Two-stage radar data enhancement method and system
Through the two-stage radar data enhancement method, combined with time-frequency structure transformation and generation of adversarial network, the lack of sample diversity and authenticity in radar signal data enhancement technology is solved, and high-quality and diverse data set generation is achieved, suitable for complex Radar signal processing in electromagnetic environment.
Patent Information
- Application Number
- CN202411121602.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-12
AI Technical Summary
The existing radar signal data enhancement technology is limited to single domain transformation, making it difficult to fully capture and enhance the multi-dimensional characteristics of radar signals, resulting in insufficient diversity and authenticity of generated samples and cannot meet the high-quality data needs in complex electromagnetic environments.
The two-stage radar data enhancement method is used to simulate the time-frequency structure and amplitude changes through primary data enhancement, and the secondary data enhancement is combined with the generation adversarial network to simulate the complex characteristics of the real signal to ensure the high quality and diversity of the generated data.
The sample diversity is enriched through time domain and amplitude domain transformation, and the sample authenticity is enhanced by using the generative adversarial network. The generated data set is highly close to the actual application scenario, solving the problem of sample scarcity and ensuring high-quality and diversified data sets.
Smart Images

Figure CN118938163A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of radar data enhancement, and in particular to a two-stage radar data enhancement method and system. Background Art
[0002] As the electromagnetic environment becomes increasingly complex, the performance of frequency-using equipment is affected. Because certain types of electromagnetic signals are difficult to capture in actual scenarios and confidentiality policies strictly restrict data sharing, although the radar signal sample library is constantly expanding, samples of certain specific signals are still very scarce. The scarcity of samples limits the comprehensive simulation and analysis of complex electromagnetic environments, which in turn affects the accurate evaluation and optimization of the performance of frequency-using equipment.
[0003] Currently, most data enhancement techniques are limited to transformations in a single domain, which limits the comprehensive capture and effective enhancement of the multi-dimensional features of radar signals, thereby weakening the diversity and authenticity of the generated samples, making it difficult to meet the demand for high-quality data in radar signal processing. Summary of the invention
[0004] In order to solve the problems existing in the prior art, the embodiments of the present disclosure provide a two-stage radar data enhancement method and system to improve the diversity and authenticity of the generated radar signal samples. The technical solution is as follows: In a first aspect, a two-stage radar data enhancement method is provided, comprising: Acquire the radar signal to be enhanced, and perform first-level data enhancement on the radar signal to be enhanced; Based on the generative adversarial network, the signal after the primary data enhancement is subjected to the secondary data enhancement; and, performing quality assessment on the signal after secondary data enhancement; The first-level data enhancement of the radar signal to be enhanced includes time domain transformation and amplitude domain transformation of the radar signal to be enhanced, wherein the time domain transformation includes time distortion of the radar signal to be enhanced, and the amplitude domain transformation includes scaling and adding noise to the signal after the time domain transformation.
[0005] In a second aspect, a two-stage radar data enhancement system is provided, comprising: A primary data enhancement module is configured to obtain a radar signal to be enhanced and perform primary data enhancement on the radar signal to be enhanced; A secondary data enhancement module is configured to perform secondary data enhancement on the signal after the primary data enhancement based on a generative adversarial network; A quality assessment module is configured to perform quality assessment on the signal after the secondary data enhancement; The first-level data enhancement of the radar signal to be enhanced includes time domain transformation and amplitude domain transformation of the radar signal to be enhanced, wherein the time domain transformation includes time distortion of the radar signal to be enhanced, and the amplitude domain transformation includes scaling and adding noise to the signal after the time domain transformation.
[0006] In a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to complete the steps of the above-mentioned two-stage radar data enhancement method.
[0007] In a fourth aspect, a computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, complete the steps of the above-mentioned two-stage radar data enhancement method.
[0008] In a fifth aspect, a computer program product is provided, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned two-stage radar data enhancement method.
[0009] The technical solution provided by the disclosed embodiment has the following beneficial effects: In the disclosed embodiment, by simultaneously transforming the radar signal to be enhanced in the time domain and the amplitude domain, simulating the time-frequency structure and amplitude variation of the radar signal, the signal is effectively enhanced from two dimensions, the dynamic range and randomness of the signal are enhanced, and the diversity of the samples is enriched; through a two-stage cascaded data enhancement mode, a generative adversarial network is used to perform a second-stage enhancement on the signal after the first-stage enhancement, simulating the complex characteristics of the real radar signal, further enhancing the authenticity of the sample, and ensuring that the generated data set can be highly close to the actual application scenario. The two-stage transformation complements each other, aiming to solve the problem of sample scarcity while ensuring the high quality and diversity of the generated data.
[0010] Advantages of additional aspects of the present disclosure will be given in part in the following description and in part will become apparent from the following description or will be learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0012] Figure 1 is a flow chart of a two-stage radar data enhancement method provided by an embodiment of the present disclosure; Figure 2 The overall flow chart of the method is shown in FIG. Figure 3 (a) is the signal waveform before time distortion; Figure 3 (b) shows the signal waveform after time distortion when d is a negative value; Figure 3 (c) shows the signal waveform after time distortion when d is a positive value; Figure 4 It is the signal waveform diagram before and after amplitude scaling; Figure 5 This is the signal waveform before and after adding Gaussian white noise; Figure 6 This is the signal waveform before and after adding impulse noise; Figure 7 This is the generator network structure diagram; Figure 8 It is the structure diagram of the discriminator network; Fig. 9 This is a comparison chart of the correlation coefficients between the enhanced data and the sample data based on the three methods. DETAILED DESCRIPTION
[0013] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.
[0014] The present disclosure provides a two-stage radar data enhancement method. Figure 1-Figure 2 , the method comprising: S101, obtaining a radar signal to be enhanced, and performing first-level data enhancement on the radar signal to be enhanced; S102, performing secondary data enhancement on the signal after primary data enhancement based on a generative adversarial network; S103, performing quality assessment on the signal after the secondary data enhancement; The first-level data enhancement of the radar signal to be enhanced includes time domain transformation and amplitude domain transformation of the radar signal to be enhanced, wherein the time domain transformation includes time distortion of the radar signal to be enhanced, and the amplitude domain transformation includes scaling and adding noise to the signal after the time domain transformation.
[0015] In order to solve the problems of scarce radar signal samples and difficulty in reconstructing complex signals, the disclosed embodiment performs data enhancement based on a two-level data enhancement method of hybrid domain transformation and generative adversarial network. Specifically, the first-level data enhancement adopts a hybrid domain transformation strategy that combines the time domain and the amplitude domain to simulate the time-frequency structure and amplitude variation of radar signals and enrich the diversity of samples; the second-level data enhancement uses a generative adversarial network to simulate the complex characteristics of real radar signals, enhance the authenticity of samples, and ensure that the generated data set is highly close to the actual application scenario. The two-level transformation complements each other and aims to solve the problem of sample scarcity while ensuring the high quality and diversity of the generated data, thereby obtaining a high-quality and diverse enhanced data set.
[0016] In step S101, the purpose of the first-level data enhancement is to simulate the time-frequency structure and amplitude variation of the radar signal through a mixed domain transformation strategy, thereby enriching the diversity of samples. Specifically, this stage combines the time warping technology in the time domain with the scaling and noise addition technology in the amplitude domain. The time warping technology simulates the variation of radar signals under different sampling rates and time offsets to capture the time-varying characteristics of the signal. Scaling and noise addition techniques enhance the dynamic range and randomness of the signal, simulating radar signals under different transmission power or receiving sensitivity conditions. Through these technical means, the first-level data enhancement can not only improve the diversity of samples, but also enhance the robustness of the signal in actual complex electromagnetic environments, providing high-quality basic samples for subsequent second-level data enhancement.
[0017] In the time domain, time warping technology is used to simulate the variation of radar signals under different sampling rates and time offsets. Time warping first divides the original data into segments, assuming that the number of data points contained in each segment is , then the amplitude value vector of this signal is It can be expressed as follows: ; in, Indicates the amplitude value of each sampling point.
[0018] Next, select an anchor point , deform the signal near this point and randomly generate a moving distance If the randomly determined moving distance If is a negative value, the signal on the left side of the anchor point is contracted and the signal on the right side is stretched; if If the value is positive, the signal on the left side of the anchor point is stretched to the right, and the signal on the right side is compressed. It can be calculated by the following formula: ; in, is the linear interpolation value of the sample signal to the left of the anchor point, is the linear interpolation value of the sample signal to the right of the anchor point. , Can be obtained according to the following formula: ; .
[0019] Time warping mainly generates new samples by deforming the original time series data in the time dimension, that is, a time point is selected as an anchor point, and then the signal is stretched or compressed on both sides of the anchor point according to the randomly determined distance, thereby generating a copy that is offset in time. When the moving distance d is a negative value, the signal on the left side of the anchor point is compressed to reduce the number of d sampling points, and the amplitude value of each sampling point after compression is obtained based on the linear interpolation of the signal before compression; the signal on the right side of the anchor point is stretched to increase the number of d sampling points, and the amplitude value of each sampling point after stretching is obtained based on the linear interpolation of the signal before stretching. When the moving distance d is a positive value, the signal on the right side of the anchor point is compressed and the signal on the left side of the anchor point is stretched in the same way. See Figure 3 (a)-Figure 3 (c). Figure 3 (a) shows a signal waveform diagram containing 50 sampling points, and the dotted line marked position is the selected anchor point τ. When d takes a negative value, take d=-5 as an example, see Figure 3 (b), the signal on the left side of τ originally containing 20 sampling points is compressed to 15, and the amplitude value of each sampling point after compression is obtained by linear interpolation of the original signal; the signal on the right side containing 30 sampling points is stretched to 35, and the amplitude value of each sampling point after stretching is obtained by linear interpolation of the original signal. When d takes a positive value, take d=5 as an example, the signal waveform after time distortion is shown in Figure 3 (c), the signal on the left side of τ is stretched to the right, and the signal on the right side is compressed.
[0020] In the amplitude domain, the strategy of scaling and adding noise is used to enhance the dynamic range and randomness of the signal. First, the scaling factor Adjust the amplitude of the signal to simulate radar signals under different transmission power or receiving sensitivity conditions. Secondly, add designed noise to the processed signal, which may include Gaussian white noise, impulse noise, etc., to simulate the noise and interference in the actual electromagnetic environment. By adjusting the intensity and type of noise, the amplitude characteristics of the signal in the sample set are further enriched, and the robustness of the signal in the actual complex electromagnetic environment is enhanced. Figure 4-Figure 6 The signal waveforms before and after scaling, adding Gaussian white noise, and adding impulse noise are shown respectively. Figure 4-Figure 6 The signal length is 50, and the scaling factor .
[0021] In step S102, secondary data enhancement is performed based on Generative Adversarial Networks (GAN). The intelligent transformation of signals is achieved by constructing a generative adversarial network model, which consists of two parts: a generator and a discriminator. The generator deeply analyzes the radar signal after hybrid domain transformation, and at the same time, uses a multi-layer convolutional neural network to deeply mine the high-order characteristics of the signal and generate new signal samples based on these characteristics; the discriminator evaluates the authenticity of the generated signal through an adversarial learning mechanism, helps the generator retain key features, and guides the improvement of its ability to generate signals close to the real signal domain.
[0022] Next, the generative adversarial network model is optimized and trained. The Adam optimizer is used to fine-tune the model, and hyperparameters such as learning rate, batch size, and iteration cycle are carefully adjusted to accelerate model convergence and optimize its performance. Finally, the trained model is used to perform secondary data enhancement on the signal after primary data enhancement.
[0023] The structure of the generator is as follows Figure 7 As shown in the figure, the generator input is random noise of length 100. In order to enable the network to represent more features, after passing through the fully connected layer, the input random noise vector is mapped to a more complex representation, and a signal of length 512 is output. The data then enters the Relu activation function and then the deconvolution layer. The convolution kernel size is 2 and the padding method is "same". The generator has a total of 2 deconvolution layers. After the first deconvolution, batch normalization is performed, and then the desired output shape is obtained after 2 deconvolution layers. In the last layer of the generator, the Tanh activation function is used to limit the output signal to between [-1,1]. This signal is the generated fake sample.
[0024] The structure of the discriminator is as follows Figure 8 As shown in the figure, the discriminator is a neural network model composed of a fully connected layer and a convolutional layer. The discriminator accepts a signal of length 4096 as input, first enters the first convolutional layer, and then passes the output data through the dropout layer to reduce the connection between neurons and then "discard" some neurons to avoid overfitting. The discarded data enters the second convolutional layer. The convolution kernel size of the two convolutional layers is 2, and the padding methods are "same" and "valid" respectively. Then, the output is flattened into a one-dimensional vector, and finally passes through a fully connected layer to map the input signal into a representation. The last layer of the discriminator uses the sigmoid activation function to limit the output value between [0,1], indicating the probability that the input signal is a real sample.
[0025] In step S103, the quality of the generated signal is evaluated. The quantitative evaluation of the quality of the generated signal is achieved by calculating the mean and standard deviation of the correlation coefficient between the generated signal and the sample signal. The calculation formula of the correlation coefficient is as follows: ; in, To generate the signal, is the sample signal.
[0026] The mean of the correlation coefficients It can be calculated by the following formula: ; Standard deviation of the correlation coefficient It can be calculated by the following formula: .
[0027] In order to prove the superiority of the method provided above, the quality of the generated signal is evaluated from two perspectives. In a specific verification process, from the perspective of qualitative observation, the time domain waveform and frequency domain spectrum of the generated signal are basically consistent with those of the sample signal, and the waveform continuity and delicacy of the generated signal are relatively good. Therefore, it can be explained that the network has learned the time domain and frequency domain characteristics of the sample; from the perspective of quantitative comparison, the average correlation coefficient between the generated signal and the sample signal can reach 0.82, and the standard deviation of the correlation coefficient between the two is about 0.043. Therefore, it can be explained that the model has roughly learned the characteristics of the sample, and the learning effect is relatively stable. Fig. 9 As shown, compared with using a single mixed domain transformation or a generative adversarial network to achieve data enhancement, the signal generated by the method proposed in the present disclosure has a better degree of similarity with the sample signal and a higher quality.
[0028] The present disclosure further provides a two-stage radar data enhancement system, the system comprising: A primary data enhancement module is configured to obtain a radar signal to be enhanced and perform primary data enhancement on the radar signal to be enhanced; A secondary data enhancement module is configured to perform secondary data enhancement on the signal after the primary data enhancement based on a generative adversarial network; A quality assessment module is configured to perform quality assessment on the signal after the secondary data enhancement; The first-level data enhancement of the radar signal to be enhanced includes time domain transformation and amplitude domain transformation of the radar signal to be enhanced, wherein the time domain transformation includes time distortion of the radar signal to be enhanced, and the amplitude domain transformation includes scaling and adding noise to the signal after the time domain transformation.
[0029] The above embodiment provides a two-stage radar data enhancement system, and the division of the above hardware modules is used as an example to illustrate when performing signal enhancement. In practical applications, the above hardware modules can be replaced with other hardware modules with similar or identical functions as needed, or the functions of a certain hardware module can be assigned to different hardware modules. In addition, the above embodiment is the same as the implementation method of a two-stage radar data enhancement method, which will not be repeated here.
[0030] The embodiment of the present disclosure further provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to complete the steps of the method provided above.
[0031] The processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and LA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.
[0032] The memory may include one or more computer-readable media, which may be non-transitory. The memory may also include high-speed random access memory, and non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable medium in the memory is used to store at least one computer program, which is used to be executed by the processor to implement a two-stage radar data enhancement method provided in an embodiment of the present disclosure.
[0033] The embodiment of the present disclosure also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of a two-stage radar data enhancement method provided in the embodiment of the present disclosure can be completed.
[0034] The embodiment of the present disclosure also provides a computer program product, including a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of a two-stage radar data enhancement method provided in the embodiment of the present disclosure are implemented.
[0035] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A two-stage radar data enhancement method, characterized in that: include: Acquire the radar signal to be enhanced, and perform first-level data enhancement on the radar signal to be enhanced; Based on the generative adversarial network, the signal after the primary data enhancement is subjected to the secondary data enhancement; and, performing quality assessment on the signal after secondary data enhancement; The first-level data enhancement of the radar signal to be enhanced includes time domain transformation and amplitude domain transformation of the radar signal to be enhanced, wherein the time domain transformation includes time distortion of the radar signal to be enhanced, and the amplitude domain transformation includes scaling and adding noise to the signal after the time domain transformation.
2. A two-stage radar data enhancement method as claimed in claim 1, characterized in that: The time warping process includes: Processing the radar signal to be enhanced in sections; For each segment of the signal after segmentation processing, an anchor point is selected and a moving distance is randomly generated; if the moving distance is a negative value, the signal on the left side of the anchor point is compressed and the signal on the right side of the anchor point is stretched; if the moving distance is a positive value, the signal on the right side of the anchor point is compressed and the signal on the left side of the anchor point is stretched.
3. A two-stage radar data enhancement method as claimed in claim 2, characterized in that: The amplitude value of each sampling point after compression is obtained based on linear interpolation of the signal before compression.
4. A two-stage radar data enhancement method as claimed in claim 2, characterized in that: The amplitude value of each sampling point after stretching is obtained based on linear interpolation of the signal before stretching.
5. A two-stage radar data enhancement method as claimed in claim 1, characterized in that: The process of amplitude domain transformation includes: scaling the signal after time domain transformation by using a proportional scaling factor; adding noise to the scaled signal, wherein the noise includes Gaussian white noise and / or impulse noise.
6. A two-stage radar data enhancement method as claimed in claim 1, characterized in that: The quality assessment is performed using the mean and standard deviation of the correlation coefficient between the signal after secondary data enhancement and the radar signal to be enhanced.
7. A two-stage radar data enhancement system, characterized in that: include: A primary data enhancement module is configured to obtain a radar signal to be enhanced and perform primary data enhancement on the radar signal to be enhanced; A secondary data enhancement module is configured to perform secondary data enhancement on the signal after the primary data enhancement based on a generative adversarial network; A quality assessment module is configured to perform quality assessment on the signal after the secondary data enhancement; The first-level data enhancement of the radar signal to be enhanced includes time domain transformation and amplitude domain transformation of the radar signal to be enhanced, wherein the time domain transformation includes time distortion of the radar signal to be enhanced, and the amplitude domain transformation includes scaling and adding noise to the radar signal to be enhanced.
8. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to complete the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1 to 6.
10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
