Sparse aperture ISAR imaging method, device and equipment based on learning assistance

By constructing an ISAR imaging model in sparse scenarios and using neural networks for iterative solution, the problem of insufficient effectiveness and timeliness in sparse aperture ISAR imaging is solved, and high-precision imaging is achieved in the absence of target echoes.

CN118688803BActive Publication Date: 2024-11-22NAT UNIV OF DEFENSE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411171711.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-22
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

The existing sparse aperture ISAR self-focusing imaging method is not good in terms of effectiveness and timeliness, and the data-driven method is difficult to adapt to complex and changeable spatial target ISAR self-focusing imaging tasks, especially in the absence of target echoes. Achieve high-precision imaging.

Method used

Using a sparse aperture ISAR imaging method based on learning assistance, a sparse scene model including ISAR two-dimensional high-resolution image matrix and primary phase error matrix is ​​constructed, and a norm regular constraint and alternating iteration minimization algorithm are used to solve it, and combined with neural networks to iterate. Calculate to obtain high resolution ISAR images.

Benefits of technology

High-precision ISAR target imaging in the absence of target echoes is achieved, which improves the effectiveness and timeliness of imaging, avoids dependence on training data, and simplifies the hyperparameter adjustment process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118688803B_ABST
    Figure CN118688803B_ABST
Patent Text Reader

Abstract

The present application relates to a sparse aperture ISAR imaging method, device and equipment based on learning assistance, by constructing an imaging model in a sparse scene, the model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved by auxiliary solution, by using the imaging model to be regularized based on the #imgabs0# norm, an optimization objective function is obtained, and after substituting the acquired one-dimensional distance signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used for iterative solution to obtain a two-dimensional high-resolution image matrix to complete sparse aperture imaging. The present method can be used to effectively and quickly perform high-precision ISAR target imaging.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of ISAR imaging technology, and in particular to a learning-assisted sparse aperture ISAR imaging method, device and equipment. Background Art

[0002] Inverse Synthetic Aperture Radar (ISAR) imaging is one of the important means of space target monitoring and information perception. Limited by radar hardware conditions, ISAR often faces sparse observation scenes, resulting in problems such as missing scattering points and image defocusing in ISAR images. The goal of sparse aperture ISAR self-focusing imaging technology is to reconstruct high-resolution and focused ISAR images from incomplete echoes.

[0003] Existing sparse aperture ISAR self-focusing imaging methods can be mainly divided into model-driven methods and data-driven methods. Model-driven sparse aperture ISAR self-focusing imaging methods often have explicit mathematical models and good interpretability. This type of algorithm can be plug-and-play and instantly solve different ISAR imaging scenarios, such as different sparsity rates, different signal-to-noise ratios, and different echo dimensions. However, the algorithm is not as effective and timely as the learning-based method, and the hyperparameters in the algorithm need to be manually adjusted. The data-driven sparse aperture ISAR self-focusing imaging method uses an implicit black box neural network to solve the ISAR imaging problem end-to-end, and usually has an efficient solution speed and effective data-driven solution performance. However, this type of method is often highly dependent on training data and pre-training processes, and it is difficult to adapt to the actual needs of complex and changeable scenes and lack of training data in space target ISAR self-focusing imaging tasks. Summary of the invention

[0004] Based on this, it is necessary to provide a learning-assisted sparse aperture ISAR imaging method, device and equipment that can effectively perform high-precision target imaging in the absence of target echoes in order to address the above technical problems.

[0005] A learning-assisted sparse aperture ISAR imaging method, the method comprising:

[0006] Acquire ISAR target echo data under sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0007] Constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved in an auxiliary manner;

[0008] By adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0009] After substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, thereby obtaining the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0010] In one embodiment, the ISAR imaging model is constructed by an ISAR one-dimensional range signal matrix, a degradation matrix, a Gaussian white noise matrix and an ISAR two-dimensional high-resolution image matrix to be solved, wherein the degradation matrix is ​​represented by a downsampling matrix, a Fourier transform matrix and the initial phase error matrix.

[0011] In one embodiment, the optimization objective function includes three items, namely:

[0012] The first term is the observation signal fidelity term, which is Norm constraints;

[0013] The second and third items are respectively The norm is used to regularize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix to be solved, and the regularization parameter is used for control;

[0014] The optimization objective function aims to minimize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix.

[0015] In one embodiment, the optimization objective function is expressed as:

[0016]

[0017] In the above formula, represents the ISAR one-dimensional range signal matrix, represents the initial phase error matrix, represents the downsampling matrix, represents the Fourier transform matrix, represents the first item, represents the second term, where represents the regularization parameter, represents the third term, wherein, represents the regularization parameter.

[0018] In one embodiment, in each iterative calculation in the process of solving the optimization objective function by using an alternating iterative minimization algorithm, two neural networks are used to solve the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix respectively;

[0019] After the ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix are solved by using the neural network, the network parameters of the neural network are updated according to the solution results.

[0020] In one embodiment, in each iterative calculation in the process of solving the optimization objective function using the alternating iterative minimization algorithm:

[0021] The intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the previous iteration are respectively used as inputs of the corresponding neural network, and the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the current iteration are output through the neural network;

[0022] Substituting the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the current iteration into the optimization objective function for calculation, respectively obtaining the loss functions of the corresponding two neural networks;

[0023] Determine whether the loss function converges. If converged, the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the current iteration are the final ISAR two-dimensional high-resolution image matrix and the initial phase error matrix;

[0024] If it does not converge, the network parameters of the neural network are updated according to the loss function, and the neural network with updated network parameters is used for the next iterative calculation.

[0025] In one embodiment, the neural network for solving the initial phase error matrix is ​​a real neural network including two layers of fully connected networks;

[0026] The neural network for solving the ISAR two-dimensional high-resolution image matrix includes a complex neural network of an eight-layer fully convolutional neural network.

[0027] The present application also provides a learning-assisted sparse aperture ISAR imaging device, the device comprising:

[0028] The ISAR echo signal acquisition module to be imaged is used to obtain the ISAR target echo data under the sparse aperture, and generate the corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0029] An ISAR imaging model building module is used to build an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix for auxiliary solution;

[0030] The module is used to optimize the objective function by adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0031] The sparse aperture ISAR imaging module is used to substitute the ISAR one-dimensional range signal matrix into the optimization objective function, and then adopt an alternating iterative minimization algorithm to perform iterative solution to obtain the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0032] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0033] Acquire ISAR target echo data under sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0034] Constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved in an auxiliary manner;

[0035] By adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0036] After substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, thereby obtaining the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0037] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0038] Acquire ISAR target echo data under sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0039] Constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved in an auxiliary manner;

[0040] By adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0041] After substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, thereby obtaining the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0042] The above-mentioned sparse aperture ISAR imaging method, device and equipment based on learning assistance construct an ISAR imaging model in a sparse scene, wherein the model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved by auxiliary means. The norm is used to regularize the ISAR imaging model to obtain the optimization objective function. After substituting the obtained ISAR one-dimensional range signal matrix into the optimization objective function, the alternating iterative minimization algorithm is used for iterative solution to obtain the ISAR two-dimensional high-resolution image matrix to complete the sparse aperture ISAR imaging. In each iterative calculation in the process of solving the optimization objective function using the alternating iterative minimization algorithm, two neural networks are used to solve the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix respectively. This method can be used to effectively and quickly perform high-precision ISAR target imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of a learning-assisted sparse aperture ISAR imaging method in one embodiment;

[0044] Figure 2 is a schematic diagram of a typical ISAR self-focusing imaging scene in an embodiment;

[0045] Figure 3 Schematic diagram of sparse aperture ISAR imaging degradation in one embodiment, wherein (a) is a full aperture one-dimensional range image, (b) is a sparse aperture one-dimensional range image, and (c) is a two-dimensional high-resolution image;

[0046] Figure 4 It is a schematic diagram of forming sparse aperture echoes by randomly sampling ISAR original echoes in the azimuth dimension in one embodiment;

[0047] Figure 5 A schematic diagram of the overall implementation flow of the algorithm of the method in one embodiment;

[0048] Figure 6 A schematic diagram of the structure of an initial phase error matrix estimation neural network in one embodiment;

[0049] Figure 7 A schematic diagram of the structure of an ISAR two-dimensional high-resolution image matrix estimation neural network in one embodiment;

[0050] Figure 8 It is a schematic diagram of the experimental comparison results of simulation data of the present method and multiple traditional methods under different sparse rate conditions in an experiment with a signal-to-noise ratio of 40dB;

[0051] Fig. 9It is a schematic diagram of the experimental comparison results of the measured data of the present method and multiple traditional methods under different sparse rate conditions in an experiment with a signal-to-noise ratio of 40dB;

[0052] Fig.10 It is a schematic diagram of the experimental comparison results of simulation data of the present method and multiple traditional methods under different signal-to-noise ratio conditions with a sparsity rate of 0.25 in an experiment;

[0053] Fig.11 It is a schematic diagram of the experimental comparison results of the measured data of the present method and multiple traditional methods under different signal-to-noise ratio conditions with a sparsity rate of 0.25 in an experiment;

[0054] Fig.12 is a structural block diagram of a sparse aperture ISAR imaging device based on learning assistance in one embodiment;

[0055] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] In view of the problems in the prior art of using algorithms or learning methods for sparse aperture ISAR self-focusing imaging, such as effectiveness, timeliness, need for hyperparameter adjustment, difficulty in obtaining training data for neural networks or small amount of training data, which leads to difficulty in convergence of neural networks, in this application, Figure 1 As shown, a learning-assisted sparse aperture ISAR imaging method is provided, which specifically includes the following steps:

[0058] Step S100, acquiring ISAR target echo data under sparse aperture, and generating a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data.

[0059] Step S110, constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix for auxiliary solution.

[0060] Step S120, by adopting The norm is used to regularize the ISAR imaging model and obtain the optimized objective function.

[0061] Step S130, after substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, and an ISAR two-dimensional high-resolution image matrix is ​​obtained to complete sparse aperture ISAR imaging.

[0062] In step S100, ISAR target echo data under sparse aperture is firstly acquired.

[0063] like Figure 2 As shown, it is a typical ISAR self-focusing imaging scene. In order to clearly analyze the movement of the monitored target, a coordinate system is first established for the target. In this coordinate system, the origin O is located at the center of gravity of the target, the y-axis is selected as the radar line of sight (LOS), and the x-axis is located on the imaging plane to form a right-handed coordinate system. In practical applications, the target performs complex non-cooperative motion, and the form of the echo is usually more complex. Since the radar coherent processing interval (CPI) is short, the motion of the target can be simplified. In this embodiment, it is assumed that the motion of the target is in the xOy plane, and the motion of the target consists of translation and rotation. Assuming that the translational motion of the target has been compensated by distance alignment and phase adjustment, the target motion can be simplified to a turntable model of uniform motion, and any point q on the target moves at an angular velocity. Rotate.

[0064] Furthermore, assuming that the echo signal has completed envelope alignment and self-focusing operations, where B represents the signal bandwidth, represents the reflection coefficient of scattering point p, and P represents the total number of scattering points. Considering the additive echo noise in the signal, the ISAR target echo data at this time It can be written as:

[0065] (1)

[0066] In formula (1), and Denote the fast time and the frequency modulation slope respectively. Any point q on the target moves at an angular velocity Rotation, where , , They represent the speed of light, slow time and echo signal amplitude respectively. represents random noise, and express Figure 2 The coordinates of the Pth scattering point on the target.

[0067] Furthermore, by performing Fourier transform on the distance direction of formula (1), the echo signal The ISAR one-dimensional range profile (High Resolution Range Profile, HRRP) signal matrix can be obtained.

[0068] In step S110, an ISAR imaging model under a sparse background is constructed. First, a general ISAR imaging model is constructed. The ISAR two-dimensional high-resolution image can be obtained by performing Fourier transform on the azimuth direction of the ISAR one-dimensional range image signal matrix. By writing this process in the form of a calculation matrix, a general observation signal model can be obtained, which is expressed as:

[0069] (2)

[0070] In formula (2), represents the ISAR one-dimensional range image signal matrix, represents the two-dimensional ISAR high-resolution image matrix, represents the degradation matrix of the ISAR image, represents the Gaussian white noise matrix of the two-dimensional ISAR image. Indicates the number of azimuth pulses under sparse echoes, Indicates the number of azimuth pulses under the original echo, Indicates the number of range image units.

[0071] Furthermore, considering ISAR imaging in sparse scenarios, some pulses in the azimuth direction cannot be used or are actively abandoned, resulting in missing echo data in the slow time domain, while the fast time domain data is not affected, such as Figure 3 As shown. Figure 3 As shown in the figure, the full aperture one-dimensional range image, sparse aperture one-dimensional range image, and two-dimensional high-resolution image of the same target are given to demonstrate the sparse aperture ISAR imaging degradation. The degradation matrix in formula (2) can be further modeled as a downsampling matrix With the Fourier transform matrix The form of multiplication , then formula (2) can be further expressed as:

[0072] (3)

[0073] In formula (3), the Fourier transform matrix With the downsampling matrix is known and determined. Among them, the downsampling matrix The form of is related to the sparse pattern and the sparse rate. The random sparse pattern is a common sparse sampling pattern, such as Figure 4 As shown in the figure, the ISAR original echo is randomly sampled in the azimuth dimension to form a sparse aperture echo. It is constructed by randomly deleting rows of the identity matrix of size M×M. The sparsity ratio represents the downsampling matrix The ratio of the number of rows with a median value of 1 to the total number of rows.

[0074] Furthermore, the above is the case when there is no residual initial phase error in the azimuth direction under ideal conditions. When there is an initial phase error in the azimuth direction, formula (1) becomes:

[0075] (4)

[0076] In formula (4), Indicates the carrier frequency, represents the phase error in azimuth and is assumed to be related to slow times only. The initial phase error matrix in the azimuth is a diagonal matrix and can be expressed as:

[0077] (5)

[0078] When in a sparse aperture scene, Represents the initial phase error matrix after downsampling.

[0079] Furthermore, the ISAR imaging model in sparse scenes can be obtained by the ISAR one-dimensional range signal matrix , degradation matrix, Gaussian white noise matrix And the ISAR two-dimensional high-resolution image matrix to be solved Constructed, where the degradation matrix is ​​composed of the downsampling matrix , Fourier transform matrix And the initial phase error matrix The specific expression is:

[0080] (6)

[0081] In formula (6), the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix All are unknowns, so the next step is to solve the optimization objective function of the corresponding degradation problem by using formula (6). Then in step S110, the constructed ISAR imaging model is shown in formula (6).

[0082] In step S120, first, the objective function of the degradation problem in this application can be expressed as:

[0083] (7)

[0084] In formula (7), and represent the regularization parameters of ISAR two-dimensional high-resolution image matrix and initial phase error matrix respectively.

[0085] However, the objective function shown in formula (7) is a non-convex and highly ill-conditioned problem. The regularization constraint of the norm relaxes the problem and obtains the final optimization objective function, which includes three items: the first item is the observation signal fidelity item, using norm constraint, the second and third terms are respectively The norm imposes regular constraints on the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix to be solved, and the regularization parameter is used for control. At the same time, the objective function is optimized to minimize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix.

[0086] Specifically, the optimization objective function is expressed as:

[0087] (8)

[0088] In formula (8), represents the ISAR one-dimensional range signal matrix, represents the initial phase error matrix, represents the downsampling matrix, represents the Fourier transform matrix, represents the first item, represents the second term, where represents the regularization parameter, represents the third term, where represents the regularization parameter.

[0089] Furthermore, when solving formula (8), the non-convex and highly ill-conditioned problems exist, and the gradient-based AM algorithm is no longer applicable. The "gray box" local neural network modeling method based on problem modeling is introduced into the imaging and self-focusing process.

[0090] In step S130, in each iterative calculation in the process of solving the optimization objective function by using the alternating iterative minimization algorithm, after solving the ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix by using the neural network, the network parameters of the neural network are also updated according to the solution results. This means that in this method, it is not necessary to train the neural network in advance. The process of iteratively solving formula (8) is the process of neural network training. When the neural network converges, the current output result is the result of the final iterative calculation, which saves a lot of work and improves efficiency. At the same time, it is not necessary to consider the problem of training data.

[0091] In this embodiment, in each iterative calculation in the process of solving the optimization objective function by using the alternating iterative minimization algorithm: the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix obtained by the previous iterative calculation are respectively used as the input of the corresponding neural network, and the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated by the current iterative calculation are output through the neural network. The intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated by the current iterative calculation are substituted into the optimization objective function for calculation, and the loss functions of the corresponding two neural networks are respectively obtained. Then, it is determined whether the loss function converges. If converged, the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated by the current iterative calculation are the final ISAR two-dimensional high-resolution image matrix and the initial phase error matrix. If not converged, the network parameters of the neural network are updated according to the loss function, and the neural network after the network parameters are updated is used for the next iterative calculation.

[0092] Specifically, the estimation neural network used to solve the initial phase error matrix is ​​expressed as:

[0093] (9)

[0094] In formula (9), the superscript is the number of iterative calculations, is the initial phase error matrix estimation neural network, which is a real neural network. When solving the problem for the first time, the input is a Gaussian random noise. , ,noise Updated with the algorithm’s iterative process, is the network parameter. At the same time, since the initial phase error matrix is ​​a diagonal matrix and the eigenvector is the initial phase error, formula (9) is to solve the phase value on the diagonal line of the initial phase error matrix, and then, the initial phase error matrix is ​​calculated based on the current iteration obtained by the solution. , which can be expressed as:

[0095] (10)

[0096] In formula (10), For the The initial phase error of a pulse.

[0097] Specifically, the neural network used to solve the ISAR two-dimensional high-resolution image matrix is ​​expressed as:

[0098] (11)

[0099] In formula (11), is the ISAR two-dimensional high-resolution image matrix estimation neural network, which is a complex neural network. Similarly, when performing the first solution, the network The input is a Gaussian random noise , the noise Updated with the algorithm’s iterative process, is the network parameter.

[0100] So far, the initial phase error matrix in sparse aperture ISAR imaging and self-focusing problems has been completed. and 2D high resolution image The local neural network modeling is based on the two neural networks that replace the solution variables in the original problem modeling. The optimization objective function can be further converted to:

[0101] (12)

[0102] Neural Network for Matrix Estimation of ISAR 2D High Resolution Image Estimation of Neural Networks Using Initial Phase Error Matrix Generate results and By network parameters and The introduction of local neural networks in formulas (9) and (11) transforms the optimization problem of multiple variables in formula (8) into the optimization problem of network parameters. , Optimization problem.

[0103] For multivariable optimization problems such as formula (12), an alternating iterative minimization algorithm can usually be used to solve it, which can be expressed as:

[0104] (13)

[0105] When using formula (13) In the first iteration, the goal of the algorithm is still to minimize the sum of the data consistency term and the regularization term. The network parameters of the two neural networks in the iteration process and , we can get The loss function of the iterative process is:

[0106] (14)

[0107] In formula (14), and The initial phase error matrix is ​​used to estimate the neural network. and ISAR 2D high resolution image matrix estimation neural network The loss function of .

[0108] Next, according to formula (14), the loss function is obtained and , can be optimized and updated by the Adaptivemoment (Adam) optimizer, as follows:

[0109] (15)

[0110] Repeat the update operation in formula (15) and iterate the ISAR two-dimensional high-resolution image matrix estimation neural network repeatedly. Estimation of Neural Networks Using Initial Phase Error Matrix Until convergence, the final predicted ISAR two-dimensional high-resolution image matrix and the predicted initial phase error matrix are obtained.

[0111] In this embodiment, an algorithm is also provided to implement this method. However, it should be noted that the algorithm for implementing this method can be expressed in multiple ways and is not limited to the following algorithm. The algorithm is as follows:

[0112]

[0113] like Figure 5 As shown in FIG. 1 , a schematic diagram of the overall implementation flow of the above algorithm is given.

[0114] In this embodiment, the initial phase error matrix estimates the neural network , set as a two-layer fully connected neural network (FCN), such as Figure 6 As shown in Figure 2, FCN is an artificial neural network structure with a relatively simple connection method and is a type of feedforward neural network.

[0115] Preferably, the initial phase error matrix estimates the neural network It consists of an input layer, a hidden layer, and an output layer. The dimensions of the input layer and the output layer are 256, the dimension of the middle hidden layer is 420, and there are 420 neurons in the hidden layer. There is a ReLU activation function between each layer to increase the expressive power of the network. Finally, the output is processed by the Tanh activation function, which limits the output range to between -1 and 1. This structure can perform complex nonlinear mapping on the input data through multiple layers of linear transformation and activation functions, thereby capturing richer feature representations.

[0116] In this embodiment, the ISAR two-dimensional high-resolution image matrix estimation neural network , set to an eight-layer fully convolutional neural network (CNN), such as Figure 7 shown.

[0117] Optimized ISAR 2D high resolution image matrix estimation neural network The number of convolution channels from input to output are 1, 16, 32, 64, 128, 64, 32, 16, and 1 respectively, and the convolution kernel size is set to , Figure 7 In the figure, arrow 1, arrow 2, arrow 3, and arrow 4 represent convolutional layers, pooling layers, transposed convolutional layers, and skip connections, respectively.

[0118] In this paper, the effect of the learning-assisted plug-and-play sparse aperture ISAR imaging and self-focusing method (hereinafter referred to as this method) proposed in this paper is also verified through experiments, which include experiments under different imaging conditions, such as different sparsity rates, different signal-to-noise ratios, etc.

[0119] 1. Experimental conditions:

[0120] 1.1) This method builds a deep learning model based on the pytorch framework.

[0121] 1.2) The dataset uses a space target simulation dataset based on physical optics (PO) and a real scene measurement dataset of three aircraft. The size of each space target ISAR image is The PO-based space target simulation dataset consists of 3,000 data pairs of 6 targets; the real-scene measurement dataset contains echo data of three aircraft (Yak-42, An-26, and Citation). The carrier frequency, bandwidth, pulse width, and PRF of the radar system in this dataset are 5520 MHz, 400 MHz, 25.6 µs, and 100 Hz, respectively.

[0122] 1.3) In the testing phase, the proposed method does not use any data for training, nor does it have any pre-training process. In the solution process, only the ISAR sparse aperture echo one-dimensional range image S is input. The comparative method CV-ADMMN method uses the training set in the simulation data set for training.

[0123] 2. Experimental results:

[0124] The experimental comparison results of simulation data of this method and several traditional methods under different sparse rate conditions with a signal-to-noise ratio of 40dB are as follows: Figure 8 shown.

[0125] The experimental comparison results of the measured data of this method and several traditional methods under different sparse rate conditions with a signal-to-noise ratio of 40dB are as follows: Fig. 9 shown.

[0126] The experimental comparison results of simulation data of this method with multiple traditional methods under different signal-to-noise ratio conditions with a sparse rate of 0.25 are as follows: Fig.10 shown.

[0127] The experimental comparison results of the measured data of this method and several traditional methods under different signal-to-noise ratio conditions with a sparse rate of 0.25 are as follows: Fig.11 shown.

[0128] 3. Experimental analysis:

[0129] Figure 8 , Fig. 9 , Fig.10 and Fig.11 The results of the self-focusing imaging methods under different sparsity rates and different signal-to-noise ratios in the simulation data set and the measured data set are compared. It can be seen from the figure that the images reconstructed by the model-driven RD method and the MEA-ADMM method are seriously defocused; the data-driven CV-ADMMN method can better restore high-quality ISAR focused images and has a good visualization effect. The method in this paper has the best visualization effect, and can still have a high resolution under low sparsity rate and low signal-to-noise ratio conditions, and the image focusing effect is the best.

[0130] In the above-mentioned learning-assisted sparse aperture ISAR imaging method, the ISAR image prediction network and the initial phase error prediction network are initialized according to the given incomplete ISAR radar echo and echo dimension, and two random noises are generated as the initial inputs of the two networks respectively. The network parameters of the ISAR image prediction network and the initial phase error prediction network are iteratively calculated based on the objective function of ISAR imaging self-focusing. After each iteration, the ISAR image and the initial phase error matrix are generated respectively until convergence, and the final predicted ISAR image and initial phase error matrix are obtained. The use of this method can ensure good effectiveness and practicability, and has the advantages of strong interpretability, good generalization ability, no data dependence, and no pre-training requirements.

[0131] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed 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 part of the sub-steps or stages of other steps.

[0132] In one embodiment, Fig.12 As shown, a sparse aperture ISAR imaging device based on learning assistance is provided, comprising: an ISAR echo signal obtaining module 200 to be imaged, an ISAR imaging model building module 210, an optimization objective function obtaining module 220 and a sparse aperture ISAR imaging module 230, wherein:

[0133] The ISAR echo signal acquisition module 200 is used to obtain ISAR target echo data under a sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0134] An ISAR imaging model building module 210 is used to build an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix for auxiliary solution;

[0135] Optimizing the objective function to obtain module 220 is used to optimize the objective function by using a The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0136] The sparse aperture ISAR imaging module 230 is used to substitute the ISAR one-dimensional range signal matrix into the optimization objective function, and then adopt an alternating iterative minimization algorithm to perform iterative solution to obtain the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0137] For the specific definition of the sparse aperture ISAR imaging device based on learning assistance, please refer to the definition of the sparse aperture ISAR imaging method based on learning assistance in the above text, which will not be repeated here. Each module in the above-mentioned sparse aperture ISAR imaging device based on learning assistance can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0138] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.13As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a sparse aperture ISAR imaging method based on learning assistance is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0139] Those skilled in the art will understand that Fig.13 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0140] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0141] Acquire ISAR target echo data under sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0142] Constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved in an auxiliary manner;

[0143] By adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0144] After substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, thereby obtaining the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0145] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0146] Acquire ISAR target echo data under sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data;

[0147] Constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved in an auxiliary manner;

[0148] By adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function;

[0149] After substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, thereby obtaining the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

[0150] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0151] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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.

[0152] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A learning-assisted sparse aperture ISAR imaging method, characterized in that: The method comprises: Acquire ISAR target echo data under sparse aperture, and generate a corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data; Constructing an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix to be solved in an auxiliary manner; By adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function, wherein the optimization objective function includes three items, namely: the first item is the observation signal fidelity item, using norm constraint, the second and third terms are respectively The norm is used to regularize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix to be solved, and the regularization parameter is used for control. The optimization objective function aims to minimize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix. The optimization objective function is expressed as: ; In the above formula, represents the ISAR one-dimensional range signal matrix, represents the initial phase error matrix, represents the downsampling matrix, represents the Fourier transform matrix, represents the ISAR two-dimensional high-resolution image matrix to be solved, represents the first item, represents the second term, where represents the regularization parameter, represents the third term, wherein, represents the regularization parameter; After substituting the ISAR one-dimensional range signal matrix into the optimization objective function, an alternating iterative minimization algorithm is used to iteratively solve the problem, thereby obtaining the ISAR two-dimensional high-resolution image matrix to complete sparse aperture ISAR imaging.

2. The sparse aperture ISAR imaging method according to claim 1, characterized in that: The ISAR imaging model is constructed by an ISAR one-dimensional range signal matrix, a degradation matrix, a Gaussian white noise matrix and an ISAR two-dimensional high-resolution image matrix to be solved, wherein the degradation matrix is ​​represented by a downsampling matrix, a Fourier transform matrix and the initial phase error matrix.

3. The sparse aperture ISAR imaging method according to claim 2, characterized in that: In each iterative calculation in the process of solving the optimization objective function by using the alternating iterative minimization algorithm, two neural networks are used to solve the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix respectively; After the ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix are solved by using the neural network, the network parameters of the neural network are updated according to the solution results.

4. The sparse aperture ISAR imaging method according to claim 3, characterized in that: In each iterative calculation in the process of solving the optimization objective function using the alternating iterative minimization algorithm: The intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the previous iteration are respectively used as inputs of the corresponding neural network, and the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the current iteration are output through the neural network; Substituting the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the current iteration into the optimization objective function for calculation, respectively obtaining the loss functions of the corresponding two neural networks; Determine whether the loss function converges. If converged, the intermediate ISAR two-dimensional high-resolution image matrix and the intermediate initial phase error matrix calculated in the current iteration are the final ISAR two-dimensional high-resolution image matrix and the initial phase error matrix; If it does not converge, the network parameters of the neural network are updated according to the loss function, and the neural network with updated network parameters is used for the next iterative calculation.

5. The sparse aperture ISAR imaging method according to claim 4, characterized in that: The neural network for solving the initial phase error matrix is ​​a real neural network including two layers of fully connected networks; The neural network for solving the ISAR two-dimensional high-resolution image matrix includes a complex neural network of an eight-layer fully convolutional neural network.

6. A learning-assisted sparse aperture ISAR imaging device, characterized in that: The device comprises: The ISAR echo signal acquisition module to be imaged is used to obtain the ISAR target echo data under the sparse aperture, and generate the corresponding ISAR one-dimensional range signal matrix according to the ISAR target echo data; An ISAR imaging model building module is used to build an ISAR imaging model in a sparse scene, wherein the ISAR imaging model includes an ISAR two-dimensional high-resolution image matrix to be solved and an initial phase error matrix for auxiliary solution; The module is used to optimize the objective function by adopting The norm is used to regularize the ISAR imaging model to obtain an optimization objective function, wherein the optimization objective function includes three items, namely: the first item is the observation signal fidelity item, using norm constraint, the second and third terms are respectively The norm is used to regularize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix to be solved, and the regularization parameter is used for control. The optimization objective function aims to minimize the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix. The optimization objective function is expressed as: ; In the above formula, represents the ISAR one-dimensional range signal matrix, represents the initial phase error matrix, represents the downsampling matrix, represents the Fourier transform matrix, represents the ISAR two-dimensional high-resolution image matrix to be solved, represents the first item, represents the second term, where represents the regularization parameter, represents the third term, wherein, represents the regularization parameter; The sparse aperture ISAR imaging module is used to substitute the ISAR one-dimensional range signal matrix into the optimization objective function, and then adopt an alternating iterative minimization algorithm to iteratively solve the problem, so as to obtain the ISAR two-dimensional high-resolution image matrix, so as to complete sparse aperture ISAR imaging. In each iterative calculation in the process of solving the optimization objective function by adopting the alternating iterative minimization algorithm, two neural networks are used to solve the ISAR two-dimensional high-resolution image matrix and the initial phase error matrix respectively.

7. 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, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.