Rotor target ISAR imaging method based on FGSR low-rank representation
Through the combination of FGSR low-rank characterization and the L-ADMM framework, the problems of high complexity and insufficient accuracy caused by nuclear norm optimization in rotor target ISAR imaging are solved, and efficient micro Doppler inhibition and ISAR enhanced imaging of rotor targets are achieved.
Patent Information
- Application Number
- CN202411001964.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-08
AI Technical Summary
The existing ISAR imaging method based on joint sparse low rank priors has problems with high algorithm complexity and insufficient imaging accuracy in rotor target imaging, especially the fuzzy approximation of matrix rank function caused by kernel norm optimization affects the low target echo. Constrained optimization of rank.
Factor group sparse regularization (FGSR) low-rank characterization is adopted, and the Schatten-1/2 norm relaxed rank function is converted to solve the optimization problem with constraints, and iteratively solves iteratively under the L-ADMM framework to avoid singular value decomposition ( SVD) calculation improves model fitting accuracy.
The micro Doppler effect in the target ISAR imaging results is effectively suppressed, and ISAR enhanced imaging with the target of rotating components is achieved, which improves the algorithm efficiency and imaging quality compared with traditional methods.
Smart Images

Figure CN118915065A_ABST
Abstract
Description
Technical Field
[0001] The disclosed embodiments relate to the field of radar imaging technology, and in particular to an ISAR imaging method for rotor targets based on FGSR low-rank characterization. Background Art
[0002] ISAR (Inverse Synthetic Aperture Radar) imaging has been widely studied and applied in military and civilian fields due to its all-day and all-weather technical advantages. However, since ISAR uses a turntable model for imaging, during the imaging coherent processing interval, the micro-motion of the target (such as the rotation of the propeller of a turboprop aircraft) will produce mD modulation on the echo, resulting in defocusing of the ISAR image.
[0003] In recent years, ISAR imaging methods for micro-moving targets based on sparse priors of targets and imaging scenes have been booming. Fully mining prior information for optimization modeling and accurately solving the problem are the keys to improving ISAR imaging quality. Therefore, ISAR imaging of micro-moving targets has continued to develop along with the evolution of prior modeling and optimization solutions.
[0004] In the related technology, the sparse constraint is combined with the minimum entropy and Laplacian Scale Mixture (LSM) prior to achieve ISAR self-focusing imaging of micro-moving targets. This capability is then optimized by introducing VBI. Then the RPCA method is introduced into the field of ISAR imaging, making the ISAR imaging method of micro-moving targets based on the joint low-rank sparse dual constraints widely studied.
[0005] With the development of multi-constrained optimization problem solving methods in the field of mathematics, optimization solution methods such as Inertial Proximal Alternating Linearized Minimization (iPALM), Accelerated PALM (APALM) and FGSR (Factor Group-Sparse Regularizations) have emerged, providing new possibilities for further promoting the development of ISAR imaging research on micro-motion targets.
[0006] From the existing high-resolution ISAR imaging research on rotor-type targets, on the one hand, most of the existing imaging methods based on joint sparse low-rank priors rely on nuclear norm optimization for modeling and solving. The SVD decomposition in the solution process makes the algorithm complexity difficult to adapt to the requirements of refined imaging of large echo matrices; on the other hand, the nuclear norm is a fuzzy approximation of the matrix rank function. The current optimization model that uses the nuclear norm to constrain the low rank of the target echo has room for improvement in imaging accuracy. Therefore, for the imaging problem of rotor-type targets, it is necessary to establish a suitable joint constraint model and study more accurate and efficient optimization solutions to further improve the ISAR imaging quality and algorithm efficiency.
[0007] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention
[0008] The purpose of the embodiments of the present disclosure is to provide a rotor target ISAR imaging method based on FGSR low-rank characterization, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.
[0009] The present disclosure provides a rotor target ISAR imaging method based on FGSR low-rank characterization, comprising the following steps:
[0010] S100, build an imaging optimization model based on FGSR low-rank representation:
[0011] S101, constructing an echo signal model of a target with a micro-moving component, and performing a sequence representation of the one-dimensional range image HRRP after translation compensation;
[0012] S102, using the multiplication form of the target ISAR imaging result and the partial Fourier matrix to represent the HRRP sequence, and obtaining a target signal model containing a rotating component;
[0013] S103, using the target signal model to model the ISAR imaging problem with rotating components as a low-rank sparse optimization problem;
[0014] S200, low-rank representation and solution based on factor group sparse regularization FGSR:
[0015] S201, using the Schatten-1 / 2 norm to relax the rank function of the low-rank sparse optimization problem to obtain a relaxed rank function;
[0016] S202, using FGSR low-rank representation to convert the relaxed rank function based on Schatten-1 / 2 norm into solving the optimization problem with constraints;
[0017] S203, solving the constrained optimization problem.
[0018] In one embodiment of the present disclosure, the target signal model is established based on three constraints: low rank of the target body echo HRRP sequence, sparsity of the target micro-movement component echo HRRP sequence, and sparsity of the target ISAR image.
[0019] In one embodiment of the present disclosure, the sequence of one-dimensional range profiles HRRP in S101 is expressed as:
[0020]
[0021] In the formula, σ p is the scattering coefficient of the pth scattering point on the target body, σ q is the scattering coefficient of the pth scattering point on the target micro-movement component, B, c and f c Represent signal bandwidth, light speed and carrier frequency respectively, R p (t m ) is the instantaneous rotation distance of the pth scattering point on the target body, R q (t m ) represents the instantaneous rotation distance of the qth scattering point on the target micro-motion component within one CPI, and j represents the imaginary unit.
[0022] In one embodiment of the present disclosure, in S102, the target signal model containing the rotating component is:
[0023] H=L+S
[0024] L=PX
[0025] In the formula, and Represent the ISAR imaging results, part of the Fourier matrix and the HRRP sequence of the target body, respectively. represents the HRRP sequence of the micro-motion component, is the target HRRP sequence; M, K and N represent the total number of pulses, the number of pulses under sparse observation and the number of range units, respectively.
[0026] In one embodiment of the present disclosure, in S103, the ISAR imaging problem with rotating components is modeled as the following low-rank sparse optimization problem:
[0027] min rank(L)+λ||S||1+μ||X||1
[0028] stH=L+S
[0029] L=PX
[0030] Where λ and μ are regularization parameters.
[0031] In one embodiment of the present disclosure, in S201, the relaxed rank function is as follows:
[0032]
[0033] s.tH=L+S
[0034] L=PX
[0035] The result of the rank function factorization based on FGSR is expressed as follows:
[0036]
[0037] Where nnzc(A) represents the number of non-zero columns of matrix A, nnzc(B T ) represents the number of non-zero rows of matrix B, A=[a1,a2,…,a d ],B=[b1,b2,…,b d ].
[0038] In one embodiment of the present disclosure, in S202, the rank function is relaxed using the Schatten-1 / 2 norm. The FGSR method is defined as follows:
[0039]
[0040] In the formula, a j and b j are the column vectors of matrix A and matrix B respectively.
[0041] In one embodiment of the present disclosure, the relaxed rank function based on FGSR is converted into the following optimization problem with constraints:
[0042]
[0043] In one embodiment of the present disclosure, in S203, the constrained optimization problem is rewritten into the following generalized Lagrangian function form:
[0044]
[0045] In the formula, <·,·> represents the inner product of two matrices, ||·|| F represents the F-norm of the matrix, Y1 and Y2 are Lagrange multipliers, ρ1>0 and ρ2>0 are penalty coefficients.
[0046] In one embodiment of the present disclosure, the optimization problem in the form of the generalized Lagrangian function is transformed into the following multiple sub-problems to be solved separately:
[0047]
[0048] Where η is the rising factor and k is the number of iterations.
[0049] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:
[0050] In the disclosed embodiment, a rotor target ISAR imaging method based on FGSR low-rank representation is established. Based on the low rank of the main component echo and the sparsity of the micro-motion component echo in the rotor target ISAR imaging echo HRRP sequence, as well as the sparsity of the target in the ISAR imaging scene, an imaging optimization model containing three regularization constraints is established; then, in order to constrain the low rank of the target main HRRP sequence, the nuclear norm problem is improved by using a method based on FGSR low-rank representation, avoiding the matrix SVD that requires a large number of complex calculations and improving the model fitting accuracy; finally, the optimization problem is iteratively solved based on the L-ADMM framework. The experimental results on simulation and measured data show that the imaging method of the present application can effectively suppress the micro-Doppler effect in the target ISAR imaging results, realize ISAR enhanced imaging of targets containing rotating components, and improve the algorithm efficiency compared with the traditional L-ADMM method based on the nuclear norm. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.
[0052] Figure 1 A schematic diagram showing a flow chart of a rotor target ISAR imaging method based on FGSR low-rank characterization in an exemplary embodiment of the present disclosure;
[0053] Figure 2 A schematic diagram of an ISAR imaging turntable model containing rotating components in an exemplary embodiment of the present disclosure is shown;
[0054] Figure 3 A schematic diagram of simulated aircraft scattering points and an aircraft photo in an exemplary embodiment of the present disclosure are shown;
[0055] Figure 4 The imaging results of the simulation data under different algorithms when the sparseness rate is 75% in the exemplary embodiment of the present disclosure are shown;
[0056] Figure 5 The imaging results of the simulation data under different algorithms when the sparseness rate is 50% in the exemplary embodiment of the present disclosure are shown;
[0057] Figure 6 The imaging results of the simulation data under different algorithms when the sparseness rate is 25% in the exemplary embodiment of the present disclosure are shown;
[0058] Figure 7 The imaging results of the simulation data under different algorithms are shown when the sparse rate is 75% and SNR=5dB in the exemplary embodiment of the present disclosure;
[0059] Figure 8 The imaging results (full pulse data) of the An-26 aircraft data under different algorithms in the exemplary embodiment of the present disclosure are shown;
[0060] Fig. 9 The imaging results of An-26 aircraft data under different algorithms in an exemplary embodiment of the present disclosure are shown (75% sparseness rate). DETAILED DESCRIPTION
[0061] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0062] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.
[0063] This example implementation provides a rotor target ISAR imaging method based on FGSR low-rank characterization, please refer to Figure 1 , may include: step S100-step S200, S100 specifically includes S101-S103, S200 specifically includes S201-S203. The details are as follows:
[0064] S100, build an imaging optimization model based on FGSR low-rank representation:
[0065] S101, constructing an echo signal model of a target with a micro-moving component, and performing a sequence representation of the one-dimensional range image HRRP after translation compensation.
[0066] S102, the HRRP sequence is represented by the multiplication form of the target ISAR imaging result and the partial Fourier matrix, and the target signal model containing the rotation component is obtained. That is, the target main body HRRP sequence is represented by the multiplication form of the target ISAR imaging result and the partial Fourier matrix, and the signal relationship between the ISAR imaging result and the target main body HRRP sequence is established.
[0067] S103, using the target signal model to model the ISAR imaging problem with rotating components as a low-rank sparse optimization problem;
[0068] S200, low-rank representation and solution based on factor group sparse regularization FGSR:
[0069] S201, using FGSR to relax the rank function of the low-rank sparse optimization problem to obtain a relaxed rank function. By utilizing the low rank of the main echo on the HRRP sequence, the sparsity of the target rotating component echo on the HRRP sequence, and the sparsity of the ISAR imaging results, the ISAR imaging problem with rotating components is modeled as a joint three-constrained low-rank sparse optimization problem based on the aforementioned target signal model, making full use of the target signal characteristics and prior information.
[0070] S202, using the Schatten-1 / 2 norm to convert the FGSR-based relaxed rank function into a constrained optimization problem. The Schatten-1 / 2 norm is used to replace the nuclear norm to relax the low-rank constraints in the above optimization problem, and the Schatten-1 / 2 norm is factorized using the FGSR method to convert the above optimization problem model into an optimization problem based on the joint L2 / 1 norm and L1 norm of the FGSR low-rank representation, avoiding the computational consumption caused by the singular value decomposition (SVD).
[0071] S203, solving the optimization problem with constraints. For example, the optimization problem with multiple regularization constraints may be solved iteratively in a linear alternating multiplier method (L-ADMM) framework.
[0072] In this embodiment, an imaging optimization model containing three regularization constraints is established based on the low rank of the main component echo and the sparsity of the micro-motion component echo in the HRRP sequence of ISAR imaging echoes of the rotor target, as well as the sparsity of the target in the ISAR imaging scene; then, in order to solve the nuclear norm problem of the low rank of the main HRRP sequence of the constrained target, an FGSR low-rank representation-based method is used to improve it, avoiding the matrix SVD that requires a large number of complex operations and improving the model fitting accuracy; finally, the optimization problem is iteratively solved based on the L-ADMM framework. The experimental results on simulation and measured data show that the imaging method of the present application can effectively suppress the micro-Doppler effect in the target ISAR imaging results, realize ISAR enhanced imaging of targets containing rotating components, and improve the algorithm speed and imaging quality compared to the traditional L-ADMM solution method based on the nuclear norm.
[0073] The specific process of the rotor target ISAR imaging method based on FGSR low-rank characterization in this application is described below.
[0074] S100, construct an imaging optimization model based on FGSR low-rank representation.
[0075] In S101, the echo signal model of the target with micro-moving components is constructed, and the process of sequentially representing the one-dimensional range image HRRP after translation compensation is as follows:
[0076] Figure 2 The scenario of radar observing a target with micro-moving parts is given. After translation compensation, the ISAR imaging model can be expressed as a two-dimensional plane turntable model. The coordinate system is established with the center O of the target as the origin, P is the pth scattering point on the target, and the coordinates are (x p ,y p ). During the azimuth accumulation imaging processing time, the instantaneous azimuth angle of point P relative to the center of target O is expressed by θ(t m ) indicates that t m =mT (m=0, 1, ..., M-1) represents the slow time, T is the pulse repetition interval (PRI). Under the condition of small angle observation, the target can be approximated as uniform motion. Assuming that the target scattering point rotates at an angular velocity w, the change in angle can be approximated as t m The linear function of Δθ(t m )=θ(t m )-θ(t0)≈ωt m Assume that the scattering point P is rotated by Δθ(t m ) and then arrive at P′, under short-time observation and small angle approximation, cosΔθ(t m )≈1, sinΔθ(t m )≈ωt mTherefore, the instantaneous rotation distance of the pth scattering point on the target can be expressed as:
[0077]
[0078] The micro-moving parts on the target are as follows Figure 2 The qth scattering point on the target micro-motion component is represented by Q, and its coordinates are (x q ,y q ). Assume that the scattered points are centered around the center O′(x O′ ,y O′ ) is ω′, then the instantaneous rotation distance of the qth scattering point on the target micro-motion component within one CPI can be expressed as
[0079]
[0080] Among them, r q , ω′ and θ q represent the rotation radius, angular velocity and angle of the qth scattering point respectively, represents the initial phase of the qth scattering point. Observing the above formula, we find that the cosine term The existence of makes the target echo signal produce micro-Doppler effect, causing the ISAR image to be defocused.
[0081] The high-resolution one-dimensional range profile (HRRP) sequence after translation compensation can be expressed as:
[0082]
[0083] The two sides of the plus sign of the polynomial in Equation 3 represent the echo signals of the target body and the micro-moving component respectively, σ p is the scattering coefficient of the pth scattering point on the target body, σ q is the scattering coefficient of the pth scattering point on the target micro-movement component, B, c and f c Represents signal bandwidth, light speed and carrier frequency respectively. p (t m ) is the instantaneous rotation distance of the pth scattering point on the target body, R q (t m ) represents the instantaneous rotation distance of the qth scattering point on the target micro-motion component within one CPI, and j represents the imaginary unit.
[0084] S102, using the multiplication form of the target ISAR imaging result and the partial Fourier matrix to represent the HRRP sequence, and obtaining a target signal model containing a rotating component.
[0085] During the azimuth accumulation imaging processing time, the main part of the target has only a small rotation angle relative to the radar, so the range image between different pulses changes little, making the HRRP sequence present a low-rank characteristic. At the same time, since the size of the target rotating part is small relative to the main body, the echo signal generated by the rotation only occupies a limited number of distance units, so it is sparse. The target HRRP sequence can be expressed as a superposition of the main signal and the micro-motion signal. The target main body HRRP sequence can be expressed in the form of multiplication of the target ISAR imaging result and the partial Fourier matrix. Therefore, the target signal model containing rotating parts can be expressed as:
[0086]
[0087] in, and Represent the ISAR imaging results, partial Fourier matrix and HRRP sequence of the target body, respectively. represents the HRRP sequence of the micro-motion component, is the target HRRP sequence. M, K, and N represent the total number of pulses, the number of pulses under sparse observation, and the number of range units, respectively. In the sparse observation scenario, the number of available pulses is less than the total number of pulses, that is, K < M. Therefore, the reconstruction of ISAR images becomes an underdetermined problem with infinite solutions.
[0088] S103, using the target signal model to model the ISAR imaging problem with rotating components as a low-rank sparse optimization problem.
[0089] According to the above properties of each signal component, the ISAR imaging problem with rotating components can be modeled as a low-rank sparse optimization problem (i.e., a joint constraint model):
[0090]
[0091] Among them, λ and μ are regularization parameters. When solving low-rank constrained problems, the rank function is usually relaxed to the nuclear norm, while when solving sparse constrained problems, the L1 norm is often used instead of the L0 norm, because the optimization problem involving the L0 norm is often an NP-hard problem.
[0092] S200, the process of low-rank representation and solution based on factor group sparse regularization FGSR is as follows.
[0093] There have been extensive studies on the solution of the above constraint problems. However, it should be noted that, on the one hand, the solution of the nuclear norm constraint requires the use of SVD decomposition, which will consume a lot of computing resources and cannot meet the requirements of fine imaging of large echo data matrices; on the other hand, the direct approximation of the matrix rank using the nuclear norm is still not accurate enough, which affects the quality of ISAR enhanced imaging. Therefore, it is of great practical significance to improve the nuclear norm constraint model in the optimization model and improve the computational speed and accuracy of the model solution. FGSR low-rank representation has shown great potential in solving the above problems.
[0094] S201, using FGSR to relax the rank function of the low-rank sparse optimization problem to obtain a relaxed rank function. The specific process is as follows.
[0095] First, we introduce the general representation of the nuclear norm, namely the Schatten-p norm:
[0096]
[0097] When p = 1, Equation 6 becomes That is, the nuclear norm. When 0<p<1, is a non-convex relaxation of rank(L). When p = 0, It can be seen that when p→0, Therefore, in the RPCA problem, when 0<p<1, the Schatten-p norm can better replace the rank function than the nuclear norm.
[0098] S202, using the Schatten-1 / 2 norm, the FGSR-based relaxed rank function is converted into a constrained optimization problem, and the specific process is as follows.
[0099] When 0<p<1, ||L|| Sp is non-convex, so it is impossible to ensure a global optimal solution. In addition, whether solving the Schatten-p norm or the nuclear norm, the matrix needs to be decomposed by SVD in each iteration, which will make it impossible to solve large-dimensional matrices. To solve this problem, a low-rank representation method without SVD decomposition, namely Factor Group Sparse Regularization (FGSR), is used to approximate the rank function. Based on FGSR, the rank function can be relaxed as:
[0100]
[0101] The result of the rank function factorization based on FGSR is expressed as follows:
[0102]
[0103] Among them, nnzc(A) represents the number of non-zero columns of matrix A. Similarly, nnzc(B T ) represents the number of non-zero rows of matrix B. A=[a1,a2,…,a d ],B=[b1,b2,…,b d ], a j and b j is a column vector. For the Schatten-1 / 2 norm:
[0104]
[0105] The SVD decomposition of L can be expressed as The main diagonal elements of Γ are the singular values of the matrix L, U L and are the left and right singular matrices respectively. and When , the equality in equation (9) holds. Therefore, based on the Schatten-1 / 2 norm, the rank function is relaxed. The representation of is defined by the FGSR method as follows:
[0106]
[0107] in, Therefore, the optimization problem based on FGSR relaxation can be rewritten as
[0108]
[0109] S203, solving the constrained optimization problem, the specific process is as follows.
[0110] The constrained optimization problem shown in formula (11) can be rewritten as the following generalized Lagrangian function:
[0111]
[0112] Among them, <·,·> represents the inner product of two matrices, ||·|| F represents the F norm of the matrix, Y1 and Y2 are the Langrange multipliers, ρ1>0 and ρ2>0 are the penalty coefficients. The above optimization problem can be transformed into multiple sub-problems as shown in formula (13) and solved separately:
[0113]
[0114] Where η is the rising factor and k is the number of iterations.
[0115] 1) Solve A (k+1) :
[0116] A (k+1)The objective function can be written as:
[0117]
[0118] Because B (k) The existence of , it is necessary to linearly expand equation (14) to approximate it as:
[0119]
[0120] where β1 is the proximal operator parameter for updating A, yes:
[0121]
[0122] In A (k) The gradient at:
[0123]
[0124] According to the proximal gradient projection, A can be updated as follows:
[0125]
[0126] in, is the group soft threshold function, β1 is the threshold parameter. For a matrix K threshold γ, It can be expressed as:
[0127]
[0128] K M. represents the Mth row of matrix K.
[0129] 2) Solve B (k+1) :
[0130] B (k+1) The objective function is:
[0131]
[0132] Similar to updating A, B can be updated as follows:
[0133]
[0134] Among them, β2 is the proximal operator parameter of updating B, yes:
[0135]
[0136] In B (k) The gradient at:
[0137]
[0138] 3) Solve for S (k+1) :
[0139] S (k+1) The objective function is:
[0140]
[0141] Formula (22) is the L1 norm minimization problem, and its solution can be obtained by the soft threshold function:
[0142]
[0143] in, is a soft threshold function, for scalars z and v:
[0144]
[0145] 4) Solve for X (k+1) :
[0146] X (k+1) The objective function can be written as:
[0147]
[0148] Due to the existence of P, equation (25) cannot be solved directly, so it is still necessary to approximate the problem equation (25) by linearizing the quadratic term of the objective function. According to the linearization derivation of the quadratic term, equation (25) can be rewritten as:
[0149]
[0150] Among them, β3 is the proximal operator parameter for updating x, yes:
[0151] In X (k) The gradient at:
[0152]
[0153] According to the proximal gradient projection, X can be updated as follows:
[0154]
[0155] 5) Solution and
[0156] Y1 and Y2 can be updated as follows:
[0157]
[0158] So far, ISAR imaging of targets with micro-moving parts based on low-rank sparse decomposition can be achieved by alternately solving the above sub-problems. The algorithm flow is shown in Table 1.
[0159] Table 1 ISAR imaging method of targets with micro-moving parts based on FGSR
[0160]
[0161]
[0162] The following experiment is carried out according to the above method and the experimental results are analyzed
[0163] The imaging method of the present application is compared with the traditional RD imaging method, RPCA method and L-ADMM method based on nuclear norm. Experiments are carried out on simulated data and measured data at different sparsity rates and signal-to-noise ratios to compare the performance of the algorithm. IE and IC are used to evaluate the imaging quality, and the calculation time is used to compare the efficiency of the algorithm. The computer configuration used for the simulation experiment is the 12th generation Intel (R) Core (TM) i7-12700, main frequency 2.10GHz, DDR4 memory 32.0GB, NVIDIA GeForce RTX 3070, Windows 11, MATLAB 2022a.
[0164] 1. Simulation Data Experiment
[0165] Please refer to Figure 3 , Figure 3 The schematic diagram of the simulated aircraft scattering points and the optical photos of the An-26 aircraft are given. The An-26 is a twin-engine turboprop aircraft. For the sake of comparison, the two propellers of the simulated aircraft are set to rotate at an angular velocity of 12πrad / s. The HRRP sequence generated by the simulation data and the ISAR images obtained by different algorithms are shown in Figure 2. Figure 4-Figure 6 As shown, they correspond to the sparse rates of 75%, 50% and 25% respectively.
[0166] The comparison results of the simulation data experimental evaluation indicators at 75% sparsity rate are shown in Table 2.
[0167] Table 2 Comparison of evaluation indicators of simulation data experiment (75% sparse rate)
[0168]
[0169] The comparison results of the simulation data experimental evaluation indicators at 50% sparsity rate are shown in Table 3.
[0170] Table 3 Comparison of evaluation indicators of simulation data experiments (50% sparse rate)
[0171]
[0172] The comparison results of the simulation data experimental evaluation indicators at 25% sparsity rate are shown in Table 4.
[0173] Table 4 Comparison of evaluation indicators of simulation data experiment (25% sparse rate)
[0174]
[0175] The following simulation data experiment is conducted with a sparse rate of 75% and SNR = 5dB. The experimental images are shown in Figure 7 ,The experimental comparison data are shown in Table 5.
[0176] Table 5 Comparison of simulation data experimental evaluation indicators (75% sparsity rate, SNR = 5dB)
[0177]
[0178]
[0179] Depend on Figure 7 The imaging results corresponding to the sparsity rate of 75% and the SNR of 5dB show that, at the same sparsity rate, the ISAR images obtained by the traditional RD algorithm and the RPCA algorithm are severely defocused, while the Nuclear-L-ADMM algorithm and the FGSR-L-ADMM algorithm of the present application can remove most of the influence of the Doppler effect and present a relatively clean ISAR image. This verifies the effectiveness of the algorithm of the present application. Comparing the imaging results under different sparsity rates, as the sparsity rate decreases, the imaging quality of the four algorithms decreases. When the sparsity rate drops to 25%, the main part of the target obtained by the four algorithms is severely defocused and difficult to identify, but the Nuclear-L-ADMM and the FGSR-L-ADMM algorithms of the present application can obtain a clearer wing image. In addition, when noise is added at a sparsity rate of 75% (SNR = 5dB), the FGSR-L-ADMM algorithm of the present application can obtain the clearest ISAR image compared with the other three algorithms, which verifies the effectiveness of the method of the present application in a noisy environment.
[0180] In order to more intuitively compare the performance of several algorithms, this application introduces three evaluation indicators: IE, IC and operation time. The specific values are shown in Tables 2-5. It is not difficult to see from the comparison results that the FGSR-L-ADMM algorithm of this application has the lowest IE and the highest IC at different sparsity rates, indicating that the FGSR-L-ADMM algorithm has the highest imaging quality compared with the other three algorithms. When considering the operation efficiency, although the FGSR-L-ADMM algorithm of this application takes more time than the traditional RD algorithm, it is more efficient than the other two iterative optimization algorithms.
[0181] The following is an experiment with measured data.
[0182] The measured An-26 echo data is used to verify the effectiveness of the algorithm of this application. The radar transmits a linear frequency modulation signal with a bandwidth of 400MHz and a carrier frequency of 5.52GHz. 512 pulses are selected as the full-aperture data of ISAR imaging, and each pulse contains 256 sampling points. To facilitate algorithm processing, the 512 pulses are downsampled to 256 pulses. Figure 8 and 9 The HRRP sequence of the An-26 aircraft at full aperture and 75% sparse rate and the ISAR imaging results under various algorithms are shown respectively. Table 6 and Table 7 are the comparison of experimental evaluation indicators of measured data (original data) and the comparison of experimental evaluation indicators of measured data (75% sparse rate). Figure 8 , Fig. 9 As can be seen from Tables 6 and 7, the micro-Doppler effect produced by the aircraft propeller is mainly concentrated in the 102nd and 116th distance units, which seriously affects the readability of the ISAR image. The traditional RD algorithm and the classic RPCA algorithm cannot eliminate the micro-Doppler effect very well, while the Nuclear-L-ADMM and the FGSR-L-ADMM algorithms of this application can remove most of the micro-Doppler effect, significantly improving the visual effect of the image. Compared with the Nuclear-L-ADMM algorithm, the ISAR image obtained by FGSR-L-ADMM is more sparse and the background is clearer. This further verifies the effectiveness of the algorithm of this application.
[0183] The FGSR-L-ADMM algorithm of the present application can not only remove the micro-Doppler effect in ISAR images, but also achieve the lowest IE and the highest IC under all conditions, indicating that it has robust imaging capabilities. In addition, compared with the Nuclear-L-ADMM algorithm, the algorithm of the present application has similar micro-Doppler removal capabilities and higher computational efficiency.
[0184] Table 6 Comparison of experimental evaluation indicators of measured data (original data)
[0185]
[0186] Table 7 Comparison of experimental evaluation indicators of measured data (75% sparse rate)
[0187]
[0188] This application proposes an ISAR imaging method based on FGSR low-rank representation for the imaging problem of medium and high altitude rotor targets. First, based on the low rank of the main components and the sparsity of the micro-motion components in the HRRP sequence of ISAR imaging echoes of rotor targets and the sparsity of targets in the ISAR imaging scene, an imaging optimization model containing three regularization constraints is established; then, in order to solve the nuclear norm problem of the low rank of the main HRRP sequence of the constrained target, an improvement based on FGSR low-rank representation is adopted to avoid the SVD that requires a large number of complex calculations and improve the model fitting accuracy; finally, the optimization problem is iteratively solved based on the L-ADMM framework. Experiments show that the method of this application can effectively suppress the micro-Doppler effect in the target ISAR imaging results, realize ISAR enhanced imaging of targets with rotating components, and improve the algorithm efficiency compared with the traditional L-ADMM method based on the nuclear norm.
[0189] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.
Claims
1. A rotor target ISAR imaging method based on FGSR low-rank characterization, characterized in that: The following steps are involved: S100, build an imaging optimization model based on FGSR low-rank representation: S101, constructing an echo signal model of a target with a micro-moving component, and performing a sequence representation of the one-dimensional range image HRRP after translation compensation; S102, using the multiplication form of the target ISAR imaging result and the partial Fourier matrix to represent the HRRP sequence, and obtaining a target signal model containing a rotating component; S103, using the target signal model to model the ISAR imaging problem with rotating components as a low-rank sparse optimization problem; S200, low-rank representation and solution based on factor group sparse regularization FGSR: S201, using the Schatten-1 / 2 norm to relax the rank function of the low-rank sparse optimization problem to obtain a relaxed rank function; S202, using FGSR low-rank representation to convert the relaxed rank function based on Schatten-1 / 2 norm into solving the optimization problem with constraints; S203, solving the constrained optimization problem.
2. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 1, characterized in that: The target signal model is established based on three constraints: low rank of the target main echo HRRP sequence, sparseness of the target micro-movement component echo HRRP sequence, and sparseness of the target ISAR image.
3. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 1, characterized in that: The sequence of one-dimensional distance image HRRP in S101 is expressed as: In the formula, σ p is the scattering coefficient of the pth scattering point on the target body, σ q is the scattering coefficient of the pth scattering point on the target micro-movement component, B, c and f c Represent signal bandwidth, light speed and carrier frequency respectively, R p (t m ) is the instantaneous rotation distance of the pth scattering point on the target body, R q (t m ) represents the instantaneous rotation distance of the qth scattering point on the target micro-motion component within one CPI, and j is an imaginary unit.
4. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 2, characterized in that: In S102, the target signal model containing rotating parts is: H=L+S L=PX In the formula, and Represent the ISAR imaging results, part of the Fourier matrix and the HRRP sequence of the target body, respectively. represents the HRRP sequence of the micro-motion component, is the target HRRP sequence; M, K and N represent the total number of pulses, the number of pulses under sparse observation and the number of range units, respectively.
5. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 4, characterized in that: In S103, the ISAR imaging problem with rotating components is modeled as the following low-rank sparse optimization problem: min rank(L;)+λ||S||1+μ||X||1 stH=L+S L=PX Where λ and μ are regularization parameters.
6. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 5, characterized in that: In S201, the relaxed rank function is as follows: stH=L+S L=PX The result of the rank function factorization based on FGSR is expressed as follows: Where nnzc(A) represents the number of non-zero columns of matrix A, nnzc(B T ) represents the number of non-zero rows of matrix B, A=[a1,a2,…,a d ],B=[b1,b2,…,b d ].
7. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 6, characterized in that: In S202, FGSR is used to relax the rank function based on the Schatten-1 / 2 norm. The method for representation is defined as follows: In the formula, a j and b j are the column vectors of matrix A and matrix B respectively.
8. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 7, characterized in that: The relaxed rank function based on FGSR is converted to the following optimization problem with constraints:
9. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 8, characterized in that: In S203, the constrained optimization problem is rewritten into the following generalized Lagrangian function form: In the formula, <·,·> represents the inner product of two matrices, ||·|| F represents the F-norm of the matrix, Y1 and Y2 are Lagrange multipliers, ρ1>0 and ρ2>0 are penalty coefficients.
10. The rotor target ISAR imaging method based on FGSR low-rank characterization according to claim 9, characterized in that: The optimization problem in the form of the generalized Lagrangian function is transformed into the following multiple sub-problems for separate solution: Where η is the rising factor and k is the number of iterations.
Citation Information
Cited By
Noise robust micro-motion echo separation method and device based on multi-domain joint constraint
CN120908803A
