifying through the small sample radar signal intelligent recognition network model.

[0049] In step S1, the collected small sample radar signal can be preprocessed by CWD, i.e., Choi-Williams time-frequency distribution, and the small sample radar signal can be converted into a corresponding CWD time-frequency image, i.e., the real sample data. By using CWD to preprocess the radar signal, it can have better energy concentration while maintaining a certain time-frequency resolution.

[0050] Specifically, the radar signal x(t) can be converted into a CWD time-frequency image CWD(t,f) by the following formula to obtain the real sample data:

[0051]

[0052] Where x(t) is the input small sample radar signal; x *(t) is the complex conjugate of the signal; τ is the time delay; ω and Ω are the angular frequency and integral variable respectively; σ is the width parameter of the kernel, which controls the degree of cross-term suppression; t is time; f is frequency. CWD(t,f) is the time-frequency representation of the input small sample radar signal at time t and frequency f. According to the above formula, two-dimensional integration is performed, and the resulting CWD(t,f) is the CWD time-frequency image obtained by converting the input small sample radar signal. Figure 2 As shown in FIG. 1 , by processing different types of radar signals through CWD, different types of CWD time-frequency images can be obtained.

[0053] Furthermore, in step S2, the converted real sample signal CWD time-frequency image, that is, the real sample data, can be input into the DCGAN network, so as to expand the small sample radar signal and obtain the corresponding generated sample signal CWD time-frequency image, that is, the generated sample data.

[0054] The DCGAN network, namely the deep convolutional generative adversarial network, is based on the theory of generative adversarial network (GAN), and innovatively combines convolutional neural network (CNN) and generative adversarial network to achieve image processing, and proposes a series of restrictions on the network structure to improve the stability of the network, so that the network can capture the hierarchical features in the image and better learn the deep features of the real sample without labels. The DCGAN network may include a discriminant network and a generative network, wherein the discriminant network is used to identify the real sample data; the generative network is used to generate the generated sample data.

[0055] Specifically, step S2 includes the following steps: step S21, scaling the real sample data and generating multiple Gaussian random noise vectors at the same time; step S22, setting the parameters of the DCGAN network; step S23, randomly extracting part of the real sample data, and selecting part of the Gaussian random noise vectors as the input of the generating network; step S24, alternately training the discriminant network and the generating network; step S25, repeating steps S21 to S22 to complete the training of the DCGAN network.

[0056] More specifically, in step S21, the real sample data obtained by CWD, that is, the real sample signal CWD time-frequency image, can be scaled to an image with a pixel size of 64*64 and a channel number of 3, and K 100-dimensional Gaussian random noise vectors are generated at the same time, which obey a uniform distribution in the range of [-1,1].

[0057] In step S22, the parameters of the DCGAN network may be set, and the parameters of the DCGAN network may include, for example, a maximum iteration step, a learning rate of the discriminant network, and a learning rate of the generator network. Specifically, the maximum iteration step may be 10,000 steps, and the learning rates of the discriminant network and the generator network may be set to 0.001. The parameters that need to be set for the DCGAN network and the specific values ​​of the parameters may be adjusted according to actual needs, and the present invention is not limited thereto.

[0058] In step S23, M real data samples can be randomly selected from the preprocessed real sample signal CWD time-frequency image for each type of signal, and N Gaussian random noise samples can be selected as the input of the generating network. That is, the same number of real sample data of each type is randomly selected as the input of the generating network. For example, Figure 2 , from 13 different categories of real sample signal CWD time-frequency images, M samples of each category are randomly selected as the input of the generation network.

[0059] In step S24, after M real data samples and N Gaussian random noise samples are input into the DCGAN network as the same batch, the discriminant network and the generative network are trained alternately. Preferably, the discriminant network is trained 5 times and the generative network is trained once each time, and the updates are performed alternately. The generative network generates the generated sample data that is closer to the distribution of the real sample data by optimizing the generation loss function; and the discriminant network optimizes the discriminant loss function so that the discriminant network can identify the real sample data as much as possible.

[0060] Finally, in step S25, the training of the DCGAN network can be completed by repeating steps S21 to S22 according to the initially set training parameters. After the training is completed, a large number of generated sample signal CWD time-frequency images can be generated and mixed with real sample signal CWD time-frequency images to form a new CWD time-frequency image data set, which can be used for training the VIT network.

[0061] The structures of the discriminant network and the generative network are as follows: Figure 3 As shown in , the input of the discriminant network is an image with 64*64 pixels and 3 channels, and the discriminant probability distribution is obtained through 4 convolution blocks D_conv (D_conv_1, D_conv_2, D_conv_3 and D_conv_4) and 1 fully connected layer D_MLP.

[0062] The input of the generative network is a 100-dimensional Gaussian random noise, which is uniformly distributed in the range of [-1,1]. It is respectively passed through a fully connected layer G_MLP and four transposed convolution layers G_deconv (G_deconv_1, G_deconv_2, G_deconv_3 and G_deconv_4) to obtain the generated CWD time-frequency image samples, i.e., the generated sample data. Among them, the convolution kernel of each transposed convolution layer is 5*5, and the stride is 2. Downsampling is achieved through the stride convolution layer to extract image features, and upsampling is achieved through the transposed convolution layer to improve the resolution of the generated image.

[0063] Specifically, the generative network and the discriminative network both use a cross entropy loss function. For the generative loss function of the generative network, the specific mathematical expression is as shown in the following formula:

[0064] Loss_G = -logD(G(z));

[0065] Where G represents the generating network, D represents the discriminative network, and z represents the random noise vector.

[0066] For the discriminant loss function of the discriminant network, the specific mathematical expression is shown as follows:

[0067] Loss_D=-log D(x)-log(1-D(G(z)));

[0068] Wherein, x is the real sample data.

[0069] Furthermore, in step S3, the real sample signal CWD time-frequency image and the generated sample signal CWD time-frequency image are input into the VIT network for training, so that the small sample radar signal intelligent recognition network model can be obtained. The VIT network, i.e., the visual transformation network, as a deep learning network model, is suitable for solving image recognition and speech recognition. Significant breakthroughs have been made in the field of computer vision, especially in processing image data.

[0070] like Figure 4 As shown in the figure, the structure of the VIT network mainly includes the image feature embedding module (Input Embedding), the positional encoding module (Positional Encoding), the Transformer encoder module and the fully connected layer classification module (MLP). The Transformer encoder module includes a multi-head attention mechanism (Multi-headAttention), a feedforward neural network (Feed Forward), a batch normalization layer and a residual network (Add&Norm).

[0071] Specifically, step S3 includes the following steps: S31, setting the parameters of the VIT network, and scaling the real sample data and the generated sample data; S32, dividing the scaled real sample data and the generated sample data into multiple image blocks, and converting each image block into a vector sequence of fixed length; S33, processing the position information of the elements in the vector sequence; S34, further processing the vector sequence, encoding and learning the image features; S35, mapping the encoded image features to the category label to generate the recognition result; S36, repeating steps S31 to S34 to form the small sample radar signal intelligent recognition network model.

[0072] In step S31, the parameters of the VIT network are first set. The parameters of the VIT network include, for example, the maximum iteration step, the learning rate of the VIT network, the hidden layer dimension, the number of encoder layers, and the number of heads of the multi-head attention mechanism. Specifically, the maximum iteration step can be set to 10,000 steps, the learning rate of the VIT network can be set to 0.001, the hidden layer dimension can be set to 256, the number of encoder layers can be set to 6, and the number of heads of the multi-head attention mechanism can be set to 12. The parameters that need to be set for the VIT network and the specific values ​​of the parameters can be adjusted according to actual needs, and the present invention is not limited to this.

[0073] Furthermore, we can Figure 4 The input Inputs in the data set, i.e., the CWD time-frequency image data set obtained by mixing the generated sample data and the real sample data, is scaled. For example, the CWD time-frequency images of all 13 radar small sample signals are scaled to obtain an image of 224*224 pixels.

[0074] In step S32, the scaled CWD time-frequency image dataset can be randomly sampled in batches and input into the VIT network. The image feature embedding module can divide the 224*224 pixel image into multiple 16*16 pixel image blocks, and each of the image blocks can be converted into a vector sequence of fixed length.

[0075] In step S33, the position information of the feature elements in the vector sequence may be processed by the position encoding module so that the model formed by the VIT network can understand the spatial structure of the data.

[0076] In step S34, a Transformer encoder module may be used to further process the vector sequence processed in step S33, so as to encode and learn the embedded image features.

[0077] In step S35, the encoded and learned image features can be mapped to the final category label through the fully connected layer classification module to generate the final recognition result, that is, Figure 4 Outputs in.

[0078] Finally, in step S36, according to the training parameters initially set, steps S31 to S34 are repeated to complete the training of the model formed by the VIT network, thereby finally forming the small sample radar signal intelligent recognition network model. By training the VIT network using the time-frequency image set of the real sample data and the generated sample data, complex image features can be effectively learned and encoded, thereby forming a small sample radar signal intelligent recognition network model and realizing the recognition of complex small sample radar signals.

[0079] After completing the training of the small sample radar signal intelligent recognition network model, in step S4 and step S5, the small sample radar intra-pulse modulation signal to be identified can be first converted into the CWD time-frequency image of the signal to be identified by using CWD, and then input into the trained small sample radar signal intelligent recognition network model to obtain the recognition result.

[0080] In other aspects, this embodiment further provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the small sample radar signal intelligent recognition method as described above is implemented.

[0081] In order to verify the effectiveness of the small sample radar signal intelligent recognition method described above, the technicians conducted a simulation test. The simulation test selected 13 small sample radar signals with a signal-to-noise ratio range of -12dB to 8dB as test radar signals, and changed the test points with a step size of 2dB. Training and recognizing radar signal data with different signal-to-noise ratios can make radar signal recognition more flexible and adaptable to changing environments.

[0082] Specifically, during the training process of the small sample radar signal intelligent recognition network model, 13 types of signal samples are selected as the training set radar signals, with 10 samples of each type; during the test, 13 types of signal samples are selected, with 700 samples of each type, to test the recognition effect of the small sample radar signal intelligent recognition network model. Table 1 shows the specific parameters of the 13 types of signal samples.

[0083]

[0084]

[0085] Table 1

[0086] In Table 1, U represents uniform distribution, fs represents the sampling frequency, B represents the bandwidth, N represents the number of sampling points, f c represents the center frequency, FH represents frequency hopping, f m Represents the lowest frequency, L c Represents the code length, c pp represents the number of code cycles per phase, M represents the frequency order, n s Represents the number of subcodes, n g Represents the number of segments, n p represents the phase number, ψ represents the path delay, G represents the average path gain, and f Dmax Represents the maximum Doppler shift.

[0087] The simulation results are as follows Figure 5 As shown, it can be seen that in the case of small sample radar signals, the present invention establishes a small sample radar signal intelligent recognition network model through the DCGAN network combined with the VIT network. When identifying 13 small sample radar modulation signals with a signal-to-noise ratio of -12dB, the recognition rate is still higher than 80%, which is much higher than the recognition rate of the traditional neural network recognition VGG16 algorithm. Moreover, with the improvement of the signal-to-noise ratio, the recognition accuracy gradually stabilizes above 95%. This is because the DCGAN network is used to achieve data enhancement of small sample radar modulation signals, so that the VIT network can have a sufficient and complete training sample set, and finally achieve good recognition results.

[0088] The present invention uses CWD to preprocess the radar signal data set, while maintaining a certain time-frequency resolution, it also has good energy aggregation, which is convenient for subsequent further processing. A small sample radar signal intelligent recognition network model is established by combining the DCGAN network with the VIT network, and a large amount of generated sample data with a distribution closer to the real sample data is generated to train the recognition model. The fitting effect is more stable, and the recognition accuracy of small sample radar signals is improved.

[0089] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0090] It should be noted that the devices and methods disclosed in the embodiments of this article can also be implemented in other ways. The device implementation described above is only schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of this article. In this regard, each box in the flowchart or block diagram can represent a part of a module, program or code, and the module, program segment or part of the code contains one or more executable instructions for implementing the specified logical function, and the module, program segment or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented by a dedicated hardware-based system for performing a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0091] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A small sample radar signal intelligent recognition method, characterized in that: include: The radar signal is preprocessed through CWD to obtain real sample data; Inputting the real sample data into the DCGAN network, expanding the real sample data, and obtaining corresponding generated sample data; The real sample data and the generated sample data are input into the VIT network to form a small sample radar signal intelligent recognition network model.

2. The small sample radar signal intelligent recognition method according to claim 1 is characterized in that: The preprocessing of the radar signal by CWD includes: The radar signal x(t) is converted into a CWD time-frequency image CWD(t,f) by the following formula to obtain the real sample data: Among them, x * x(t) is the complex conjugate of the radar signal x(t); τ is the time delay; ω is the angular frequency; Ω is the integration variable; σ is the width parameter of the kernel; t is time; and f is frequency.

3. The small sample radar signal intelligent recognition method according to claim 1 is characterized in that: The DCGAN network includes a discriminant network and a generator network; The discriminant network is used to identify the real sample data; and the generative network is used to generate the generated sample data.

4. The small sample radar signal intelligent recognition method according to claim 3 is characterized in that: The inputting the real sample data into the DCGAN network and expanding the real sample data comprises the following steps: S21, scaling the real sample data and generating multiple Gaussian random noise vectors; S22, setting parameters of the DCGAN network; S23, randomly extracting part of the real sample data, and selecting part of the Gaussian random noise vector as the input of the generating network; S24, alternately training the discriminant network and the generative network; S25. Repeat steps S21 to S22 to complete the training of the DCGAN network.

5. The small sample radar signal intelligent recognition method according to claim 4 is characterized in that: The parameters of the DCGAN network include: maximum iteration step size; learning rate of the discriminant network; learning rate of the generator network.

6. The small sample radar signal intelligent recognition method according to claim 4 is characterized in that: The real sample data of the random sampling part includes: The same amount of the real sample data is randomly selected from the real sample data of each category as the input of the generation network.

7. The small sample radar signal intelligent recognition method according to claim 1 is characterized in that: The step of inputting the real sample data and the generated sample data into the VIT network to form the small sample radar signal intelligent recognition network model comprises the following steps: S31, setting the parameters of the VIT network, and performing scaling processing on the real sample data and the generated sample data; S32, dividing the real sample data and the generated sample data after scaling processing into a plurality of image blocks, and converting each of the image blocks into a vector sequence of a fixed length; S33, processing the position information of the elements in the vector sequence; S34, further processing the vector sequence to encode and learn image features; S35, mapping the encoded image features to category labels to generate recognition results; S36, repeating steps S31 to S34 to form the small sample radar signal intelligent recognition network model.

8. The small sample radar signal intelligent recognition method according to claim 7 is characterized in that: The parameters of the VIT network include: maximum iteration step size; learning rate of the VIT network; hidden layer dimension; number of encoder layers; number of heads of the multi-head attention mechanism.

9. The small sample radar signal intelligent recognition method according to claim 1, characterized in that: The small sample radar signal intelligent recognition method also includes: The radar signal to be identified is preprocessed by CWD, and then identified by the small sample radar signal intelligent identification network model.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the small sample radar signal intelligent recognition method according to any one of claims 1 to 9 is implemented.

Images