ISAR sparse imaging method, device and equipment based on DA-ISTA network

Abstract

The application relates to an ISAR sparse imaging method, device and equipment based on a DA-ISTA network. An iterative threshold convergence algorithm (ISTA) is used to construct a deep neural network for ISAR sparse imaging. In the ISTA, the iteration process is tiled into a multi-layer network structure, and the linear sparse transformation in the ISTA algorithm is replaced by a nonlinear convolution operation in each layer of the network structure. In the DA-ISTA network, a large number of iterations are not required as in the ISTA algorithm. Only a small number of network structure layers are required to process ISAR sparse aperture echo data, and a high-quality ISAR imaging result can be obtained. The method improves the imaging quality and the calculation efficiency, and makes the DA-ISTA network interpretable.

Description

Technical Field

[0001] This application relates to the field of radar signal processing technology, and in particular to an ISAR sparse imaging method, apparatus and device based on DA-ISTA network. Background Technology

[0002] Inverse Synthetic Aperture Radar (ISAR) can acquire high-resolution radar images of moving targets in all weather conditions and is widely used in military and civilian fields. However, sparse aperture effects can occur when strong environmental noise interference disrupts echo integrity, multi-function radar switching causes partial loss of received echoes, and limited radar system data storage and processing capabilities lead to the active discarding of some data. For sparse aperture echoes, traditional Range Doppler (RD) algorithms suffer from severe defocusing and cannot effectively recover the ISAR image.

[0003] Compressive sensing (CS) is an important method for sparse aperture ISAR imaging. Due to the sparsity of ISAR images in the image domain, CS methods can be used to reconstruct ISAR images. Currently, commonly used CS methods can be broadly classified into three categories: greedy algorithms, convex optimization methods, and sparse Bayesian learning methods. Common greedy algorithms include Matching Pursuit (MP) and Orthogonal Matching Pursuit (OMP). These algorithms are computationally simple, but generally require many iterations to converge. Sparse Bayesian learning methods solve the sparse aperture ISAR imaging problem within a probabilistic framework. However, this method suffers from high computational complexity and difficulty in quantitatively analyzing its convergence performance. The Iterative Shrinkage Thresholding Algorithm (ISTA) is a convex optimization method that updates results through soft thresholding operations. It has low complexity and a simple structure, and has been widely studied and applied. However, the ISTA algorithm converges slowly, and the image quality is greatly affected by the hyperparameters, requiring manual adjustment. Summary of the Invention

[0004] Therefore, it is necessary to provide an ISAR sparse imaging method, apparatus, and device based on DA-ISTA network that can improve computational efficiency and achieve better imaging results, in order to address the aforementioned technical problems.

[0005] A sparse imaging method for ISAR based on DA-ISTA network, the method comprising:

[0006] Acquire ISAR sparse aperture echo data to be imaged;

[0007] The ISAR sparse aperture echo data is processed to obtain the corresponding high-resolution one-dimensional range image;

[0008] The high-resolution one-dimensional range image is input into the trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0009] In one embodiment, the data processing procedure for each layer of the network structure is represented as follows:

[0010]

[0011] In the above formula, the superscript k represents the k-th layer of the network structure, ρ (k) The iteration step size parameter is represented by y, and the initial input data is represented by x. (k-1) This represents the output data of the (k-1)th layer network structure, soft(,α) (k) ) represents the soft thresholding function, where α (k) Indicates the threshold parameter, x (k) This represents the output data of the k-th layer, which is the current layer of the network structure. as well as These represent different convolutional units in the current layer's network structure.

[0012] In one embodiment, the convolutional unit as well as Each includes a linear convolution operator;

[0013] The convolutional unit as well as Each includes two linear convolution operators and an activation layer embedded between the two linear convolution operators, and the convolution unit Convolutional unit The left inverse transform.

[0014] In one embodiment, after the high-resolution one-dimensional distance image is input into the trained DA-ISTA network, the high-resolution one-dimensional distance image is processed by the least squares estimation method before being input into the first layer of the neural network, and the processing result is input into the first layer of the neural network structure.

[0015] In one embodiment, the loss function is expressed as follows when training the DA-ISTA network:

[0016]

[0017] In the above formula, This represents the training imaging structure obtained after processing the training sample data via the DA-ISTA network, x n γ represents the true label corresponding to the training sample data, and γ represents the regularization parameter.

[0018] In one embodiment, when training the DA-ISTA network, adaptive matrix estimation is used to optimize the loss function until the loss function converges, thus obtaining the trained DA-ISTA network.

[0019] In one embodiment, the trainable parameters in the DA-ISTA network include the parameters in each convolutional unit of each layer of the network structure, the iteration stride parameter, and the threshold parameter in the soft thresholding function.

[0020] In one embodiment, the DA-ISTA network includes a 3-layer network structure.

[0021] This application also provides an ISAR sparse imaging device based on a DA-ISTA network, the device comprising:

[0022] The echo data acquisition module is used to acquire ISAR sparse aperture echo data to be imaged;

[0023] A high-resolution one-dimensional range image acquisition module is used to process the ISAR sparse aperture echo data to obtain the corresponding high-resolution one-dimensional range image.

[0024] The ISAR sparse imaging module is used to input the high-resolution one-dimensional range image into a trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0025] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0026] Acquire ISAR sparse aperture echo data to be imaged;

[0027] The ISAR sparse aperture echo data is processed to obtain the corresponding high-resolution one-dimensional range image;

[0028] The high-resolution one-dimensional range image is input into the trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0029] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0030] Acquire ISAR sparse aperture echo data to be imaged;

[0031] The ISAR sparse aperture echo data is processed to obtain the corresponding high-resolution one-dimensional range image;

[0032] The high-resolution one-dimensional range image is input into the trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0033] The aforementioned ISAR sparse imaging method, apparatus, and device based on the DA-ISTA network constructs a deep neural network for ISAR sparse imaging based on the Iterative Shrinkage Thresholding Algorithm (ISTA). This network flattens the multiple iterations of the ISTA algorithm across a multi-layered network structure, and replaces the linear sparse transformation in the ISTA algorithm with nonlinear convolution operations in each layer. This eliminates the need for extensive iterations in the DA-ISTA network compared to the ISTA algorithm, requiring only a few network layers to process the ISAR sparse aperture echo data to obtain high-quality ISAR imaging results. This method improves imaging quality and computational efficiency while also making the DA-ISTA network interpretable. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating an ISAR sparse imaging method based on a DA-ISTA network in one embodiment.

[0035] Figure 2 This is a schematic diagram of the structure of a DA-ISTA network in one embodiment;

[0036] Figure 3 This is a schematic diagram illustrating the geometric relationship between the target and the radar in one embodiment;

[0037] Figure 4 This is a schematic diagram of complete radar echo data used in a simulation experiment.

[0038] Figure 5 This is a high-resolution ISAR image used in a simulation experiment.

[0039] Figure 6 This is a schematic diagram of echo data with 25% sparsity in a simulation experiment;

[0040] Figure 7 This is a schematic diagram showing the results of a simulation experiment using different ISAR imaging methods. Figure 7 (a) Figure 7 (b) Figure 7 (c) Figure 7 (d) Schematic diagrams of the imaging results of the four methods: RD, ISTA, ADMM, and DA-ISTA;

[0041] Figure 8 This is a structural block diagram of an ISAR sparse imaging device based on a DA-ISTA network in one embodiment;

[0042] Figure 9 This is an internal structural diagram of a computer device 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] To address the shortcomings of existing technologies, such as the need for numerous iterative steps, computational complexity, and slow convergence in compressive sensing methods for ISAR sparse aperture imaging, while artificial neural network methods improve image reconstruction accuracy and accelerate convergence, they cannot process complex radar signals. In one embodiment, for example... Figure 1 As shown, an ISAR sparse imaging method based on a DA-ISTA network is provided, including the following steps:

[0045] Step S100: Acquire the ISAR sparse aperture echo data to be imaged;

[0046] Step S110: Process the ISAR sparse aperture echo data to obtain the corresponding high-resolution one-dimensional range image.

[0047] Step S120: Input the high-resolution one-dimensional range image into the trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through the sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0048] In this embodiment, addressing the problems of poor generalization, high computational cost, and inapplicability to complex signal processing in existing ISAR sparse aperture imaging methods, a sparse aperture ISAR imaging method based on the deep enhancement-iterative shrinkage thresholding algorithm (DA-ISTA) network is proposed. This method constructs the iterative steps of ISTA as hidden layers of the network and adds nonlinear layers to effectively recover neglected high-frequency components. Compared with traditional model-driven algorithms, this method avoids the manual parameter tuning process, achieves faster convergence, higher imaging quality, and better generalization ability for data with large feature differences.

[0049] In step S100, the radar echo data to be imaged is sparse aperture echo data that has caused the sparse aperture effect under circumstances such as strong environmental noise interference causing the echo integrity to be destroyed, the multi-function radar function switching process causing the loss of some received echoes, and the radar system's data storage and processing capabilities being limited, resulting in the active discarding of some data.

[0050] In step S110, before inputting the sparse aperture echo data into the DA-ISTA network, motion compensation and range compression are performed on it to obtain the corresponding high-resolution one-dimensional range image, which is then input into the DA-ISTA network.

[0051] In this embodiment, the input data to the DA-ISTA network is in the data domain, while the output results are in the image domain. Since the sparsity of the azimuth dimension in ISAR imaging is more practical, this method only focuses on the sparse data of the azimuth dimension.

[0052] In step S120, the structure of the DA-ISTA network is as follows: Figure 2 As shown, the data processing procedure for each layer of the network structure is represented as follows:

[0053]

[0054] In formula (1), the superscript k denotes the k-th layer network structure, ρ (k) The iteration step size parameter is represented by y, and the initial input data is represented by x. (k-1) This represents the output data of the (k-1)th layer network structure, soft(,α) (k) ) represents the soft thresholding function, where α (k) Indicates the threshold parameter, x (k) This represents the output data of the k-th layer, which is the current layer of the network structure. as well as These represent different convolutional units in the current layer's network structure.

[0055] Furthermore, convolutional units as well as Each includes a linear convolution operator, and the convolution unit... as well as Each includes two linear convolution operators and an activation layer embedded between the two linear convolution operators, and the convolution units... Convolutional unit The left inverse transform.

[0056] In this embodiment, after the high-resolution one-dimensional range image is input into the trained DA-ISTA network, a least-squares estimation method is used to process the high-resolution one-dimensional range image before it is input into the first layer of the neural network. The processing result is then input into the first layer of the neural network structure. The least-squares estimation uses the following formula:

[0057]

[0058] In formula (2), y n Represents a high-resolution one-dimensional range image. This represents the initial value input to the first layer of the network structure.

[0059] In this embodiment, the DA-ISTA network is a CNN convolutional neural network.

[0060] In this embodiment, when training the DA-ISTA network, the training dataset Y = [y1, y2, ..., y] is constructed by processing existing ISAR sparse aperture echo data to obtain corresponding high-resolution one-dimensional range images. N], and construct a labeled dataset X = [x1, x2, ..., x] from the imaging results corresponding to each ISAR sparse aperture echo data. N In the DA-ISTA network, the initial values ​​of each trainable parameter are randomly generated.

[0061] Because ISAR data is complex, CNN neural networks currently only support real number processing; complex numbers need to be processed as real numbers before being input into the network. Complex multiplication can be represented as:

[0062]

[0063] In formula (3), Re(·) and Im(·) are the operators for extracting the real and imaginary parts, respectively. Before the data is input into the network, the complex data and the measurement matrix are first decomposed, and then reassembled after the network output to obtain the final recovery result.

[0064] To increase the sparsity of ISAR images and reduce the impact of noise, the loss function is defined as follows:

[0065]

[0066] In formula (4), This represents the training imaging result obtained after processing the training sample data through the DA-ISTA network, x n This represents the true label corresponding to the training sample data, and γ represents the regularization parameter.

[0067] In one embodiment, the regularization parameter γ can take the value 0.01.

[0068] The loss function described above comprises two parts: the first part calculates the error between the image reconstructed by the network and the true label, and the second part represents... and The error between them, to ensure the inverse matrix The assumption.

[0069] In this embodiment, when training the DA-ISTA network, the loss function is optimized using Adaptive Moment Estimation (Adam) until the loss function converges, thus obtaining the trained DA-ISTA network.

[0070] In this embodiment, the trainable parameters in the DA-ISTA network include the parameters in each convolutional unit of each network layer, the iteration stride parameter, and the threshold parameter in the soft thresholding function.

[0071] In this embodiment, to avoid overfitting, a 3-layer network structure is set in the DA-ISTA network.

[0072] In fact, the focus of this method is more on the construction of the DA-ISTA network. How to construct the structure of the neural network based on the ISTA algorithm is the focus of this paper, that is, how to obtain formula (1), because its construction and derivation process further demonstrates that the construction of the DA-ISTA network has a solid basis and is reasonable. In the following text, the derivation process is explained in three parts, including the ISAR imaging model, ISAR sparse imaging based on ISTA, and ISAR sparse imaging based on DA-ISTA.

[0073] First, in the ISAR imaging model section, it is assumed that the radar transmits a linear frequency modulated signal as follows:

[0074]

[0075] In formula (5), T represents the pulse width, γ represents the modulation frequency, A represents the signal amplitude, and τ = tmod(T) r ) represents fast time, t represents slow time, and f represents fast time. c Represents the carrier frequency, and rect(·) represents a rectangular window.

[0076] Next, assuming the translational motion is fully compensated, a maneuvering target with K strong scattering centers rotates around center O in a two-dimensional plane, where the geometric relationship between the radar and the target is referenced. Figure 3 In CPI containing M-pulses, the instantaneous rotation angle of the target is defined as Δθ(t), and under the far-field assumption, the scattering center (x) k ,y k The instantaneous distance to the radar is:

[0077]

[0078] In formula (5), ω represents the rotation rate, which is an acceptable approximation because the rotation angle can be very small during a short CPI period, and the Doppler frequency shift remains almost constant. Therefore, the complex envelope of the echo signal can be written as:

[0079]

[0080] In formula (7), c represents the speed of light, and A k This represents the backscattering amplitude, which can be considered stationary during CPI. Assuming the scatterer migration through the resolution cell (MTRC) is absent or corrected, the received signal is obtained after range compression through matched filtering and ignoring constant terms:

[0081]

[0082] When τ=2(R0+y k When ) / c, further assuming that the corresponding range cell contains Q≤K scattering centers at different span positions, the signal in the range cell is obtained by combining the following isophase terms:

[0083]

[0084] In formula (10), and f q =f c x q ω / c represent the complex amplitude and Doppler frequency of the q-th scattering center, respectively. Considering unavoidable measurement noise and background clutter, the signal corresponding to the range cell in equation (9) can be rewritten as:

[0085]

[0086] In formula (10), vector e represents the synthesized additive noise in the distance cell. The time series is represented as t = [1:M]. T T r , [·] T This represents the transpose of a vector or matrix. For super-resolution, the pulse repetition frequency f r Divided into N parts (much larger than M), Doppler frequency domain f D =[1:N] T Δf-f r / 2, Δf=f r / N can be considered as the Doppler frequency domain resolution. Therefore, formula (10) can be rewritten as:

[0087] s r =Ax+e (11)

[0088] Next, after obtaining the echo signal model, when performing ISAR sparse imaging based on ISTA, since the maximum number of scattering centers projected in the vertical direction of radar line of sight is usually much smaller than the number of span cells, the ISAR signal can be considered sparse to some extent. This essentially provides a basis for applying sparse recovery theory to the construction of high-resolution ISAR images.

[0089] In this case, the formula (11) for the undetermined linear system can be solved by introducing the following constraint sparsity enhancement objective function:

[0090]

[0091] In formula (12), ε=||e||2 represents the noise power, and Φ represents the sparse transformation. This means that after range compression acquisition of the echo signal of the target scene, as long as the number of scattering centers is small enough or the number of pulses is large enough, a super-resolution ISAR image can be generated by simply solving the sparse recovery problem for each range cell. The regularization minimization under the CS theorem can be expressed as follows: If the one-dimensional range image and the ISAR image are expanded in the azimuth dimension, then formula (12) can be expressed as:

[0092]

[0093] In formula (13), λ represents the regularization parameter.

[0094] Assuming range compression has been performed and MTRC does not exist, the cross-range compression algorithm for ISAR imaging can be solved using ISTA, with the following iterative steps:

[0095]

[0096] In formula (14), k represents the number of iterations of ISTA, ρ represents the iteration step size, and soft(·,α) (k) ) represents the soft thresholding function, and α represents the threshold. soft(·,α) (k) ) is defined as:

[0097] soft(x i ,α (k) )=(x i / |x i |)·max(|x i |-α,0) (15)

[0098] By iterating through formulas (14) and (15), conventional ISTA can obtain satisfactory imaging results. However, it requires a large amount of computation and predefined parameters (e.g., threshold α and step size ρ) to obtain satisfactory results, which are not easily determined manually.

[0099] Finally, based on the above two parts, the structure of the DA-ISTA network is constructed. From formula (13), it is reasonable to assume that:

[0100] x = r + w + e (16)

[0101] In formula (16), e represents the noise term, and w represents the missing high-frequency component in r, which can be obtained through linear transformation, expressed as: Define linear transformation in There are 32 filters, each with a size of 3×3. There is one filter, and each filter has a size of 3×3.

[0102] To map ISTA into a convolutional neural network and improve the generalization ability of the CNN, a nonlinear transformation is employed. Instead of the sparse transformation Φ, equation (14) can be reformulated as follows:

[0103]

[0104] Due to the existence of the following approximation relationship:

[0105]

[0106] In formula (18), the scalar t is only related to Relatedly, formula (17) can be reformulated as:

[0107]

[0108] In formula (19), α = t·λ. Therefore, the closed-form solution is as follows:

[0109]

[0110] Through modeling in It consists of two linear convolution operators, each corresponding to 32 filters, each 3×3 in size, and includes an embedded activation layer, namely an LReLU layer.

[0111] Next, in order to restore the signal, it is necessary to obtain The left inverse transformation makes It is the identity matrix. At this point, we can introduce... get:

[0112]

[0113] The single-step iterative steps of the recovery process described above are strictly mapped to a single hidden layer of the DA-ISTA network, and multiple hierarchical structures are connected to the entire network, ensuring the theoretical feasibility of the network. Through the above deduction, the forward propagation process of the DA-ISTA network can be obtained, which is Equation (1).

[0114] Meanwhile, the DA-ISTA network solves the problem of manual parameter tuning by automatically learning parameters.

[0115] In this paper, simulation experiments of ISAR sparse imaging are also conducted using the DA-ISTA network proposed in this paper, and comparative experiments are also conducted using traditional methods.

[0116] Dataset preparation and DA-ISTA network setup: To match the size of the actual measured ISAR data, the simulation scene size is set to N. r ×N a , where N r =256 is the distance dimension, N a =256 is the azimuth dimension.

[0117] In accordance with Figure 2 The network structure shown illustrates the construction of the DA-ISTA network. During training, 200 scenarios with random points are generated for training, and another 50 scenarios with specific points are generated for testing. The carrier frequency, bandwidth, pulse width, and pulse repetition frequency of the simulated radar signal are set to 10 GHz, 600 MHz, 100 μs, and 200 Hz, respectively. During DA-ISTA training, the number of iterations is set to 3, and the mini-batch size is set to 64. The Adam optimizer with a learning rate of 0.0001 is used for training.

[0118] Meanwhile, the quantification standards used to evaluate the results include: runtime, normalized mean square error (NMSE), and image entropy (IE). The definition of NMSE is as follows:

[0119]

[0120] In formula (22), N represents the total number of training samples. Indicates the imaging result, x n This represents the corresponding real label.

[0121] Image focusing characteristics (IE) reflect the focusing properties of an ISAR image. The IE is defined as follows:

[0122]

[0123] In formula (23), x i yes elements,

[0124] The experimental equipment used was an Intel(R) Core(TM) i7-10510U CPU@1.80GHz.

[0125] like Figure 4 The image shows the complete data echo. Figure 5 The image shown is a high-resolution ISAR image. This experiment considers echo data with a random downsampling rate of 25%, such as... Figure 6 As shown.

[0126] Figure 7 middle, Figure 7 (a) Figure 7 (b) Figure 7 (c) Figure 7 (d) The imaging results of the four methods, RD, ISTA, ADMM and DA-ISTA, are presented respectively, and the corresponding quantitative results are shown in Table 1.

[0127] Table 1 Evaluation Indicators

[0128]

[0129] Experimental results show that, compared with traditional CS-based methods, the proposed method achieves good reconstruction results while maintaining high computational efficiency. Furthermore, as an improved version of ISTA, DA-ISTA's faster convergence and better reconstruction performance have demonstrated its superiority over the other two traditional algorithms (ISTA and ADMM).

[0130] In the aforementioned ISAR sparse imaging method based on the DA-ISTA network, the iterative steps of ISTA are constructed as hidden layers of the network, and nonlinear layers are added to effectively recover neglected high-frequency components. Compared with traditional model-driven algorithms, this method avoids the manual parameter tuning process, achieves faster convergence, higher imaging quality, and better generalization ability for data with large feature differences. During the inference construction of the DA-ISTA network, the modeling process takes into account the lost high-frequency components, improving the model's robustness. In terms of parameter setting, this method combines the advantages of convolutional neural networks and traditional ISTA, enabling end-to-end automatic learning of key parameters. In terms of application scenarios, this method uses nonlinear convolution operations instead of linear sparse transformations, making the proposed method more flexible and better suited for scenarios such as non-cooperative ISAR imaging of targets with undersampled or incomplete data.

[0131] 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.

[0132] In one embodiment, such as Figure 8 As shown, an ISAR sparse imaging device based on a DA-ISTA network is provided, including: an echo data acquisition module 200, a high-resolution one-dimensional range image acquisition module 210, and an ISAR sparse imaging module 220, wherein:

[0133] The echo data acquisition module 200 is used to acquire ISAR sparse aperture echo data to be imaged;

[0134] The high-resolution one-dimensional range image acquisition module 210 is used to process the ISAR sparse aperture echo data to obtain the corresponding high-resolution one-dimensional range image.

[0135] The ISAR sparse imaging module 220 is used to input the high-resolution one-dimensional range image into a trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0136] Specific limitations regarding the DA-ISTA network-based ISAR sparse imaging device can be found in the limitations of the DA-ISTA network-based ISAR sparse imaging method described above, and will not be repeated here. Each module in the aforementioned DA-ISTA network-based ISAR sparse imaging device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0137] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an ISAR sparse imaging method based on a DA-ISTA network. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0138] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0139] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0140] Acquire ISAR sparse aperture echo data to be imaged;

[0141] The ISAR sparse aperture echo data is processed to obtain the corresponding high-resolution one-dimensional range image;

[0142] The high-resolution one-dimensional range image is input into the trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0143] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0144] Acquire ISAR sparse aperture echo data to be imaged;

[0145] The ISAR sparse aperture echo data is processed to obtain the corresponding high-resolution one-dimensional range image;

[0146] The high-resolution one-dimensional range image is input into the trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

[0147] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0148] 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.

[0149] 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 DA-ISTA network-based ISAR sparse imaging method, characterized in that, The method includes: Acquire ISAR sparse aperture echo data to be imaged; The ISAR sparse aperture echo data is processed to obtain the corresponding high-resolution one-dimensional range image; The high-resolution one-dimensional range image is input into a trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process for solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operations. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result. The data processing process of each layer is represented as follows: In the above formula, the superscript Indicates the first Layered network structure, Indicates the iteration step size parameter, Indicates initial input data, Indicates the first Output data of layered network structure Represents the soft thresholding function, where, Indicates threshold parameter, Indicates the first A layer is the output data of the current layer of the network structure. , , as well as These represent different convolutional units in the current layer network structure. as well as Each includes a linear convolution operator, and the convolution unit as well as Each includes two linear convolution operators and an activation layer embedded between the two linear convolution operators, and the convolution unit Convolutional unit The left inverse transform; After the high-resolution one-dimensional distance image is input into the trained DA-ISTA network, the low-squares estimation method is used to process the high-resolution one-dimensional distance image before it is input into the first layer of the neural network, and the processing result is input into the first layer of the neural network structure. The DA-ISTA network comprises a three-layer network structure.

2. The ISAR sparse imaging method according to claim 1, characterized in that, When training the DA-ISTA network, the loss function is expressed as: In the above formula, This represents the training imaging structure obtained after processing the training sample data via the DA-ISTA network. This represents the true label corresponding to the training sample data. This represents the regularization parameter.

3. The ISAR sparse imaging method according to claim 2, characterized in that, When training the DA-ISTA network, adaptive matrix estimation is used to optimize the loss function until the loss function converges, thus obtaining the trained DA-ISTA network.

4. The ISAR sparse imaging method according to claim 3, characterized in that, The trainable parameters in the DA-ISTA network include the parameters in each convolutional unit of each layer of the network structure, the iteration stride parameter, and the threshold parameter in the soft thresholding function.

5. A sparse imaging device for ISAR based on a DA-ISTA network, characterized in that, The device implements the ISAR sparse imaging method based on DA-ISTA network as described in any one of claims 1-4, including: The echo data acquisition module is used to acquire ISAR sparse aperture echo data to be imaged; A high-resolution one-dimensional range image acquisition module is used to process the ISAR sparse aperture echo data to obtain the corresponding high-resolution one-dimensional range image. The ISAR sparse imaging module is used to input the high-resolution one-dimensional range image into a trained DA-ISTA network to obtain the corresponding ISAR sparse imaging result. In the DA-ISTA network, the iterative process of solving the ISAR sparse imaging problem is spread across a multi-layer network structure based on the ISTA algorithm. In each layer of the network structure, the linear sparse transformation in the ISTA algorithm is replaced by nonlinear convolution operation. The high-resolution one-dimensional range image is processed through a sequentially connected multi-layer neural network structure to obtain the ISAR sparse imaging result.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

Images