THz-SAR moving target high-resolution imaging method, device and equipment

Abstract

The invention relates to a THz-SAR moving target high-resolution imaging method, device and equipment, and the method comprises the steps: constructing a target echo model, representing the model as a distance image sequence in a matrix form based on a compressed sensing theory, constructing a distance image sequence model, constructing an azimuth phase error matrix through employing a maximized image contrast criterion, and carrying out the high-resolution imaging of a THz-SAR moving target. Constructing a target function according to the target range profile matrix, the background range profile matrix and a to-be-solved moving target image, introducing an imaging model and an auxiliary variable to restrain the target function, and converting the to-be-optimized target function into a plurality of sub-problems for iterative solution by adopting an alternating direction multiplier method; and expanding the moving target into a moving target self-focusing imaging network based on robust principal component analysis, preprocessing a target echo signal obtained by actual measurement, and inputting the preprocessed target echo signal into the network to obtain an SAR high-resolution imaging result of the moving target. By adopting the method, high-precision SAR imaging can be carried out on the moving target under the condition of high error sensitivity.

Description

Technical Field

[0001] The present application relates to the field of radar target recognition technology, and in particular to a THz-SAR moving target high-resolution imaging method, device and equipment. Background Art

[0002] Compared with traditional microwave SAR, terahertz synthetic aperture radar (THz-SAR) has significant advantages of high frame rate, low latency and high resolution, which is very suitable for dynamic monitoring of maneuvering targets and can meet the application requirements of high-resolution imaging, attitude estimation and intention inversion of maneuvering targets at the component level. However, due to the non-ideal motion of the target and the airborne platform in the scene, THz-SAR imaging has problems such as blur, defocusing and low resolution.

[0003] With the rapid development of deep learning technology, learning-based SAR imaging methods have attracted much attention due to their advantages such as strong adaptability, high resolution and high efficiency, especially their ability to meet the needs of high-performance imaging under non-ideal conditions such as unknown motion errors in echoes and sparse sampling of echoes.

[0004] However, since the THz-SAR carrier aircraft's flight attitude and speed often have certain instabilities, and the target also has non-cooperative translation and micro-motion, it brings challenges to the non-parametric imaging method. Summary of the invention

[0005] Based on this, it is necessary to provide a THz-SAR moving target high-resolution imaging method, device and equipment that can solve the problem of high error sensitivity of terahertz synthetic aperture radar in response to the above technical problems.

[0006] A THz-SAR moving target high-resolution imaging method, the method comprising:

[0007] Constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp;

[0008] Based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion;

[0009] Splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved;

[0010] An imaging model and auxiliary variables are introduced to constrain the objective function, and an alternating direction multiplier method is used to transform the objective function to be optimized into a plurality of sub-problems to be solved iteratively;

[0011] Expanding the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and training the network to obtain a trained moving target self-focusing imaging network;

[0012] A target echo signal is obtained, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, and the target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0013] In one embodiment, the azimuth phase error matrix constructed using the maximization image contrast criterion is expressed as:

[0014]

[0015] Where diag(·) represents a diagonal matrix consisting of column vectors, and the phase error The optimal estimate of is expressed as:

[0016]

[0017] In the above formula, X represents the moving target image to be solved, the size of the moving target image to be solved is M×N, the value of each pixel in the image is expressed as X(m,n), * represents taking conjugate, S i ' represents the i-th column of the distance image matrix S, F H is the Fourier transform matrix.

[0018] In one embodiment, in the objective function, the target range image matrix, the background range image matrix and the moving target image to be solved are constrained using the nuclear norm, the F norm and the l1 norm in sequence.

[0019] In one embodiment, the objective function to be optimized is expressed as:

[0020] min||T|| * +α1||N|| F +α2||Z||1

[0021] S=T+N

[0022] stT=EFX

[0023] X=Z

[0024] In the above formula, ||T||* Indicates the use of nuclear norm to constrain the target distance image matrix, ||N|| F Indicates that the background range image matrix is ​​constrained by the F norm, ||Z||1 indicates that the auxiliary variables are constrained by the l1 norm, and α1 and α2 represent regularization parameters.

[0025] In one embodiment, when the objective function to be optimized is converted into a plurality of sub-problems to be solved iteratively using the alternating direction multiplier method:

[0026] Introducing a Lagrangian multiplier matrix and a penalty term coefficient, and constructing an augmented Lagrangian function according to the objective function to be optimized;

[0027] Using the alternating direction multiplier method, the augmented Lagrangian function is transformed into a plurality of sub-problems to be solved iteratively;

[0028] In one iterative calculation, multiple sub-problems are included, including updating the target range image matrix, the background range image matrix, the moving target image to be solved, the auxiliary variables, the azimuth phase error matrix, the Lagrange multiplier matrix and the penalty term coefficients in turn.

[0029] In one embodiment, the moving target self-focusing imaging network includes a multi-layer neural network corresponding to the number of iterative calculations performed on the multiple sub-problems;

[0030] In each layer of the neural network, the target range image matrix, the background range image matrix, the moving target image to be solved, the auxiliary variables, the azimuth phase error matrix, the Lagrange multiplier matrix and the penalty term coefficient are updated in turn.

[0031] In one embodiment, when training the moving target self-focusing imaging network:

[0032] Simulating multiple moving point targets according to preset parameters to obtain simulation data;

[0033] Dechirp-demodulated range image sequences of the simulation data are used as training data;

[0034] The range Doppler algorithm imaging results of the stationary point targets used by each moving point target are used as the true value label;

[0035] The moving target self-focusing imaging network is trained using the training data and the true value labels.

[0036] In one embodiment, a root mean square error loss function is used to train the moving target self-focusing imaging network.

[0037] The present application also provides a THz-SAR moving target high-resolution imaging device, the device comprising:

[0038] A target echo model building module is used to build a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp;

[0039] A range image sequence model building module is used to represent the target echo model as a range image sequence in matrix form based on the compressed sensing theory, and to build a range image sequence model according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion;

[0040] An objective function construction module is used to split the range image sequence into a target range image matrix and a background range image matrix, and to construct an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved;

[0041] An alternating direction multiplier method solution module, which constrains the objective function by introducing an imaging model and auxiliary variables, and uses an alternating direction multiplier method to convert the objective function to be optimized into a plurality of sub-problems to be solved iteratively;

[0042] A moving target self-focusing imaging network obtaining module is used to expand the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and train the network to obtain a trained moving target self-focusing imaging network;

[0043] The SAR high-resolution imaging result obtaining module is used to obtain a target echo signal, wherein the target echo signal is obtained by detecting a moving target using a terahertz synthetic aperture radar, and the target echo signal is pre-processed and then input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0044] 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:

[0045] Constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp;

[0046] Based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion;

[0047] Splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved;

[0048] An imaging model and auxiliary variables are introduced to constrain the objective function, and an alternating direction multiplier method is used to transform the objective function to be optimized into a plurality of sub-problems to be solved iteratively;

[0049] Expanding the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and training the network to obtain a trained moving target self-focusing imaging network;

[0050] A target echo signal is obtained, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, and the target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0051] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0052] Constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp;

[0053] Based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion;

[0054] Splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved;

[0055] An imaging model and auxiliary variables are introduced to constrain the objective function, and an alternating direction multiplier method is used to transform the objective function to be optimized into a plurality of sub-problems to be solved iteratively;

[0056] Expanding the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and training the network to obtain a trained moving target self-focusing imaging network;

[0057] A target echo signal is obtained, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, and the target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0058] The above-mentioned THz-SAR moving target high-resolution imaging method, device and equipment construct a target echo model obtained by detecting a moving target with a terahertz synthetic aperture radar, and then express the target echo model as a range image sequence in matrix form based on the compressed sensing theory, and construct a range image sequence model according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by maximizing the image contrast criterion, the range image sequence is split into a target range image matrix and a background range image matrix, the objective function is constructed according to the target range image matrix, the background range image matrix and the moving target image to be solved, the imaging model and auxiliary variables are introduced to constrain the objective function, and the alternating direction multiplier method is used to convert the objective function to be optimized into multiple sub-problems to be solved iteratively, the process of iteratively solving each sub-problem is expanded into a moving target self-focusing imaging network based on robust principal component analysis, and it is trained to obtain the trained moving target self-focusing imaging network, the measured target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain the SAR high-resolution imaging result of the moving target. This method can be used to perform high-precision SAR imaging of moving targets to address the problem of high error sensitivity of terahertz synthetic aperture radar. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic diagram of a process of a THz-SAR moving target high-resolution imaging method in one embodiment;

[0060] Figure 2 A schematic diagram of a SAR ground target motion model in one embodiment;

[0061] Figure 3 A schematic diagram of a line frequency modulation pulse pressure in an embodiment;

[0062] Figure 4 A schematic diagram of the structure of a moving target self-focusing imaging network in one embodiment;

[0063] Figure 5 A schematic diagram of input data and corresponding label images during network simulation training in one embodiment;

[0064] Figure 6 A schematic diagram of input data and corresponding label images during network actual test training in one embodiment;

[0065] Figure 7A schematic diagram of a gradient propagation process of a moving target self-focusing imaging network in one embodiment;

[0066] Figure 8 A schematic diagram of a simulated motion point target position and corresponding network input and label in one embodiment;

[0067] Fig. 9 A schematic diagram of imaging results of different algorithm simulation data comparison in one embodiment;

[0068] Fig.10 A schematic diagram of a measured moving target ROI and corresponding network input and label in one embodiment;

[0069] Fig.11 A schematic diagram of imaging results of different algorithms in one embodiment comparing measured data;

[0070] Fig.12 A structural block diagram of a THz-SAR moving target high-resolution imaging device in one embodiment;

[0071] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0072] 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.

[0073] In view of the high sensitivity of THz-SAR to errors, the existing non-parametric imaging methods cannot solve this problem well. In this application, a THz-SAR moving target high-resolution imaging method is provided, which specifically includes the following steps:

[0074] Step S100, constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compressed echo signal obtained after demodulation processing using Dechirp.

[0075] Step S110, based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion.

[0076] Step S120 , splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved.

[0077] Step S130, introducing an imaging model and auxiliary variables to constrain the objective function, and using an alternating direction multiplier method to transform the objective function to be optimized into a plurality of sub-problems to be solved iteratively.

[0078] Step S140, expanding the process of iteratively solving each sub-problem into a moving target self-focusing imaging network based on robust principal component analysis, and training it to obtain a trained moving target self-focusing imaging network.

[0079] Step S150, obtaining a target echo signal, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, preprocessing the target echo signal and inputting it into a trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0080] In this embodiment, in order to solve the problem of high sensitivity of THz-SAR to errors, a moving target learning imaging method based on sparse recovery is proposed. The method first derives the self-focusing module based on the maximum contrast criterion, and embeds it into the iterative solution process of the alternating direction multiplier method (ADMM) based on sparse recovery, instead of directly using the traditional phase gradient autofocus (PGA) algorithm to compensate for the azimuth motion error. Furthermore, considering that the THz-SAR moving target range image matrix has low-rank characteristics, the background clutter range image and the moving target image matrix are sparse, the idea of ​​robust principal component analysis (RPCA) is introduced, and an iterative solution process based on ADMM is proposed. Subsequently, a deep unfolding imaging network is established to realize the moving target background separation and imaging driven by the joint model data.

[0081] First, in step S100, a model is built for the echo pattern obtained by detecting a moving target using a terahertz synthetic aperture radar.

[0082] like Figure 2 As shown in the figure, the ground target motion model based on THz-SAR is built, where the X-axis is the coordinate axis of the airborne platform's motion direction, the Z-axis is the coordinate axis perpendicular to the ground target's motion plane, and the Y-axis is perpendicular to the X-axis and Z-axis respectively. Assuming that the airborne platform works in the positive side view mode, in order to better explore the target imaging characteristics, the single-point imaging process is analyzed. The initial position of point P is (x0, y0, 0), the aircraft's motion direction is the azimuth direction (slow time direction), the radar's electromagnetic wave emission direction is the distance direction (fast time direction), the aircraft's altitude is H, and the motion speed is v a, the shortest slant distance from the aircraft platform to the target point P is

[0083] If the target moves in a straight line with a speed of v and an acceleration of a, it can be decomposed into v along the X-axis and the Y-axis respectively. x 、v y 、a x 、a y According to the distance model, the instantaneous slant range R(η) from the aircraft to the target is expanded by Taylor series as follows:

[0084]

[0085] Due to the influence of the characteristics of the carrier aircraft and external airflow, the radar will inevitably produce vibration errors during movement. Assuming that the vibration of the carrier aircraft can be regarded as the superposition of periodic simple harmonic components of different frequencies, this vibration can be modeled along the radar line of sight as:

[0086]

[0087] In formula (2), η represents the slow time, I represents the number of vibration components, and A i 、f i , represent the amplitude, frequency and initial phase of the i-th vibration component respectively. When the absolute value of the product of the vibration frequency and the synthetic aperture time is greater than or equal to 1, that is, |f v ·T s |≥1, the vibration is high-frequency vibration.

[0088] Furthermore, the vibration frequency range of the helicopter platform is about 10 to 30 Hz, while the synthetic aperture time of THz-SAR is about 0.2 s. Therefore, the vibration of the helicopter platform is high-frequency vibration. Due to the short wavelength of terahertz, it is very sensitive to the tiny high-frequency vibration of the radar platform. If not processed, this type of motion error will seriously affect the results of THz-SAR imaging. Therefore, when the radar platform has high-frequency vibration, the instantaneous slant range formula (1) from point P on the moving target in the scene to the radar is rewritten as:

[0089]

[0090] The THz-SAR system transmits a linear frequency modulated (LFM) signal for target detection, as shown below:

[0091]

[0092] In formula (4), rect(·) represents the rectangular envelope function, T p Indicates the signal pulse width, fc represents the center frequency of the linear frequency modulation signal, γ represents the modulation frequency, τ=t-mT represents the fast time, T represents the pulse repetition period, m represents the number of pulses, η=mT represents the slow time, and t represents the full time.

[0093] The echo signal of the target received by the radar is:

[0094]

[0095] Due to the high carrier frequency and large bandwidth of the THz-SAR system, the matched filtering method commonly used in the traditional microwave frequency band requires a higher sampling frequency for reception, which not only results in a huge amount of data but also increases the hardware cost. Therefore, the THz-SAR system uses the dechirp method to receive echoes, such as Figure 3 The reference distance is defined as Then the reference signal is:

[0096]

[0097] De-skewing is also called de-slanting reception. It is to perform frequency difference processing on the echo signal formula (5) and the reference signal formula (6). Let R Δ =R(η)-R ref , then its difference frequency output is:

[0098]

[0099] The last term in formula (7) is the residual video phase (RVP) introduced by the Dechirp process. After correcting the RVP term, the distance compressed signal can be obtained, which is expressed as:

[0100]

[0101] In formula (8), f i =-2γR Δ / c represents the instantaneous frequency of distance compression.

[0102] From formula (8), we can get the phase of the azimuth signal is -4πR Δ / λ, the frequency of the azimuth signal can be obtained by taking the phase derivative of the azimuth signal:

[0103]

[0104] Next, in step S110, the self-focusing is set based on the maximum image contrast. According to the theory of compressed sensing, the target echo signal after distance compression can be expressed in the form of a matrix: Where S represents the range image sequence, N is the number of azimuth sampling points, and M is the number of range sampling points. In the ISAR turntable imaging model, the ISAR image can be obtained by directly applying the Inverse Fast Fourier Transform (IFFT) to the range image sequence S. However, for SAR imaging, the signal S needs to be multiplied by the matched filter matrix E in azimuth H Then IFFT is performed in azimuth to obtain a focused SAR image.

[0105] At this time, the range image sequence can be expressed as S=EFX, where S is the received range image echo sequence, is the azimuth phase error matrix, which is associated with the motion parameters of the moving target and the flight trajectory of the carrier aircraft. diag(·) represents a diagonal matrix composed of column vectors. represents the phase error at each pulse. F represents the Fourier transform matrix, that is, The Fourier transform operator can be used to quickly solve the problem. X represents the SAR moving target image to be solved.

[0106] In this embodiment, the maximization image contrast criterion is used to construct the azimuth matched filter matrix E. H , achieving self-focusing imaging of moving targets.

[0107] Specifically, image contrast is a commonly used indicator for evaluating image quality, which is defined as the ratio of the standard deviation of the square of the image amplitude to the mean. Let the mean of the square of the SAR moving target image amplitude be The formula for image contrast is as follows:

[0108]

[0109] In formula (10), the size of the SAR image to be generated is M×N, and the value of each pixel in the image is represented by X(m,n).

[0110] Next, let Calculate the image contrast with respect to the phase error corresponding to the i-th pulse The derivative of is shown in formula (11):

[0111]

[0112] In formula (11), Re(·) indicates taking the real part, and * indicates taking the conjugate.

[0113] make

[0114] At this time, the phase error can be obtained The optimal estimate of is expressed as:

[0115]

[0116] In formula (12), X represents the moving target image to be solved, the size of the moving target image to be solved is M×N, the value of each pixel in the image is expressed as X(m,n), * represents taking conjugate, F H is the Fourier transform matrix. S' i represents the i-th column of the distance image matrix S, expressed as:

[0117]

[0118] Through the above analysis, we can get the azimuth matched filter matrix E of the SAR moving target image X(m,n) H The expression, that is Thereby realizing the function of phase self-focusing.

[0119] Then in step S120, an objective function based on low-rank sparse constraints may be constructed.

[0120] In this embodiment, in the objective function, the nuclear norm, the F norm and the l1 norm are used in sequence to constrain the target range image matrix, the background range image matrix and the moving target image to be solved.

[0121] Specifically, for the range image S, it includes the moving target range image Distance to background That is: S = T + N. The columns of the target range image matrix T are highly correlated, so T has a low-rank feature. The energy of the ground background range image matrix N is scattered in different distance units and is relatively dense, so it can be constrained by the F norm. In order to better extract the moving target and image it, it can be separated from the background based on the above two properties.

[0122] In addition, the moving target image X consists of only a few scattered points compared to the background, and we hope that the imaging result is as sparse as possible, so the l0 norm is used to constrain this sparse matrix.

[0123] The objective function that can be constructed based on the above requirements is expressed as: min rank(T)+α1||N|| F +α2||X||0, where α1 and α2 are regularization parameters.

[0124] However, since the l0 norm constraint in the above objective function leads to an NP-hard problem that cannot find an exact solution, the l1 norm is usually used to replace the l0 norm, making the above optimization problem a solvable convex optimization problem. In addition, the nuclear norm is usually used to represent low-rank constraints. Therefore, the final objective function constructed can be expressed as:

[0125] min||T|| * +α1||N|| F +α2||X||1(14)

[0126] It can be seen that the objective function expressed by formula (14) is a convex optimization problem with three variables, and the ADMM framework cannot be used directly. Therefore, in step S130, the imaging model is introduced as a constraint and auxiliary variable. To make it conform to the standard ADMM framework, we get an objective function to be optimized, which is expressed as:

[0127]

[0128] In formula (15), ||T|| * Indicates the use of nuclear norm to constrain the target distance image matrix, ||N|| F represents the constraint on the background range image matrix using the F norm, ||Z||1 represents the constraint on the auxiliary variable using the l1 norm, α1 and α2 represent regularization parameters. The imaging model is T=EFX.

[0129] Furthermore, when the alternating direction multiplier method is used to transform the objective function to be optimized into multiple sub-problems to be solved iteratively: a Lagrangian multiplier matrix and a penalty term coefficient are introduced, and an augmented Lagrangian function is constructed according to the objective function to be optimized, and the augmented Lagrangian function is transformed into multiple sub-problems to be solved iteratively using the alternating direction multiplier method. In one iterative calculation, multiple sub-problems are included, including sequentially updating the target range image matrix, the background range image matrix, the moving target image to be solved, the auxiliary variables, the azimuth phase error matrix, the Lagrangian multiplier matrix and the penalty term coefficient.

[0130] Specifically, the augmented Lagrangian function constructed according to formula (15) is expressed as:

[0131]

[0132] In formula (16), u1, u2 and u3 represent Lagrange multiplier matrices, and ρ1, ρ2 and ρ3 represent penalty coefficients. Then the ADMM algorithm is used to solve the optimization problem shown in formula (16), which is converted into multiple sub-problems, expressed as:

[0133]

[0134] Next, the sub-problems shown in formula (17) are solved one by one.

[0135] First, update the target range image matrix, omitting the variables unrelated to T in formula (16), and express it as:

[0136]

[0137] It can be seen that formula (18) is a nuclear norm convex optimization problem, which can be solved by a singular value shrinkage operator To solve it, the specific method is as follows:

[0138] make

[0139] Then the update expression of the target distance image matrix T is expressed as:

[0140]

[0141] In formula (20), H (k) =U (k) ·Σ (k) ·(V (k) ) H , U (k) and V (k) They are H (k) The left and right singular vectors, Σ (k) is a diagonal matrix consisting of singular values. represents the soft threshold operator for any variable x, expressed as where sgn(·) is a sign operator.

[0142] Then, the background noise matrix N is updated, and the variables irrelevant to N are omitted in formula (16) and expressed as:

[0143]

[0144] Formula (21) shows a standard F-norm minimization problem, and its analytical solution can be obtained by direct matrix derivation. By calculating the first-order derivative of N and making it equal to 0, the equation can be solved as follows:

[0145]

[0146] Then, the target main part X in the SAR image is updated, and the variables irrelevant to X are omitted in formula (16), which is expressed as:

[0147]

[0148] Similarly, by calculating the derivative of formula (23) with respect to X and setting it equal to 0, the update expression of X can be obtained as:

[0149]

[0150] It can be seen that the update of X involves finding the inverse of an N×N matrix, and its computational complexity is at least Furthermore, note that the full Fourier matrix F is a unitary matrix, that is, F H F=FF H =I, so formula (24) can be further simplified as:

[0151]

[0152] In formula (25), the F matrix and F H The matrix can be quickly calculated by Fast Fourier Transform (FFT) and IFFT respectively to improve the calculation efficiency.

[0153] Further updating the auxiliary variable Z and omitting the terms in formula (16) that are not related to Z, we can obtain:

[0154]

[0155] This is a standard l1 norm minimization problem, which can be directly derived using the soft threshold operator:

[0156]

[0157] Then, the azimuth error matrix E obtained in step S110 is used to perform the autofocusing process. Finally, the update formulas of u1, u2, u3, ρ1, ρ2 and ρ3 are expressed as:

[0158]

[0159] In summary, the solution process of each sub-problem in each iteration of THz-SAR moving target imaging based on ADMM with self-focusing function can be expressed as:

[0160]

[0161] The above-mentioned whole THz-SAR moving target high-resolution imaging method is named as AF-RPCA-ADMM. Meanwhile, in this embodiment, a specific processing flow for implementing AF-RPCA-ADMM is also provided, see Algorithm 1.

[0162]

[0163] In this embodiment, the above-mentioned AF-RPCA-ADMM method can be used to perform multiple iterative calculations on the measured target echo signal (after Dechirp demodulation preprocessing) until it converges, thereby obtaining a high-resolution imaging result of the moving target. In this method, the AF-RPCA-ADMM method is embedded in a neural network, and the neural network is used to solve the high-resolution imaging result of the moving target, so as to overcome the problems of difficult parameter search and time-consuming iterative calculation in the traditional algorithm.

[0164] In step S140, the above-mentioned iterative process of the THz-SAR moving target imaging algorithm based on low-rank sparse constraints is carried out to obtain a moving target self-focusing imaging network AF-RPCA-Net based on PRCA separation, and its network structure is as follows: Figure 4 shown.

[0165] In this embodiment, consistent with the N-times iterative solution process of the AF-RPCA-ADMM algorithm, the moving target self-focusing imaging network includes N layers of neural networks corresponding to the number of iterative calculations for multiple sub-problems. In each layer of the neural network, the target range image matrix, the background range image matrix, the moving target image to be solved, the auxiliary variables, the azimuth phase error matrix, the Lagrange multiplier matrix, and the penalty term coefficient are updated in turn.

[0166] In this embodiment, when training the moving target self-focusing imaging network: multiple moving point targets are simulated according to preset parameters to obtain simulation data, and the range image sequence after Dechirp demodulation of each simulation data is used as training data, and then the range Doppler algorithm imaging result of each moving point target to the static point target is used as the true value label. The moving target self-focusing imaging network is trained using the training data and the true value label.

[0167] Specifically, in the neural network model, the data set plays a vital role, which directly affects the quality of the model. First, a data set of simulated point targets is constructed. For simplicity, the azimuth offset caused by the radial velocity of the target is avoided, the radial velocity of the target is ignored, and only the factors affecting the azimuth defocus of the target are considered. In the 40×40 scene, 1 to 300 azimuth motion speeds are randomly set to [-10, 10] m / s, and the range acceleration range is [-5, 5] m / s. 2 The radar parameters are set as shown in Table 1 for simulation, and 1100 pairs of Dechirp demodulated range image sequences are generated as the input of the network. The range Doppler (RD) algorithm imaging results of the corresponding stationary point targets are used as the labels of the network, of which the first 1000 pairs of data are used for training and the last 100 pairs are used for testing. Figure 5(a) is the rough imaging result of the RD algorithm for a moving target. It can be seen that the imaging result is defocused in the azimuth direction due to the motion of the target. A set of input range image sequences and corresponding label image data pairs are randomly selected from the data set as the display, as shown in Figure 2. Figure 5 As shown in (b) and (c).

[0168] Table 1 Radar simulation parameter setting table

[0169]

[0170] Furthermore, since THz-SAR technology involves highly sensitive data, which are often confidential in nature, there is currently no publicly available measured data set that can be used for the analysis of moving targets. In this embodiment, the THz beam SAR raw data acquired during laboratory flight is used, and the radar parameters are consistent with the simulation, as shown in Table 1, to construct the required moving target imaging data set. The specific approach is as follows: first, RD coarse imaging is performed frame by frame for each piece of data, and a 256×256 area of ​​the moving target in each frame of the image is selected as the ROI, and an IFFT operation is performed on the ROI area to obtain a range image sequence before azimuth compensation, which is the input of the network. Unlike the simulation situation, there is no corresponding static situation for the measured data, so we will temporarily use the image obtained by the sparse recovery algorithm in the previous section as the label. Figure 6 Several sets of network training input distance images and corresponding label images randomly selected from the measured data set are shown.

[0171] In this embodiment, a root mean square error loss function is used to train the moving target self-focusing imaging network.

[0172] Specifically, the loss function is an indicator to measure the effectiveness of the model. The choice of loss function depends on many factors, including whether there are outliers, the choice of objective function and algorithm, the time efficiency of running gradient descent, whether it is easy to find the derivative of a given function, and the confidence of the predicted results. In order to enable the network to learn as much relevant information as possible between the input and label images, the following root mean square error (RMSE) loss function is defined for the network after comprehensive consideration, expressed as:

[0173]

[0174] In formula (29), It represents the imaging result obtained by using the input distance image through the network. represents the corresponding label image, ||·|| Frepresents the F norm of the matrix to ensure that the output image is similar to the label. In addition, the l1 norm is introduced to constrain the output image to better restore the sparse features of SAR moving targets. Among them, λ is a parameter that controls the l1 loss weight.

[0175] Furthermore, once the loss function is determined, the parameters can be automatically updated by using an optimization algorithm. Taking the update of X as an example, Figure 7 This is the process diagram of forward propagation and backward propagation of the network. The whole process is implemented through the chain rule.

[0176] In step S150, the measured target echo signal is subjected to Dechirp demodulation processing to obtain a range-compressed echo signal, which is then input into a trained moving target self-focusing imaging network to obtain a high-resolution SAR imaging result of the moving target.

[0177] In this method, the effectiveness of this method is also proved through experiments.

[0178] In the experiment, a tank target consisting of 28 scattering points was set to test the effectiveness of the algorithm. The simulated scattering point positions are as follows: Figure 8 (a) shown. Figure 8 (b) shows the corresponding range image sequence. Since the target has a velocity component, its range image is blurred. Fig. 9 As shown in Figure 2, the proposed network is compared with several classic imaging algorithms, including Fig. 9 (a) is a schematic diagram of the imaging results of the RDA algorithm. Fig. 9 (b) is a schematic diagram of the imaging results of the PGA algorithm. Fig. 9 (d) is a schematic diagram of the imaging results of the minimum entropy self-focusing ADMM algorithm. Fig. 9 (c) is a schematic diagram of the imaging results of the Orthogonal Matching Pursuit (OMP) algorithm. Fig. 9 (e) is a schematic diagram of the imaging results of the AF-RPCA-ADMM algorithm and Fig. 9 (f) is a schematic diagram of the imaging results of the AF-RPCA-Net algorithm.

[0179] Similarly, we use the measured data for verification. We select another frame of data outside the training set for imaging testing. The original moving target ROI image and the corresponding network input and output are shown in Fig.10 As shown, Fig.10 (a) is the original moving target ROI image, Fig.10 (b) is the range image sequence of the moving target. Fig.10(c) is the label image. The proposed network is compared with several classic imaging algorithms including RDA, the algorithm based on short-time Fourier transform (STFT) motion parameter estimation combined with PGA, OMP algorithm and minimum entropy self-focusing ADMM imaging algorithm. Since the label uses the imaging result of AF-RPCA-ADMM, it is not compared with this algorithm here. Fig.11 As shown, Fig.11 (a) is a schematic diagram of the imaging results of the algorithm based on short-time Fourier transform (STFT) motion parameter estimation combined with PGA. Fig.11 (b) is a schematic diagram of the imaging results of the OMP algorithm. Fig.11 (c) is a schematic diagram of the imaging results of the minimum entropy self-focusing ADMM imaging algorithm. Fig.11 (d) is a schematic diagram of the imaging results of the AF-RPCA-Net algorithm.

[0180] The imaging results of both simulation and measured data show that the imaging results of AF-RPCA-Net are closest to the label image and have the best target background separation ability and focusing effect. The OMP algorithm apparently reduces the background noise, but in fact it only zeroes the weaker scattering points on the original image and cannot restore the moving target. The introduction of the self-focus module significantly improves the effect of the ADMM algorithm in processing the focusing of moving targets. Compared with the traditional algorithms of RDA and STFT+PGA, its imaging quality has been significantly improved, but there is still strong background noise in the image. Table 2 gives a quantitative comparison of PSLR, ISLR, IE and SSIM of the five algorithms of measured data to quantitatively evaluate the performance of the algorithm. The IE index of AF-RPCA-Net is only 3.1433, which is significantly lower than the results of the RD algorithm and the traditional STFT+PGA. In addition, on the Intel(R) Core(TM) i9-13900HXCPU@5.4GHz computing platform, the OMP algorithm based on the unfocused module takes 6.513394s, the AF-RPCA-ADMM takes 4.626387s, and the proposed algorithm only takes 2.2282s, which is faster than the compressed sensing algorithm, further verifying the efficiency of the proposed algorithm. The AF-RPCA-Net algorithm has the advantages of high-quality and fast imaging, which not only reduces the waiting time, but also has important practical significance for application scenarios that require fast response, such as real-time imaging or dynamic monitoring.

[0181] Table 2 Comparison of imaging performance of different algorithms in measured data

[0182]

[0183] In the above-mentioned THz-SAR moving target high-resolution imaging method, a THz-SAR moving target high-resolution learning imaging method based on ADMM-Net is provided. This algorithm designs a deep learning method, namely RPCA-ADMM-Net, for the imaging task of single-channel SAR ground moving targets in order to solve the problems of difficult parameter search and time-consuming iteration in sparse recovery algorithms. This method uses the low rank and sparse characteristics of moving targets in SAR images to separate moving targets from the background, and uses the designed maximum image contrast module to iteratively compensate for the azimuth phase error to achieve fast sparse reconstruction of moving targets. Applying the model-data jointly driven deep learning imaging technology to SAR end-to-end imaging can effectively improve the above-mentioned problems existing in traditional algorithms. The network has strong interpretability while avoiding the need for a large number of samples, giving full play to the advantages of THz-SAR refined imaging.

[0184] 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.

[0185] In one embodiment, Fig.12 As shown, a THz-SAR moving target high-resolution imaging device is provided, comprising: a target echo model construction module 200, a range image sequence model construction module 210, an objective function construction module 220, an alternating direction multiplier method solution module 230, a moving target self-focusing imaging network acquisition module 240 and a SAR high-resolution imaging result acquisition module 250, wherein:

[0186] The target echo model building module 200 is used to build a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar. The target echo model models a range compressed echo signal obtained after demodulation processing using Dechirp.

[0187] The range image sequence model construction module 210 is used to represent the target echo model as a range image sequence in matrix form based on the compressed sensing theory, and to construct the range image sequence model according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion.

[0188] The objective function construction module 220 is used to split the range image sequence into a target range image matrix and a background range image matrix, and construct an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved.

[0189] The alternating direction multiplier method solving module 230 introduces an imaging model and auxiliary variables to constrain the objective function, and uses the alternating direction multiplier method to transform the objective function to be optimized into multiple sub-problems to be solved iteratively.

[0190] The moving target self-focusing imaging network obtaining module 240 is used to expand the process of iteratively solving each sub-problem into a moving target self-focusing imaging network based on robust principal component analysis, and train it to obtain a trained moving target self-focusing imaging network.

[0191] The SAR high-resolution imaging result obtaining module 250 is used to obtain the target echo signal. The target echo signal is obtained by detecting the moving target with the terahertz synthetic aperture radar. The target echo signal is pre-processed and then input into the trained moving target self-focusing imaging network to obtain the SAR high-resolution imaging result of the moving target.

[0192] For the specific definition of the THz-SAR moving target high-resolution imaging device, please refer to the definition of the THz-SAR moving target high-resolution imaging method above, which will not be repeated here. Each module in the above-mentioned THz-SAR moving target high-resolution imaging device 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 of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0193] 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 THz-SAR moving target high-resolution imaging method 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 shell, or an external keyboard, touchpad or mouse, etc.

[0194] 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.

[0195] 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:

[0196] Constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp;

[0197] Based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion;

[0198] Splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved;

[0199] An imaging model and auxiliary variables are introduced to constrain the objective function, and an alternating direction multiplier method is used to transform the objective function into a plurality of sub-problems to be solved iteratively;

[0200] Expanding the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and training the network to obtain a trained moving target self-focusing imaging network;

[0201] A target echo signal is obtained, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, and the target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0202] 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:

[0203] Constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp;

[0204] Based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion;

[0205] Splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved;

[0206] An imaging model and auxiliary variables are introduced to constrain the objective function, and an alternating direction multiplier method is used to transform the objective function into a plurality of sub-problems to be solved iteratively;

[0207] Expanding the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and training the network to obtain a trained moving target self-focusing imaging network;

[0208] A target echo signal is obtained, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, and the target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

[0209] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed 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 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. As an 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).

[0210] The technical features of the above embodiments may be arbitrarily combined. 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.

[0211] 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 THz-SAR moving target high-resolution imaging method, characterized in that: The method comprises: Constructing a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp; Based on the compressed sensing theory, the target echo model is represented as a range image sequence in matrix form, and a range image sequence model is constructed according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion; Splitting the range image sequence into a target range image matrix and a background range image matrix, and constructing an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved; An imaging model and auxiliary variables are introduced to constrain the objective function, and an alternating direction multiplier method is used to transform the objective function to be optimized into a plurality of sub-problems to be solved iteratively; Expanding the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and training the network to obtain a trained moving target self-focusing imaging network; A target echo signal is obtained, where the target echo signal is obtained by detecting a moving target with a terahertz synthetic aperture radar, and the target echo signal is pre-processed and input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

2. The THz-SAR moving target high-resolution imaging method according to claim 1, characterized in that: The azimuth phase error matrix constructed by maximizing the image contrast criterion is expressed as: Where diag(·) represents a diagonal matrix consisting of column vectors, and the phase error The optimal estimate of is expressed as: In the above formula, X represents the moving target image to be solved, the size of the moving target image to be solved is M×N, the value of each pixel in the image is expressed as X(m,n), * represents taking conjugate, S i ' represents the i-th column of the distance image matrix S, F H is the Fourier transform matrix.

3. The THz-SAR moving target high-resolution imaging method according to claim 2, characterized in that: In the objective function, the target range image matrix, the background range image matrix and the moving target image to be solved are constrained by using the nuclear norm, the F norm and the l1 norm in sequence.

4. The THz-SAR moving target high-resolution imaging method according to claim 3, characterized in that: The objective function to be optimized is expressed as: min||T|| * +α1||N|| F +α2||Z||1 S=T+N stT=EFX X=Z In the above formula, ||T|| * Indicates the use of nuclear norm to constrain the target distance image matrix, ||N|| F Indicates that the background range image matrix is ​​constrained by the F norm, ||Z||1 indicates that the auxiliary variables are constrained by the l1 norm, and α1 and α2 represent regularization parameters.

5. The THz-SAR moving target high-resolution imaging method according to claim 4, characterized in that: When the alternating direction multiplier method is used to transform the objective function to be optimized into multiple sub-problems to be solved iteratively: Introducing a Lagrangian multiplier matrix and a penalty term coefficient, and constructing an augmented Lagrangian function according to the objective function to be optimized; Using the alternating direction multiplier method, the augmented Lagrangian function is transformed into a plurality of sub-problems to be solved iteratively; In one iterative calculation, multiple sub-problems are included, including updating the target range image matrix, the background range image matrix, the moving target image to be solved, the auxiliary variables, the azimuth phase error matrix, the Lagrange multiplier matrix and the penalty term coefficients in turn.

6. The THz-SAR moving target high-resolution imaging method according to any one of claims 1 to 5, characterized in that: The moving target self-focusing imaging network includes a multi-layer neural network corresponding to the number of iterative calculations performed on the multiple sub-problems; In each layer of the neural network, the target range image matrix, the background range image matrix, the moving target image to be solved, the auxiliary variables, the azimuth phase error matrix, the Lagrange multiplier matrix and the penalty term coefficient are updated in turn.

7. The THz-SAR moving target high-resolution imaging method according to claim 6, characterized in that: When training the moving target self-focusing imaging network: Simulating multiple moving point targets according to preset parameters to obtain simulation data; Dechirp-demodulated range image sequences of the simulation data are used as training data; The range Doppler algorithm imaging results of the stationary point targets used by each moving point target are used as the true value label; The moving target self-focusing imaging network is trained using the training data and the true value labels.

8. The THz-SAR moving target high-resolution imaging method according to claim 7, characterized in that: The moving target self-focusing imaging network is trained using a root mean square error loss function.

9. A THz-SAR moving target high-resolution imaging device, characterized in that: The device comprises: A target echo model building module is used to build a target echo model obtained by detecting a moving target using a terahertz synthetic aperture radar, wherein the target echo model models a range compression echo signal obtained after demodulation processing using Dechirp; A range image sequence model building module is used to represent the target echo model as a range image sequence in matrix form based on the compressed sensing theory, and to build a range image sequence model according to the azimuth phase error matrix, the Fourier transform matrix and the moving target image to be solved, wherein the azimuth phase error matrix is ​​constructed by adopting the maximization image contrast criterion; An objective function construction module is used to split the range image sequence into a target range image matrix and a background range image matrix, and to construct an objective function according to the target range image matrix, the background range image matrix and the moving target image to be solved; An alternating direction multiplier method solution module, which constrains the objective function by introducing an imaging model and auxiliary variables, and uses an alternating direction multiplier method to convert the objective function to be optimized into a plurality of sub-problems to be solved iteratively; A moving target self-focusing imaging network obtaining module is used to expand the process of iteratively solving each of the sub-problems into a moving target self-focusing imaging network based on robust principal component analysis, and train the network to obtain a trained moving target self-focusing imaging network; The THz-SAR high-resolution imaging result acquisition module is used to obtain a target echo signal, wherein the target echo signal is obtained by detecting a moving target using a terahertz synthetic aperture radar, and the target echo signal is pre-processed and then input into the trained moving target self-focusing imaging network to obtain a SAR high-resolution imaging result of the moving target.

10. 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 8 are implemented.

Images