Sparse Aperture ISAR Maneuvering Target Imaging and Calibration Method

Abstract

The present invention discloses a sparse aperture ISAR maneuvering target imaging and calibration method, comprising the following steps: establishing a maneuvering target ISAR sparse aperture signal model, characterizing the maneuverability of the target by constructing a variable scale non-uniform Fourier two-dimensional dictionary containing MTRC; constructing a parameterized sparse imaging model by maximum a posteriori estimation, expressed as a cost function of joint target parameter estimation and sparse imaging; finding the target rotation parameter by the minimum entropy method, and implementing a rough estimation of the maneuvering target rotation parameter by performing parameter iteration by the coordinate descent method; and implementing high-quality imaging with MTRC correction and accurate parameter estimation by combining sparse ISAR imaging and target rotation parameter estimation, decomposing into sparse imaging and dictionary learning, and iterating and solving in sequence. The present invention realizes the use of sparse aperture echo data to solve the problem of high-resolution imaging of ISAR maneuvering targets, greatly increasing the implementation scope and application scenarios of radar imaging.

Description

Technical Field

[0001] The present invention relates to a sparse aperture ISAR high-resolution imaging and calibration technology for maneuvering targets, and belongs to the field of radar signal processing technology. Specifically, the invention aims to accurately estimate the rotation parameters of maneuvering targets to complete the two-dimensional MTRC correction of maneuvering targets, thereby realizing high-resolution imaging, and is suitable for sparse aperture ISAR high-resolution imaging and calibration of maneuvering targets. Background Art

[0002] Inverse synthetic aperture radar (ISAR) can provide two-dimensional (2D) high-resolution images of non-cooperative moving targets and has become an important type of remote sensing application. In order to achieve high-resolution azimuth imaging, ISAR needs to obtain continuous sampling of large lateral angle changes between the radar and the target, which may be difficult to meet in some practical applications. In general, multi-function radar systems usually have the ability to search, detect, track and image multiple targets, and need to use a time-sharing mechanism to optimally allocate system resources for different tasks. In this case, due to the limitation of multi-task operation, broadband signal pulses that are continuously emitted and collected over a long period of time will not achieve ISAR imaging. For example, a radar beam tracks the observed target by emitting narrowband signal pulses that are scanned at a low speed. Therefore, the discontinuity of broadband measurements forms a sparse aperture (SA). Multi-source interference often corrupts the echo data. In addition, in network radar systems, the measurement data collected by multiple sensors distributed at different angles are often discontinuous. Therefore, the task of achieving high-resolution ISAR imaging through SA data synthesis or spectrum estimation is crucial to promote the development of modern radar systems.

[0003] Due to the motion and maneuvering characteristics of maneuvering targets, their echo signal models have characteristics such as two-dimensional unit migration. For maneuvering targets, it is difficult to obtain well-focused images using conventional range-Doppler algorithms. In view of the maneuvering non-stationary characteristics of target signals, existing literature has proposed some effective algorithms to solve such problems, such as time-frequency methods, decomposition algorithms based on Chirp signals, etc. However, most of these algorithms do not take into account the effects of sparse aperture and two-dimensional unit migration, and their imaging performance will decrease or even no longer be applicable under sparse aperture. Therefore, the study of sparse aperture ISAR high-resolution imaging of maneuvering targets has become particularly important. Summary of the invention

[0004] The purpose of the present invention is to provide a sparse aperture ISAR maneuvering target imaging and calibration method, which can correct the two-dimensional cross-range migration of the maneuvering target by estimating the rotation parameters of the maneuvering target, thereby realizing sparse aperture ISAR high-resolution imaging and calibration.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A sparse aperture ISAR maneuvering target imaging and calibration method comprises the following steps:

[0007] Step 1: Establish a maneuvering target ISAR sparse aperture signal model, and construct a variable-scale non-uniform Fourier two-dimensional dictionary containing MTRC to characterize the target maneuverability;

[0008] Step 2: construct a parameterized sparse imaging model through maximum a posteriori estimation, which is expressed as a cost function of joint target parameter estimation and sparse imaging;

[0009] Step 3: Find the target rotation parameters by using the minimum entropy method, and perform parameter iteration by using the coordinate descent method to achieve a rough estimate of the maneuvering target rotation parameters;

[0010] Step 4: By combining sparse ISAR imaging and target rotation parameter estimation, the two processes are decomposed into sparse imaging and dictionary learning, which are solved iteratively in turn to achieve high-quality imaging with MTRC correction and accurate parameter estimation.

[0011] In step 1, the causes of two-dimensional cross-distance migration are first analyzed:

[0012] Assume that the radar system transmits linear frequency modulation (LFM) signal pulses and uses matched filtering technology to perform distance compression. The echo data compressed in the distance dimension is expressed as

[0013]

[0014] in, R p =R0+y p , is the fast time in the distance dimension, t m is the slow time in the azimuth dimension, x p and p They represent the horizontal and vertical coordinates of the scattering point p in the turntable model, R p is the distance between the origin of the coordinate system and the radar, ω is the angular velocity of the maneuvering target in the turntable model, j is a unit imaginary number, σ p is the scattering coefficient of the scattering point p, B and f c are the bandwidth and carrier frequency of the transmitted signal, respectively, and c represents the transmission speed of the electromagnetic wave;

[0015] In equation (1), only the linear range offset in the envelope is retained as the range-dimensional MTRC, while the quadratic term is too small to generate the range-dimensional MTRC. The linear frequency modulation and quadratic phase in equation (1) correspond to the azimuth-dimensional MTRC, which can be decomposed into components related to the range and components related to the azimuth. Then equation (1) is transformed to the range-frequency domain as follows:

[0016]

[0017] Among them, f r is the frequency domain representation of the distance-fast time of the echo signal; from formula (2), we can see that the mobility of the target causes non-uniform spatial sampling, that is, it brings about chirp modulation related to the azimuth, which is very different from the echo under non-maneuvering targets. In essence, formula (2) is r ,t m ) domain exhibits unique characteristics of MTRC.

[0018] In the step 1, the method of constructing a variable-scale non-uniform Fourier two-dimensional dictionary including MTRC is: modeling a sparse aperture (SA) discrete signal:

[0019]

[0020]

[0021] Where a and s represent the ISAR image and SA data to be restored, respectively. A is the matrix form of the ISAR image, with the rows being the azimuth dimension and the columns being the distance dimension. n is the noise. vec() is a matrix vector operation, which means that the SA data of I sub-apertures are connected one by one according to the aperture number to form a long vector. represents the total number of pulses of sparse aperture echo data, where L i is the number of pulses of the i-th sub-aperture, N represents the echo data sampling length in the range dimension, k represents the k-th pulse in the echo pulse sequence, n represents the n-th sampling data in an echo pulse, and j is a unit imaginary number; is the quadratic phase error related to distance in equation (1) or (2), F r represents the Fourier transform of the distance dimension, F r H Indicates F r The conjugate transpose of; In formula (4), the diagonal elements of E are all non-zero, and the others are all zero; The element at n∈[1:N],m∈[1:M], where m and m' are the mth and m'th pulses of FA and SA under the same pulse, respectively; in addition, Θ represents the scaled non-uniform Fourier transform of the azimuthal dimension, where and are scaling factor and uneven step size, respectively. Here F s is the sampling frequency in the distance dimension; when d(m)=1, equation (4) can be simplified to the signal model of non-maneuvering target.

[0022] In step 2, the SA-ISAR image is sparsely represented as:

[0023]

[0024]

[0025] Where a is the ISAR image to be restored, s is the sparse aperture echo data (SA data), is the total number of pulses of sparse aperture echo data, L i is the number of pulses of the ith subaperture, a i is the i-th pixel data of the ISAR image to be restored, η is the inverse of the variance of each component of a, is the variance of n, p(s|a) represents the likelihood function distribution model of s with respect to a, p(a|η) represents the sparse prior probability distribution model of a, and the final formula for SA-ISAR image generation is obtained by using the maximum a posteriori estimation:

[0026]

[0027] in, is the estimated value of the ISAR image a to be restored, is the estimated value of the phase error E, is the estimate of the scaled non-uniform Fourier transform Θ in the azimuthal dimension, is the coefficient that characterizes the sparse constraint. In equation (7), sparse aperture imaging and rotation parameter estimation are performed simultaneously by considering the non-cooperative motion of the target.

[0028] In step 3, the rotation parameters are found by the minimum entropy method:

[0029]

[0030] in

[0031]

[0032] Where a(n,k) is the (n,k)th element of the ISAR image of a, where α = 2f c ω 2 / c, Δy and Δk correspond to the rotation center offset caused by translation compensation, These unknown parameters are represented by is the estimate of α, is the estimate of β, is the estimate of Δy, is an estimate of Δk, using express estimates;

[0033] To compensate for the imaging error, in each iteration of the coordinate descent method, The other parameters in are only estimated In this way, we estimate one parameter in turn. For each parameter in, each such process is defined as an iteration; define is the estimate of the first q-1 parameters of the i-th iteration, The qth parameter estimate of

[0034]

[0035]

[0036] in, and yes The estimated values ​​at the (i-1)th and i-th iterations, respectively, and They are about The gradient vector and Hessian matrix of (i,q) is the search step size of the quasi-Newton method based on the Armijo criterion estimation; Using equations (10) and (11), the parameters are updated at each iteration. To achieve convergence, finally get a final estimated parameter.

[0037] In step 4, combined with sparse ISAR imaging, it is decomposed into sparse imaging and dictionary learning, which are respectively expressed as:

[0038]

[0039] and

[0040]

[0041] In formula (12), To solve the non-differentiability of ||a||1, δ is a positive constant close to infinitesimal. In the present invention, |a i | 2 1 / 1000 of the average value;

[0042] First, use the quasi-Newton method to solve equation (12):

[0043]

[0044] in and are the estimated values ​​of a at the i-th and (i+1)-th iterations, respectively, and is an approximate Hessian matrix; and are the estimated values ​​of E and Θ at the i-th iteration;

[0045] Considering The dimension of is large, and its inverse usually requires a large amount of calculation. Convert equation (14) to solve the following problem

[0046]

[0047] Use the conjugate gradient method to solve equation (15);

[0048] Then, equation (13) is transformed into The estimate of

[0049]

[0050] In fact, equation (16) is solved by gradient descent method or linear search method;

[0051] At the i-th iteration, the dynamic search is set at and Nearby, including and is ω and The estimated value in the (i-1)th iteration; after one iteration of parameter estimation, use and The estimate of ω is used to update the dictionary of E and Θ, and then the next sparse signal imaging iteration is performed using equation (15); thus, when the iterations of equations (15) and (16) converge, high-quality SA imaging with MTRC correction and accurate rotation parameter estimation can be achieved. The final iterative value of ω is used to calibrate the restored ISAR image in the azimuth dimension.

[0052] Beneficial effects: In order to effectively suppress the image "defocus" and "blur" caused by incomplete data and solve the two-dimensional range migration problem of maneuvering targets, the present invention proposes a parameterized sparse imaging method, which constructs a maneuvering target imaging signal model, jointly realizes sparse imaging and target parameter estimation, and obtains a coherent, two-dimensional migration corrected calibration image. This sparse imaging and calibration method has good robustness, can effectively suppress the influence of noise, and has a higher performance improvement than traditional algorithms.

[0053] Compared with the existing methods, the present invention has the following advantages:

[0054] First, the present invention realizes sparse aperture ISAR imaging of maneuvering targets, greatly increasing the realization scope and application scenarios of radar imaging.

[0055] Second, the present invention realizes accurate parameter estimation and uses the accurately estimated parameters to calibrate the restored ISAR image in the azimuth dimension.

[0056] Third, the sparse imaging and calibration method in the algorithm of the present invention has good robustness, can effectively suppress the influence of noise, has a high tolerance to strong noise or clutter, and has a higher performance improvement compared with traditional algorithms.

[0057] Fourth, the present invention uses the target-background ratio (TBR) indicator to accurately evaluate the error in parameter estimation, quantitatively evaluates the algorithm performance, and provides accurate data for comparison with the performance advantages of traditional algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of the present invention;

[0059] Figure 2 is the target model of Yak42 aircraft; (a) is a two-dimensional model, (b) is the ISAR image realized by using RD algorithm on FA data, (c) is half of FA data, and (d) is one quarter of FA data;

[0060] Figure 3 It is an ISAR image reconstructed using the algorithm of the present invention on simulated data;

[0061] Figure 4 It is an ISAR image reconstructed using traditional algorithms;

[0062] Figure 5 It is the measured data of Yak42 plane; among them, (a) is the distance envelope after translation compensation; (b) is the partial enlarged interval of (a); (c) is the RD image without two-dimensional MTRC correction; (d) is the ISAR image reconstructed from FA data using the algorithm proposed in this invention.

[0063] Figure 6 The algorithm of the present invention is used to perform SA-ISAR imaging on measured data; wherein (a) and (b) are the ISAR images reconstructed by SA1 and SA2 respectively; (c) and (d) are the aligned distance envelopes of SA1 and SA2 respectively. DETAILED DESCRIPTION

[0064] The present invention will be further described below in conjunction with the accompanying drawings.

[0065] Reference Figure 1 , a sparse aperture ISAR maneuvering target imaging and calibration method, comprising the following steps:

[0066] Step 1: Establish a maneuvering target ISAR sparse aperture signal model, and construct a variable-scale non-uniform Fourier two-dimensional dictionary containing MTRC to characterize the target maneuverability;

[0067] Analysis of the causes of two-dimensional cross-distance migration:

[0068] Assume that the radar system transmits linear frequency modulation (LFM) signal pulses and uses matched filtering technology to perform distance compression. The echo data compressed in the distance dimension is expressed as

[0069]

[0070] in, R p =R0+y p , is the fast time in the distance dimension, t m is the slow time in the azimuth dimension, x p and p They represent the horizontal and vertical coordinates of the scattering point p in the turntable model, R p is the distance between the origin of the coordinate system and the radar, ω is the angular velocity of the maneuvering target in the turntable model, j is a unit imaginary number, σ p is the scattering coefficient of the scattering point p, B and f c are the bandwidth and carrier frequency of the transmitted signal, respectively, and c represents the transmission speed of the electromagnetic wave;

[0071] In equation (1), only the linear range offset in the envelope is retained as the range-dimensional MTRC, while the quadratic term is too small to generate the range-dimensional MTRC. The linear frequency modulation and quadratic phase in equation (1) correspond to the azimuth-dimensional MTRC, which can be decomposed into components related to the range and components related to the azimuth. Then equation (1) is transformed to the range-frequency domain as follows:

[0072]

[0073] Among them, f r It is the frequency domain representation of the distance-fast time of the echo signal; From formula (2), we can see that the mobility of the target causes non-uniform spatial sampling, that is, it brings about chirp modulation related to the azimuth, which is very different from the echo under non-maneuvering targets. In essence, formula (2) is r ,t m ) domain exhibits unique characteristics of MTRC.

[0074] The method of constructing a scaled non-uniform Fourier two-dimensional dictionary including MTRC is to model the sparse aperture (SA) discrete signal:

[0075]

[0076]

[0077] Where a and s represent the ISAR image and SA data to be restored, respectively. A is the matrix form of the ISAR image, with the rows being the azimuth dimension and the columns being the distance dimension. n is the noise. vec() is a matrix vector operation, which means that the SA data of I sub-apertures are connected one by one according to the aperture number to form a long vector. represents the total number of pulses of sparse aperture echo data, where L i is the number of pulses of the i-th sub-aperture, N represents the echo data sampling length in the range dimension, k represents the k-th pulse in the echo pulse sequence, n represents the n-th sampling data in an echo pulse, and j is a unit imaginary number; is the quadratic phase error related to distance in equation (1) or (2), F r represents the Fourier transform of the distance dimension, F r H Indicates F r The conjugate transpose of; In formula (4), the diagonal elements of E are all non-zero, and the others are all zero; The element at n∈[1:N],m∈[1:M], where m and m' are the mth and m'th pulses of FA and SA under the same pulse, respectively; in addition, Θ represents the scaled non-uniform Fourier transform of the azimuthal dimension, where and are scaling factor and uneven step size, respectively. Here F s is the sampling frequency in the distance dimension; when d(m)=1, equation (4) can be simplified to the signal model of non-maneuvering target.

[0078] Step 2: construct a parameterized sparse imaging model through maximum a posteriori (MAP) estimation, which is expressed as a cost function of joint target parameter estimation and sparse imaging;

[0079] The sparse representation of SA-ISAR image is:

[0080]

[0081]

[0082] Where a is the ISAR image to be restored, s is the sparse aperture echo data (SA data), is the total number of pulses of sparse aperture echo data, L i is the number of pulses of the ith subaperture, a iis the i-th pixel data of the ISAR image to be restored, η is the inverse of the variance of each component of a, is the variance of n, p(s|a) represents the likelihood function distribution model of s with respect to a, and p(a|η) represents the sparse prior probability distribution model of a. Using the maximum a posteriori estimation, the final formula for SA-ISAR image generation is:

[0083]

[0084] in, is the estimated value of the ISAR image a to be restored, is the estimated value of the phase error E, is the estimate of the scaled non-uniform Fourier transform Θ in the azimuthal dimension, is the coefficient that characterizes the sparse constraint. In equation (7), sparse aperture imaging and rotation parameter estimation are performed simultaneously by considering the non-cooperative motion of the target.

[0085] Step 3: Find the target rotation parameters by using the minimum entropy method, and perform parameter iteration by using the coordinate descent method to achieve a rough estimate of the maneuvering target rotation parameters;

[0086] Find the rotation parameters using the minimum entropy method:

[0087]

[0088] in

[0089]

[0090] Where a(n,k) is the (n,k)th element of the ISAR image of a, where α = 2f c ω 2 / c, Δy and Δk correspond to the rotation center offset caused by translation compensation, These unknown parameters are represented by is the estimate of α, is the estimate of β, is the estimate of Δy, is an estimate of Δk, using express estimates;

[0091] To compensate for the imaging error, in each iteration of the coordinate descent method, The other parameters in are only estimated In this way, we estimate one parameter in turn. For each parameter in, each such process is defined as an iteration; define is the estimate of the first q-1 parameters of the i-th iteration, The qth parameter estimate of

[0092]

[0093]

[0094] in, and yes The estimated values ​​at the (i-1)th and i-th iterations, respectively, and They are about The gradient vector and Hessian matrix of (i,q) is the search step size of the quasi-Newton method based on the Armijo criterion estimation; Using equations (10) and (11), the parameters are updated at each iteration. To achieve convergence, finally get a final estimated parameter.

[0095] Step 4: By combining sparse ISAR imaging and target rotation parameter estimation, the algorithm is decomposed into sparse imaging and dictionary learning, and the solutions are solved iteratively to achieve high-quality imaging with MTRC correction and accurate parameter estimation.

[0096] Combined with sparse ISAR imaging, it is decomposed into sparse imaging and dictionary learning, which are expressed as:

[0097]

[0098] and

[0099]

[0100] In formula (12), To solve the non-differentiability of ||a||1, δ is a positive constant close to infinitesimal. In the present invention, |a i | 2 1 / 1000 of the average value;

[0101] First, use the quasi-Newton method to solve equation (12):

[0102]

[0103] in and are the estimated values ​​of a at the i-th and (i+1)-th iterations, respectively, and is an approximate Hessian matrix; and are the estimated values ​​of E and Θ at the i-th iteration;

[0104] Considering The dimension of is large, and its inverse usually requires a large amount of calculation. Convert equation (14) to solve the following problem

[0105]

[0106] Use the conjugate gradient method to solve equation (15);

[0107] Then, equation (13) is transformed into The estimate of

[0108]

[0109] In fact, equation (16) is solved by gradient descent method or linear search method;

[0110] At the i-th iteration, the dynamic search is set at and Nearby, including and is ω and The estimated value in the (i-1)th iteration; after one iteration of parameter estimation, use and The estimate of ω is used to update the dictionary of E and Θ, and then the next sparse signal imaging iteration is performed using equation (15); thus, when the iterations of equations (15) and (16) converge, high-quality SA imaging with MTRC correction and accurate rotation parameter estimation can be achieved. The final iterative value of ω is used to calibrate the restored ISAR image in the azimuth dimension.

[0111] In order to illustrate the effectiveness of the present invention for sparse aperture ISAR high-resolution imaging and calibration of maneuvering targets, the following experiment on simulation data is further explained:

[0112] Example 1: Simulation data verification

[0113] 1. Experimental Setup

[0114] The Yak42 aircraft target model with a size of 36.47m×33.68m is used. The receiver adopts the de-chirp mode, and the ISAR system parameters are shown in Table 1. To indicate the maneuverability, the angular velocity of the target is set to 0.03rad / s, and the angular acceleration is set to 0.003rad / s 2 Meanwhile, the pulse number of FA was set to 400, and the total rotation angle of the target was 7.6°.

[0115] Radar system parameter settings:

[0116] Table 1 Simulated ISAR system parameters

[0117]

[0118] In de-chirp mode, the length of a sampling range unit is Where γ is the modulation frequency of the transmitted LFM signal, and it can be calculated that Δy = 0.293m. The imaging performance needs to be analyzed from two aspects: rotation parameter estimation and SA imaging. In order to verify the robustness of the algorithm, the number of FA samples is defined as all, half, and one quarter of the number of FA samples. In the simulation, a Gaussian noise signal is added to the echo data, and the signal-to-noise ratio of the echo signal is set to SNR = 20dB, 10dB, and 0dB.

[0119] 2. Experimental content

[0120] Based on the MATLAB software platform, 400, 200 and 100 pulse signals were used to perform range-Doppler two-dimensional imaging under SNR=20dB, 10dB and 0dB respectively, and the method of the present invention was used to estimate the rotation speed of the maneuvering target. Table 2 shows the target speed estimation results under SNR=20dB, 10dB and 0dB.

[0121] 3. Experimental results analysis

[0122] pass Figure 2 It can be seen that SA1 and SA2 are sampled data under FA, and are respectively one-half and one-quarter of the number of FA samples. Figure 3 It can be seen that the ISAR image effect is better in the case of FA and SA1, and worse in the case of SA2. The higher the SNR, the closer the estimated value of the parameter is to the actual value. At the same time, the accuracy of parameter estimation for FA, SA1, and SA2 decreases in turn. Figure 3 and Figure 4 From the comparison of the ISAR image imaging effects in the two figures and the comparison of the parameter estimation accuracy in Tables 2 and 3, it can be seen that the algorithm proposed in the present invention is superior to the traditional algorithm, and its advantages are mainly reflected in the sparse signal recovery and accurate parameter estimation.

[0123] Table 2 Rotation parameter estimation using the algorithm proposed by the present invention

[0124]

[0125] Table 3 Rotation parameter estimation using traditional algorithm

[0126]

[0127] Example 2: Measured data verification

[0128] In order to illustrate the effectiveness of the present invention in using measured data to estimate rotation parameters and reconstruct ISAR images, the following experiment on measured data is further described:

[0129] 1. Experimental Setup

[0130] The setting is the same as the previous simulation echo experiment.

[0131] 2. Experimental content

[0132] The present invention is first applied to FA data and then to SA data. In the case of FA, the ISAR image generated by the traditional RD algorithm is first used, and then the ISAR image is reconstructed from the FA data using the algorithm proposed by the present invention. The algorithm proposed by the present invention is used to image SA1 and SA2 respectively, and the range envelope after the ISAR image alignment is generated at the same time to observe the effect of the algorithm of the present invention in correcting MTRC.

[0133] 3. Experimental results analysis

[0134] pass Figure 5 In (a) and (b), it can be seen that the maneuvering target still has obvious distance MTRC after translation compensation, so the traditional RD algorithm imaging Figure 5 There is an obvious two-dimensional MTRC as shown in (c). Figure 5 In (d), it can be seen that the two-dimensional MTRC is well corrected for the ISAR image reconstructed using the algorithm of the present invention. The imaging and calibration results of the algorithm proposed in the present invention for SA1 and SA2 are shown in Figure 2. Figure 6 As shown in (a) and (b), it can be seen that by correcting the 2D MTRC, the SA echo data can obtain a well-focused image. In the SA case, the size of the reconstructed aircraft is roughly the same as that of the FA. Figure 6 The range images of the ISAR images generated in (a) and (b) are as follows: Figure 6 For clarity, the Figure 5 (b) The same region, Figure 6 (c) and (d) show that by successfully correcting the distance dimension MTRC, the distance envelope is completely aligned on a straight line. The above results confirm the effectiveness of the present invention.

[0135] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications should also be regarded as within the protection scope of the present invention.

Claims

1. A sparse aperture ISAR maneuvering target imaging and calibration method, characterized by: The steps include: Step 1: Establish a maneuvering target ISAR sparse aperture signal model, and construct a variable-scale non-uniform Fourier two-dimensional dictionary containing MTRC to characterize the target maneuverability; Step 2: construct a parameterized sparse imaging model through maximum a posteriori estimation, which is expressed as a cost function of joint target parameter estimation and sparse imaging; Step 3: Find the target rotation parameters by using the minimum entropy method, and perform parameter iteration by using the coordinate descent method to achieve a rough estimate of the maneuvering target rotation parameters; Step 4: By combining sparse ISAR imaging and target rotation parameter estimation, the algorithm is decomposed into sparse imaging and dictionary learning, and the solutions are iteratively solved in sequence to achieve high-quality imaging with MTRC correction and accurate parameter estimation. In step 1, the causes of two-dimensional cross-distance migration are first analyzed: Assume that the radar system transmits linear frequency modulation signal pulses and uses matched filtering technology to perform distance compression. The echo data compressed in the distance dimension is expressed as in, R p =R0+y p , is the fast time in the distance dimension, t m is the slow time in the azimuth dimension, x p and p They represent the horizontal and vertical coordinates of the scattering point p in the turntable model, R p is the distance between the origin of the coordinate system and the radar, ω is the angular velocity of the maneuvering target in the turntable model, j is a unit imaginary number, σ p is the scattering coefficient of the scattering point p, B and f c are the bandwidth and carrier frequency of the transmitted signal, respectively, and c represents the transmission speed of the electromagnetic wave; In equation (1), only the linear range offset in the envelope is retained as the range-dimensional MTRC, while the quadratic term is too small to generate the range-dimensional MTRC. The linear frequency modulation and quadratic phase in equation (1) correspond to the azimuth-dimensional MTRC, which can be decomposed into components related to the range and components related to the azimuth. Then equation (1) is transformed to the range-frequency domain as follows: Among them, f r It is the frequency domain representation of the distance-fast time of the echo signal; From formula (2), we can see that the mobility of the target causes non-uniform spatial sampling, that is, it brings about chirp modulation related to the azimuth, which is very different from the echo under non-maneuvering targets. In essence, formula (2) is r ,t m ) domain exhibits characteristics unique to MTRC; In the step 1, the method of constructing a variable-scale non-uniform Fourier two-dimensional dictionary including MTRC is: modeling the sparse aperture discrete signal: Where a and s represent the ISAR image and SA data to be restored, respectively. A is the matrix form of the ISAR image, with the rows being the azimuth dimension and the columns being the distance dimension. n is the noise. vec() is a matrix vector operation, which means that the SA data of I sub-apertures are connected one by one according to the aperture number to form a long vector. represents the total number of pulses of sparse aperture echo data, where L i is the number of pulses of the i-th sub-aperture, N represents the echo data sampling length in the range dimension, k represents the k-th pulse in the echo pulse sequence, n represents the n-th sampling data in an echo pulse, and j is a unit imaginary number; is the quadratic phase error related to distance in equation (1) or (2), F r represents the Fourier transform of the distance dimension, F r H Indicates F r The conjugate transpose of; In formula (4), the diagonal elements of E are all non-zero, and the others are all zero; The element at n∈[1:N],m∈[1:M] where m and m' are the mth and m'th pulses of FA and SA under the same pulse, respectively; in addition, Θ represents the scaled non-uniform Fourier transform of the azimuthal dimension, where and are scaling factor and uneven step size, respectively. Here F s is the sampling frequency in the distance dimension; when d(m) = 1, equation (4) can be simplified to the signal model of non-maneuvering targets; In step 2, the SA-ISAR image is sparsely represented as: Among them, a is the ISAR image to be restored, s is the sparse aperture echo data, is the total number of pulses of sparse aperture echo data, L i is the number of pulses of the ith subaperture, a i is the i-th pixel data of the ISAR image to be restored, η is the inverse of the variance of each component of a, is the variance of n, p(s|a) represents the likelihood function distribution model of s with respect to a, and p(a|η) represents the sparse prior probability distribution model of a. Using the maximum a posteriori estimation, the final formula for SA-ISAR image generation is: in, is the estimated value of the ISAR image a to be restored, is the estimated value of the phase error E, is the estimate of the scaled non-uniform Fourier transform Θ in the azimuthal dimension, is the coefficient that characterizes the sparse constraint. In equation (7), sparse aperture imaging and rotation parameter estimation are performed simultaneously by considering the non-cooperative motion of the target; In step 3, the rotation parameters are found by the minimum entropy method: in Where a(n,k) is the (n,k)th element of the ISAR image of a, where α = 2f c ω 2 / c, Δy and Δk correspond to the rotation center offset caused by translation compensation, These unknown parameters are represented by is the estimate of α, is the estimate of β, is the estimate of Δy, is an estimate of Δk, using express estimates; To compensate for the imaging error, in each iteration of the coordinate descent method, The other parameters in are only estimated In this way, we estimate one parameter in turn. For each parameter in, each such process is defined as an iteration; define is the estimate of the first q-1 parameters of the i-th iteration, The qth parameter estimate of in, and yes The estimated values ​​at the (i-1)th and i-th iterations, respectively, and They are about The gradient vector and Hessian matrix of (i,q) is the search step size of the quasi-Newton method based on the Armijo criterion estimation; using equations (10) and (11), the parameters are updated at each iteration To achieve convergence, and finally get a final estimated parameter; In step 4, combined with sparse ISAR imaging, it is decomposed into sparse imaging and dictionary learning, which are respectively expressed as: and In formula (12), To solve the non-differentiability of ||a||1, δ is a positive constant close to infinitesimal. In the present invention, |a i | 2 1 / 1000 of the average value; First, use the quasi-Newton method to solve equation (12): in and are the estimated values ​​of a at the i-th and (i+1)-th iterations, respectively, and is an approximate Hessian matrix; and are the estimated values ​​of E and Θ at the i-th iteration; Considering The dimension of is large, and its inverse usually requires a large amount of calculation. Convert equation (14) to solve the following problem Use the conjugate gradient method to solve equation (15); Then, equation (13) is converted into The estimate of In fact, equation (16) is solved by gradient descent method or linear search method; At the i-th iteration, the dynamic search is set at and Nearby, including and is ω and The estimated value in the (i-1)th iteration; after one iteration of parameter estimation, use and The estimate of ω is used to update the dictionary of E and Θ, and then the next sparse signal imaging iteration is performed using equation (15). In this way, when the iterations of equations (15) and (16) converge, high-quality SA imaging with MTRC correction and accurate rotation parameter estimation can be achieved. The final iterative value of ω is used to calibrate the restored ISAR image in the azimuth dimension.

Images