Metamaterial aperture-encoding denoising imaging method based on unsupervised generative prior

Abstract

The application relates to a metamaterial aperture coding denoising imaging method based on unsupervised prior generation. The method comprises the following steps: constructing a reference signal matrix by using a distance-time delay-based signal deduction method; processing a pre-acquired high signal-to-noise ratio echo data set by using a generator network-based aperture coding imaging method to obtain a target scattering data set; constructing a metamaterial aperture coding imaging model by using echo data, a target scattering coefficient vector, a reference signal matrix and a measurement noise vector; taking the difference between a high signal-to-noise ratio target scattering coefficient distribution and a low signal-to-noise ratio target scattering coefficient distribution as a penalty term to construct a difference loss function, and solving and optimizing the metamaterial aperture coding imaging model according to the target scattering data set and the difference loss function by using a generative adversarial network to obtain optimal estimation of a target pattern. By using the method, high-precision reconstruction of a target scattering coefficient under low signal-to-noise ratio can be realized.

Description

Technical Field

[0001] This application relates to the field of radar denoising technology, and in particular to a metamaterial aperture coding denoising imaging method based on unsupervised generation of priors. Background Technology

[0002] Metamaterial aperture-coded imaging technology is a novel forward-looking radar imaging technique. It boasts a simple structure, ease of integration, and numerous advantages, including high resolution and all-weather capability. Unlike traditional radar imaging, which relies on relative motion to accumulate aperture for lateral high resolution, metamaterial aperture-coded imaging utilizes a metamaterial-coded antenna to randomly modulate electromagnetic waves, creating a spatiotemporally independent radiation field in the forward-looking direction. The principle involves a transmitter emitting a radar signal that illuminates a reflective metamaterial-coded antenna. The antenna's coding control module applies different amplitude, phase, or frequency random modulation factors to the incident electromagnetic wave at different times, ultimately forming a spatiotemporally independent radiation field in the forward-looking target area. After reflection from the target, the radar echo is received by a receiving antenna. The received signal is then sent to a signal processing module, where a reference signal matrix is ​​constructed by gridding the imaging area, and the target scattering coefficient is calculated using computational imaging. While metamaterial aperture-coded imaging technology offers numerous advantages, several challenges remain to be overcome before it can be successfully applied in practice. For example, in complex electromagnetic environments, strong scatterers outside the grid plane can interfere with the solution of the target scattering coefficient and exist in the form of noise in the imaging grid plane. Therefore, it is necessary to propose new solution models and imaging methods to improve the imaging quality under low signal-to-noise ratio.

[0003] However, current research on metamaterial aperture coding imaging technology under low signal-to-noise ratio (SNR) mainly falls into two categories. The first is a paper titled "Terahertz Aperture Coding Imaging Based on Convolutional Neural Networks," which proposes a terahertz aperture coding enhancement method using convolutional neural networks. This method implicitly models the imaging system by constructing an end-to-end neural network, leveraging the network's powerful inverse calculation capabilities and noise resistance to achieve target reconstruction under low SNR, thus addressing the problems of poor robustness and high computational complexity of traditional imaging algorithms. Results show that this method achieves implicit modeling of the imaging system, does not rely on complex prior knowledge, significantly reduces imaging computational complexity, and can reconstruct targets with different sparsity at different SNRs. Furthermore, compared to classic optimization iterative algorithms, this method can achieve higher resolution target reconstruction under low SNR. The second category is a novel depth-prior-based imaging method proposed in a foreign paper. Its scheme is as follows: first, the system is analyzed and modeled based on a coherent detection array; then, a depth-prior-based TCAI model is proposed within the generator's scope by modeling the target. Then, a Deep Alternating Minimization (Deep-Am) algorithm was designed, which solves the model by alternating projections between the target space and the latent variable space. The results show that this method can achieve high-resolution reconstruction of targets with different sparsity while improving robustness under compressed measurement. This method utilizes array reception and depth prior information, significantly reducing the system's acquisition and encoding time. Furthermore, the alternating projection method improves the accuracy of latent vector recovery. However, when dealing with the inverse problem of estimating the scattering coefficients from echo to target, although the end-to-end neural network-based method can achieve rapid recovery of the target from the echo signal, a well-trained network requires a large number of echo-image data pairs, which is difficult to obtain. Also, this network is only applicable to the specific imaging scene; when the scene changes, new data is needed to train a new network. In the second method, the extraction of depth prior information can achieve high-resolution reconstruction of targets with different sparsity under compressed measurement and low signal-to-noise ratio conditions, exhibiting good noise resistance and stability. In addition, this method overcomes scene constraints and can be applied to different imaging scenarios. However, it has certain limitations because it cannot recover targets outside the scope of the generated model. Furthermore, when the signal-to-noise ratio is high, generating priors can interfere with image quality, making its performance lower than that of regularization methods. Summary of the Invention

[0004] Therefore, it is necessary to provide a metamaterial aperture coding denoising imaging method based on unsupervised generative priors that can improve the reconstruction accuracy of the target scattering coefficient under low signal-to-noise ratio conditions, addressing the aforementioned technical problems.

[0005] A metamaterial aperture coding denoising imaging method based on unsupervised prior generation, the method comprising:

[0006] Construct a metamaterial aperture-coded imaging system; the metamaterial aperture-coded imaging system includes a control terminal, a coding control module, a metamaterial aperture-coded antenna, and a receiving antenna;

[0007] The imaging plane is divided into grids to obtain multiple target grid cells; the linear frequency modulated signal transmitted by the radar is acquired in advance, and the phase random modulation factor is loaded into the metamaterial coded antenna by the coding control module to modulate the phase of the linear frequency modulated signal, forming a spatiotemporally uncorrelated random radiation field in the forward-looking region; the reference signal of the target grid cell at any time is derived by the signal extrapolation method based on range time delay.

[0008] A reference signal matrix is ​​constructed based on the reference signals of all grid cells at multiple times; an aperture coding imaging method based on generator network is used to process the pre-acquired high signal-to-noise ratio echo dataset to obtain the target scattering dataset;

[0009] A metamaterial aperture-coded imaging model was constructed using echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector.

[0010] The difference between the target scattering coefficient distribution with high signal-to-noise ratio and that with low signal-to-noise ratio is used as a penalty term to construct a difference loss function. Based on the target scattering dataset and the difference loss function, a generative adversarial network is used to solve and optimize the metamaterial aperture coding imaging model to obtain the optimal estimate of the target pattern.

[0011] In one embodiment, the process of acquiring echo data includes: the control terminal drives the radar transmitter to transmit a radar signal that illuminates the surface of the metamaterial coded antenna; at the same time, the control terminal drives the coding control module to load different phase random modulation factors onto the radar signal at different times to perform random phase modulation on the incident radar signal, and finally form a detection signal with significant spatiotemporal uncorrelated characteristics in the target area. Multiple targets are selected, and after the detection signal is reflected by the target, the radar receiver obtains the corresponding echo data through multiple sampling and reception.

[0012] In one embodiment, the linear frequency modulated signal is

[0013]

[0014] Among them, f c A is the center frequency, K is the signal amplitude, t is the frequency modulation, and j is the symbol for the imaginary unit.

[0015] In one embodiment, after the linear frequency modulated signal is phase modulated by loading a phase random modulation factor onto the metamaterial coded antenna using an encoding control module, a spatiotemporally uncorrelated random radiation field is formed in the forward-looking region. The reference signal of the target grid cell at any time is derived using a range-delay-based signal extrapolation method, including:

[0016] After loading a phase random modulation factor into the metamaterial coded antenna using the coding control module to perform phase modulation on the linear frequency modulated signal, a spatiotemporally uncorrelated random radiation field is formed in the forward-looking region. The reference signal of the target grid cell at any given time is derived using a range-delay-based signal extrapolation method.

[0017]

[0018] Where Q represents the total number of modulation elements in the aperture-coded antenna. Is the q-th coding unit in t? m Phase modulation factor at time d Tx,q,n d represents the sum of the distance from the transmitter to the q-th coding unit and the distance from the q-th coding unit to the n-th imaging grid. n,Rx f represents the distance from the nth imaging grid to the receiver. c A is the center frequency, A is the signal amplitude, K is the frequency modulation frequency, t represents the time, j represents the symbol of the imaginary unit, and m represents the time sequence number.

[0019] In one embodiment, a reference signal matrix is ​​constructed based on reference signals of all grid cells at multiple times, including:

[0020] The reference signal matrix is ​​constructed based on the reference signals of all grid cells at multiple times.

[0021]

[0022] Where N represents the total number of target grid cells, and M represents the total number of time points.

[0023] In one embodiment, a metamaterial aperture-coded imaging model is constructed using echo data, a target scattering coefficient vector, a reference signal matrix, and a measurement noise vector, including:

[0024] A metamaterial aperture-coded imaging model is constructed using echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector.

[0025] S r =S·β+ω

[0026] Where, ω∈R M×1 It measures the noise vector, S r ∈R M×1This indicates that the radar receiver obtains the corresponding echo data after M samplings and receptions, β represents the target scattering coefficient vector, and S represents the reference signal matrix.

[0027] In one embodiment, the difference between the target scattering coefficient distribution with high signal-to-noise ratio and the target scattering coefficient distribution with low signal-to-noise ratio is used as a penalty term to construct a difference loss function, which further includes:

[0028] The difference between the target scattering coefficient distributions with high signal-to-noise ratio and those with low signal-to-noise ratio is used as a penalty term to construct the loss function.

[0029]

[0030] in, S represents the predicted echo of the metamaterial aperture-coded imaging model. r Indicates target echo data, This represents a convolutional neural network with θ parameters. This represents the difference between the target scattering coefficient distribution with a high signal-to-noise ratio and the target scattering coefficient distribution with a low signal-to-noise ratio on the grid plane. This represents the distribution of the target scattering coefficients at a high signal-to-noise ratio. β represents the distribution of target scattering coefficients with low signal-to-noise ratio. set Represents the target scattering dataset. denoted as the optimal estimate of the target pattern, λ represents the parameter balancing distortion and adversarial loss, and β0 is the pattern generated by random initialization.

[0031] In one embodiment, the generative adversarial network includes a generator network and a discriminator network; based on the target scattering dataset and the difference loss function, the generative adversarial network is used to solve and optimize the metamaterial aperture-coded imaging model to obtain the optimal estimate of the target pattern, including:

[0032] The network parameters of the generative adversarial network are iteratively updated based on the RMSProp algorithm, the target scattering dataset, and the pre-set generator network loss function and discriminator network loss function to obtain a trained generative adversarial network.

[0033] The process of solving and optimizing the metamaterial aperture coding imaging model using the difference loss function is analogous to the process of solving and optimizing the metamaterial aperture coding imaging model using a trained generative adversarial network, thereby obtaining the optimal estimate of the target pattern.

[0034] In one embodiment, the generator network loss function is pre-set to be...

[0035]

[0036] in, express Deep convolutional neural networks with specific parameters.

[0037] In one embodiment, the discriminator network loss function is pre-set to be...

[0038]

[0039] Where γ represents the gradient penalty coefficient. This indicates gradient calculation.

[0040] The aforementioned metamaterial aperture coding denoising imaging method based on unsupervised generative priors is further improved in this application by incorporating an unsupervised recovery learning criterion and using a generative adversarial network (GAN) to achieve the desired result. A discriminator network is used to mine the difference between the scattering coefficient distributions of targets with high signal-to-noise ratios (SNR) and those with low SNR, and a generator network is used to achieve high-precision reconstruction of the target scattering coefficients under low SNR conditions. Compared to end-to-end neural network methods based on supervised learning, this invention adopts an unsupervised approach, eliminating the need to construct paired training datasets, thus reducing workload. Furthermore, this method exhibits generalization ability, enabling imaging of targets outside the dataset. In addition, compared to traditional compressed sensing algorithms, this invention significantly improves the reconstruction accuracy of target scattering coefficients under low SNR conditions. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating a metamaterial aperture coding denoising imaging method based on unsupervised prior generation in one embodiment.

[0042] Figure 2 This is a structural block diagram of a metamaterial aperture-coded imaging system in one embodiment. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0044] In one embodiment, such as Figure 1 As shown, a metamaterial aperture coding denoising imaging method based on unsupervised prior generation is provided, including the following steps:

[0045] Step 102: Construct a metamaterial aperture coding imaging system; the metamaterial aperture coding imaging system includes a control terminal, a coding control module, a metamaterial aperture coding antenna, and a receiving antenna.

[0046] The control terminal is used to drive the coding control module to apply a phase random modulation factor to the metamaterial coding antenna; the coding control module is used to apply a phase random modulation factor to the metamaterial coding antenna; the metamaterial aperture coding antenna is used to perform phase modulation on the linear frequency modulated signal, and the receiving antenna is used for echo data.

[0047] Step 104: Grid division is performed on the imaging plane to obtain multiple target grid cells; the linear frequency modulated signal transmitted by the radar is acquired in advance, and the phase random modulation factor is loaded into the metamaterial coded antenna using the coding control module to modulate the phase of the linear frequency modulated signal, forming a spatiotemporally uncorrelated random radiation field in the forward-looking region; the reference signal of the target grid cell at any time is derived using the signal extrapolation method based on range time delay; a reference signal matrix is ​​constructed based on the reference signals of all grid cells at multiple times.

[0048] Step 106: The aperture coding imaging method based on generator network processes the pre-acquired high signal-to-noise ratio echo dataset to obtain the target scattering dataset; and constructs a metamaterial aperture coding imaging model using the echo data, target scattering coefficient vector, reference signal matrix and measurement noise vector.

[0049] The imaging plane is divided into N grids. Under good electromagnetic conditions, multiple targets are selected. After the detection signal is reflected by the target, the radar receiver obtains the corresponding echo data S through M sampling and reception. r i ∈R 1×M i = 1, 2, ..., L, using echo data to construct echo data S r set The echo data was processed using a generator network-based aperture coding imaging method to obtain the target scattering dataset β. set =[β 1 ,β 2 ,…,β L Under low signal-to-noise ratio conditions, utilizing Figure 2 The system's target echo S is shown. r And construct the reference signal matrix S using the deductive method, and use the dataset β set To improve echo S r The target scattering coefficient is recovered to achieve metamaterial aperture coding denoising imaging based on unsupervised generative priors.

[0050] Under high signal-to-noise ratio (SNR) conditions, the radar system transmits a linear frequency modulated pulse waveform, which is then modulated by a reflective metamaterial array antenna to illuminate a known target. After reflection from the target, the echo data of the known target is acquired by a receiving antenna to construct a high SNR echo dataset.

[0051] By loading a phase random modulation factor onto a metamaterial coded antenna using a metamaterial antenna driving module to modulate the phase of a linear frequency modulated signal, a reference signal for the target grid cell at any given time is obtained, thereby constructing a reference signal matrix. The reference signal matrix enables accurate estimation of the radiation field distribution. After obtaining the reference signal matrix, a metamaterial aperture coded imaging model is constructed using echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector, which enables high-resolution reconstruction of targets with different sparsity under compressed measurement.

[0052] Step 108: The difference between the target scattering coefficient distribution with high signal-to-noise ratio and the target scattering coefficient distribution with low signal-to-noise ratio is used as a penalty term to construct a difference loss function. Based on the target scattering dataset and the difference loss function, the metamaterial aperture coding imaging model is solved and optimized using a generative adversarial network to obtain the optimal estimate of the target pattern.

[0053] Based on the metamaterial aperture-coded imaging model, the difference between the target scattering coefficient distribution with high signal-to-noise ratio and that with low signal-to-noise ratio on the mesh plane is incorporated as a penalty term into the imaging model. Generative adversarial networks are then used to achieve the aforementioned objective. Under unsupervised learning, this significantly improves the quality of target scattering coefficient recovery under low signal-to-noise ratio conditions. Furthermore, this method is not constrained by the model and exhibits generalization performance.

[0054] In the aforementioned metamaterial aperture coding denoising imaging method based on unsupervised generative priors, this application utilizes a metamaterial aperture coding imaging system, integrates unsupervised recovery learning criteria, and employs a generative adversarial network to achieve this goal. A discriminator network is used to mine the difference between the scattering coefficient distributions of targets with high signal-to-noise ratios (SNR) and those with low SNR, and a generator network is used to achieve high-precision reconstruction of the target scattering coefficients under low SNR conditions. Compared to end-to-end neural network methods based on supervised learning, this invention adopts an unsupervised approach, eliminating the need to construct paired training datasets, thus reducing workload. Furthermore, this method exhibits generalization ability, enabling imaging of targets outside the dataset. Additionally, compared to traditional compressed sensing algorithms, this invention significantly improves the reconstruction accuracy of target scattering coefficients under low SNR conditions.

[0055] In one embodiment, the process of acquiring echo data includes: the control terminal drives the radar transmitter to transmit a radar signal that illuminates the surface of the metamaterial coded antenna; at the same time, the control terminal drives the coding control module to load different phase random modulation factors onto the radar signal at different times to perform random phase modulation on the incident radar signal, and finally form a detection signal with significant spatiotemporal uncorrelated characteristics in the target area. Multiple targets are selected, and after the detection signal is reflected by the target, the radar receiver obtains the corresponding echo data through multiple sampling and reception.

[0056] In a specific embodiment, such as Figure 2 As shown, the control terminal drives the radar transmitter to emit radar signals that illuminate the surface of the reflective metamaterial coded antenna. Simultaneously, the control terminal drives the coding control module to apply a phase random modulation factor to the metamaterial coded antenna, achieving random phase modulation of the incident radar signal, thus forming a spatially uncorrelated random radiation field in the imaging region. Furthermore, different phase random modulation factors are applied to the radar signal at different times, making the radiation field in the imaging region temporally uncorrelated. Ultimately, a detection signal with significant spatiotemporal uncorrelated characteristics is formed in the target region. Multiple targets are selected, and after the detection signal is reflected by the targets, the radar receiver obtains the corresponding echo data through multiple sampling and reception processes; the echo data is the target echo data.

[0057] In one embodiment, the linear frequency modulated signal is

[0058]

[0059] Among them, f c A is the center frequency, K is the signal amplitude, t is the frequency modulation, and j is the symbol for the imaginary unit.

[0060] In one embodiment, after the linear frequency modulated signal is phase modulated by loading a phase random modulation factor onto the metamaterial coded antenna using an encoding control module, a spatiotemporally uncorrelated random radiation field is formed in the forward-looking region. The reference signal of the target grid cell at any time is derived using a range-delay-based signal extrapolation method, including:

[0061] After loading a phase random modulation factor into the metamaterial coded antenna using the coding control module to perform phase modulation on the linear frequency modulated signal, a spatiotemporally uncorrelated random radiation field is formed in the forward-looking region. The reference signal of the target grid cell at any given time is derived using a range-delay-based signal extrapolation method.

[0062]

[0063] Where Q represents the total number of modulation elements in the aperture-coded antenna, ψ tm,q Is the q-th coding unit in t? m Phase modulation factor at time d Tx,q,n d represents the sum of the distance from the transmitter to the q-th coding unit and the distance from the q-th coding unit to the n-th imaging grid. n,Rx f represents the distance from the nth imaging grid to the receiver. c A is the center frequency, A is the signal amplitude, K is the frequency modulation frequency, t represents the time, j represents the symbol of the imaginary unit, and m represents the time sequence number.

[0064] In one embodiment, a reference signal matrix is ​​constructed based on reference signals of all grid cells at multiple times, including:

[0065] The reference signal matrix is ​​constructed based on the reference signals of all grid cells at multiple times.

[0066]

[0067] Where N represents the total number of target grid cells, and M represents the total number of time points.

[0068] In one embodiment, a metamaterial aperture-coded imaging model is constructed using echo data, a target scattering coefficient vector, a reference signal matrix, and a measurement noise vector, including:

[0069] A metamaterial aperture-coded imaging model is constructed using echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector.

[0070] S r =S·β+ω

[0071] Where, ω∈R M×1 It measures the noise vector, S r ∈R M×1 This indicates that the radar receiver obtains the corresponding echo data after M samplings and receptions, β represents the target scattering coefficient vector, and S represents the reference signal matrix.

[0072] In one embodiment, the difference between the target scattering coefficient distribution with high signal-to-noise ratio and the target scattering coefficient distribution with low signal-to-noise ratio is used as a penalty term to construct a difference loss function, which further includes:

[0073] The difference between the target scattering coefficient distributions with high signal-to-noise ratio and those with low signal-to-noise ratio is used as a penalty term to construct the loss function.

[0074]

[0075] in, S represents the predicted echo of the metamaterial aperture-coded imaging model. r Indicates target echo data, This represents a convolutional neural network with θ parameters. This represents the difference between the target scattering coefficient distribution with a high signal-to-noise ratio and the target scattering coefficient distribution with a low signal-to-noise ratio on the grid plane. This represents the distribution of the target scattering coefficients at a high signal-to-noise ratio. β represents the distribution of target scattering coefficients with low signal-to-noise ratio. set Represents the target scattering dataset. denoted as the optimal estimate of the target pattern, λ represents the parameter balancing distortion and adversarial loss, and β0 is the pattern generated by random initialization.

[0076] In one embodiment, the generative adversarial network includes a generator network and a discriminator network; based on the target scattering dataset and the difference loss function, the generative adversarial network is used to solve and optimize the metamaterial aperture-coded imaging model to obtain the optimal estimate of the target pattern, including:

[0077] The network parameters of the generative adversarial network are iteratively updated based on the RMSProp algorithm, the target scattering dataset, and the pre-set generator network loss function and discriminator network loss function to obtain a trained generative adversarial network.

[0078] The process of solving and optimizing the metamaterial aperture coding imaging model using the difference loss function is analogous to the process of solving and optimizing the metamaterial aperture coding imaging model using a trained generative adversarial network, thereby obtaining the optimal estimate of the target pattern.

[0079] In a specific embodiment, the process of solving and optimizing the metamaterial aperture-coded imaging model using the difference loss function involves inputting a random image β0, which is then processed by a network. The predicted pattern is obtained after processing. Predicted echoes obtained using metamaterial aperture-coded imaging models The difference loss function is used to update the generator network parameters θ through an optimization algorithm. After multiple iterations, the optimal estimate of the target pattern is obtained.

[0080] In one embodiment, the generator network loss function is pre-set to be...

[0081]

[0082] in, express Deep convolutional neural networks with specific parameters.

[0083] In a specific embodiment, the process of solving and optimizing the difference loss function for the metamaterial aperture-coded imaging model is similar to the training process of a Generative Adversarial Network (GAN). Therefore, a GAN can be used to achieve the above objective. The loss functions of the generator network and the discriminator network are updated using the RMSProp algorithm to update the parameters of the generator network and the discriminator network. After multiple iterations, the optimal estimate of the target image is achieved. The process of updating the generator network parameters and discriminator network parameters using the RMSProp algorithm is existing technology and will not be elaborated upon in this application. The generator network-based aperture coding imaging method for high signal-to-noise ratio echo data can be implemented using a generator network when λ in the generator network loss function formula is zero, i.e., aperture coding imaging without prior generation conditions.

[0084] In one embodiment, the discriminator network loss function is pre-set to be...

[0085]

[0086] Where γ represents the gradient penalty coefficient, and ▽ represents the gradient calculation.

[0087] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0088] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for denoising imaging of metamaterial aperture coding based on unsupervised generation prior, characterized in that The method includes: Constructing a metamaterial aperture encoding imaging system; the metamaterial aperture encoding imaging system includes a control terminal, an encoding control module, a metamaterial aperture encoding antenna, and a receiving antenna; Dividing the imaging plane into grids to obtain multiple target grid units; pre-acquiring the linearly chirped signal transmitted by the radar, using the encoding control module to load a phase random modulation factor onto the metamaterial encoding antenna to perform phase modulation on the linearly chirped signal, forming a spatio-temporally uncorrelated random radiation field in the forward-looking area, and using a signal deduction method based on distance-time delay to deduce the reference signal of the target grid unit at any moment; Constructing a reference signal matrix based on the reference signals of all grid units at multiple moments; processing the pre-acquired high signal-to-noise ratio echo dataset using the aperture encoding imaging method based on the generator network to obtain a target scattering dataset; Constructing a metamaterial aperture encoding imaging model using the echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector; Taking the difference between the high signal-to-noise ratio target scattering coefficient distribution and the low signal-to-noise ratio target scattering coefficient distribution as a penalty term to construct a difference loss function, and using the target scattering dataset and the difference loss function and the generative adversarial network to solve and optimize the metamaterial aperture encoding imaging model to obtain the optimal estimate of the target pattern; Using the encoding control module to load a phase random modulation factor onto the metamaterial encoding antenna to perform phase modulation on the linearly chirped signal, forming a spatio-temporally uncorrelated random radiation field in the forward-looking area, and using a signal deduction method based on distance-time delay to deduce the reference signal of the target grid unit at any moment, including: Among them, represents the total number of modulation units of the aperture-coded antenna, is the th phase modulation factor of the coding unit at moment, represents the sum of the distance from the transmitter to the th coding unit and the distance from the th coding unit to the th imaging grid, represents the distance from the th imaging grid to the receiver, is the center frequency, is the signal amplitude, is the frequency modulation rate, represents the moment, represents the imaginary unit symbol, and m represents the serial number of the moment.

2. The method according to claim 1, wherein Using the encoding control module to load a phase random modulation factor onto the metamaterial encoding antenna to perform phase modulation on the linearly chirped signal, forming a spatio-temporally uncorrelated random radiation field in the forward-looking area, and using a signal deduction method based on distance-time delay to deduce the reference signal of the target grid unit at any moment as The process of acquiring the echo data includes:

3. The method according to claim 1, wherein The control terminal drives the radar transmitter to emit a radar signal to irradiate the surface of the metamaterial encoding antenna, and at the same time, the control terminal drives the encoding control module to load different phase random modulation factors onto the metamaterial encoding antenna at different moments to perform random phase modulation on the incident radar signal, finally forming a detection signal with significant spatio-temporal non-correlation characteristics in the target area. Multiple targets are selected, and after the detection signal is reflected by the target, the radar receiver obtains the corresponding echo data through multiple samplings and receptions. Among them, is the center frequency, is the signal amplitude, is the frequency modulation rate, represents the time, represents the imaginary unit symbol.

4. The method according to claim 1, wherein The linearly chirped signal is Constructing a reference signal matrix based on the reference signals of all grid units at multiple moments, including: Among them, represents the total number of target grid cells, M represents the total number of time instances.

5. The method according to claim 1, wherein Constructing a reference signal matrix based on the reference signals of all grid units at multiple moments as Constructing a metamaterial aperture encoding imaging model using the echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector, including: Among them, is the measurement noise vector, represents that the radar receiver obtains the corresponding echo data after M samplings and receptions, represents the target scattering coefficient vector, represents the reference signal matrix.

6. The method according to claim 1, wherein Constructing a metamaterial aperture encoding imaging model using the echo data, target scattering coefficient vector, reference signal matrix, and measurement noise vector as Taking the difference between the high signal-to-noise ratio target scattering coefficient distribution and the low signal-to-noise ratio target scattering coefficient distribution as a penalty term to construct a difference loss function, further including: Construct a loss function by taking the difference between the target scattering coefficient distribution with high signal-to-noise ratio and the target scattering coefficient distribution with low signal-to-noise ratio as the penalty term, as follows Among them, represents the predicted echo of the metamaterial aperture encoding imaging model, represents the target echo data, represents the convolutional neural network under the parameter represents the difference between the target scattering coefficient distribution with high signal-to-noise ratio and the target scattering coefficient distribution with low signal-to-noise ratio on the grid plane, represents the target scattering coefficient distribution with high signal-to-noise ratio, represents the target scattering coefficient distribution with low signal-to-noise ratio, represents the target scattering data set, represents the optimal estimate of the target pattern, represents the parameter for balancing distortion and adversarial loss, is a pattern randomly initialized and generated.

7. The method according to claim 6, wherein The generative adversarial network includes a generator network and a discriminator network; solving and optimizing the metamaterial aperture encoding imaging model using the generative adversarial network according to the target scattering data set and the difference loss function, to obtain the optimal estimate of the target pattern, including: Iteratively update the network parameters of the generative adversarial network according to the RMSProp algorithm, the target scattering data set, and the pre-set generator network loss function and discriminator network loss function, to obtain a trained generative adversarial network; Compare the process of solving and optimizing the metamaterial aperture encoding imaging model using the difference loss function with the process of solving and optimizing the metamaterial aperture encoding imaging model using the trained generative adversarial network, to obtain the optimal estimate of the target pattern.

8. The method according to claim 7, wherein The pre-set generator network loss function is Among them, denotes the deep convolutional neural network under the parameter.

9. The method according to claim 7, wherein The pre-set discriminator network loss function is Among them, represents the gradient penalty coefficient, represents the gradient calculation.

Images