A multi-frame super-resolution imaging method for airborne forward-looking scanning radar

Through the multi-frame super-resolution imaging method, combined with sparse constraints and data fidelity terms, the sub-pixel-level displacement is accurately estimated, which solves the problem that the single-frame super-resolution method is susceptible to noise, and significantly improves the imaging quality of the radar forward-view area, overcoming the The blind spots and noise fluctuations of traditional radar imaging in the foreview area are solved.

CN118795472BActive Publication Date: 2024-12-03NANJING MASHANG ANJIA TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411268677.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-12-03
Estimated Expiration
2044-09-11

AI Technical Summary

Technical Problem

The existing radar technology has formed a 'blind spot' in the imaging area due to the lack of effective Doppler bandwidth and 'left and left-right blur' phenomenon when imaging the forward-view area, and the single-frame super-resolution method is susceptible to noise and target scattering coefficients. Influence, poor imaging effect.

Method used

Using multi-frame super-resolution imaging method, the objective function of multi-frame super-resolution imaging is combined with sparse constraint terms by distance-dimensional pulse compression, essay correction, multi-frame sub-pixel displacement estimation, sub-pixel displacement frame echo convolution model construction, and multi-frame joint super-resolution imaging. and data fidelity terms, optimized solutions using iterative threshold shrinkage algorithm, accurately estimate subpixel-level displacement and improve imaging quality.

Benefits of technology

Break through the limitations of single-frame echo data information, accurately estimate the subpixel-level displacement between multi-frame echo data, and significantly improve the imaging quality of airborne forward-view scanning radar through redundant complementary information reorganization of time and space dimensions. , improves image resolution and noise suppression capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118795472B_ABST
    Figure CN118795472B_ABST
Patent Text Reader

Abstract

The present invention discloses a multi-frame super-resolution imaging method for airborne forward-looking scanning radar. The method utilizes the redundant complementary information extended in the spatial dimension of radar multi-frame scanning and the suppression characteristics of the target scattering coefficient matrix and the receiver noise fluctuation in the time dimension, and constructs the imaging objective function by combining sparse prior constraints, and uses an iterative threshold contraction method to complete the optimization solution of the objective function, including a distance-dimensional pulse compression step, a motion correction step, a multi-frame sub-pixel displacement estimation step, a sub-pixel displacement frame echo model construction step, a multi-frame joint super-resolution imaging objective function construction step, and a super-resolution imaging result output step. The present invention can break through the limitations of the information carried by single-frame echo data in the existing method, accurately estimate the sub-pixel displacement movement between multiple frames of echo data, complete the reorganization of multiple frames of redundant complementary information from the two dimensions of time and space, and effectively improve the imaging quality of the airborne forward-looking scanning radar.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to radar technology, and in particular relates to multi-frame super-resolution imaging technology of airborne forward-looking scanning radar. Background Art

[0002] Radar imaging systems have the advantages of all-day, all-weather, long detection distance and high imaging resolution, and play an increasingly important role in military and civilian fields such as terrain mapping and battlefield surveillance. Traditional synthetic aperture radar SAR technology and Doppler sharpening DBS technology cannot perform high-resolution imaging of the forward-looking area directly in front of the aircraft due to the lack of effective Doppler bandwidth and the "left-right blur" phenomenon, forming a "blind spot" in the imaging area.

[0003] Radar forward-looking super-resolution imaging refers to building an echo convolution model based on the convolution of the azimuth scanning echo, which is the antenna pattern and the target scattering coefficient matrix, on the premise of acquiring the forward-looking echo of the real aperture scanning radar, and using the deconvolution method to estimate the target scattering coefficient matrix, thereby improving the azimuth resolution of the forward-looking scanning radar.

[0004] In the past decade, domestic and foreign scholars have been increasingly interested in the research of radar forward-looking super-resolution imaging. A variety of forward-looking super-resolution algorithms have been proposed, and the degree of image quality enhancement has been continuously improved. The paper "Sparse super-resolution imaging for airborne single channel forward-looking radar in expanded beam space vialp regularisation" (Chen HM, Li M., Wang Z, et al. Electronics Letters, 2015, 51(11): 863-865) proposed a sparse super-resolution imaging method, which uses the sparse characteristics of key targets in the scene and models the targets through the L1 norm to achieve effective resolution of cluster targets. However, due to the over-emphasis on scene sparsity, this method will cause certain damage to the background contour while resolving targets; the paper "A TV forward-looking super-resolution imaging method based on TSVD strategy for scanning radar" (Zhang Y., Tuo X., Huang Y., Yang J., IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(7): 4517-4528) proposed a sparse TV regularization method, which introduced a total variation constraint to solve the problem of scene contour recovery. The target scene was modeled and solved by the regularization method, which can effectively preserve the contour features of the scene.

[0005] However, whether it is the super-resolution method based on sparse prior or total variation constraint, it is based on the prior information of a single frame image to guide the super-resolution process. The scene is observed only once, the information that can be mined is limited, and it is affected by the receiver noise and the fluctuation of the target scattering coefficient. It is difficult to achieve a fundamental breakthrough in the super-resolution effect.

[0006] The idea of ​​sub-pixel multi-frame super-resolution is to observe the scene multiple times and from different perspectives through the movement of the airborne platform, which not only alleviates the influence of target scattering coefficient and noise fluctuation from the time dimension, but also introduces sub-pixel redundant information in the spatial dimension, making it possible to improve super-resolution imaging results. Therefore, how to accurately estimate the sub-pixel displacement between multi-frame image data, and how to construct the sub-pixel displacement frame echo convolution model and super-resolution objective function are the core issues that need to be solved. Summary of the invention

[0007] The technical problem to be solved by the present invention is to propose a multi-frame super-resolution imaging method for an airborne forward-looking scanning radar that can effectively improve the imaging quality, in view of the problem that the super-resolution of the prior art is subject to the limitations of single-frame information and is easily affected by noise and target echo fluctuations, resulting in poor imaging effect.

[0008] The technical solution adopted by the present invention to solve the above technical problems is a multi-frame super-resolution imaging method of an airborne forward-looking scanning radar, comprising the steps of:

[0009] Range dimension pulse compression step: performing pulse compression on the real beam scanning echo along the range dimension to obtain a pulse compressed echo;

[0010] Motion correction step: performing range motion correction on the echo after pulse compression along the range dimension to obtain a motion-corrected echo;

[0011] Multi-frame sub-pixel displacement estimation steps: Fourier transform the two frames of echo data after motion correction respectively, multiply the conjugate of the two Fourier transform results and divide them by the absolute value of the multiplication result to obtain a normalized cross-correlation spectrum, then perform inverse Fourier transform on the normalized cross-correlation spectrum to obtain a two-dimensional Dirac kernel function, and solve the sub-pixel distance displacement and sub-pixel azimuth displacement between the two frames of echo data according to the peak point of the two-dimensional Dirac kernel function and the sampling values ​​around it;

[0012] The steps of constructing the sub-pixel displacement frame echo convolution model are as follows: the convolution matrix is ​​obtained from the antenna pattern; the displacement matrix is ​​constructed using the sub-pixel range displacement and the sub-pixel azimuth displacement; the word-typical vector of each frame echo data after motion correction is expressed as the sum of the product of the downsampling matrix, the convolution matrix, the displacement matrix and the target scattering coefficient matrix and the word-typical vector of the radar receiver noise;

[0013] The steps of constructing the objective function of multi-frame joint super-resolution imaging are as follows: the objective function of multi-frame joint super-resolution imaging is constructed based on the sub-pixel displacement frame echo convolution model to solve the minimum value of the sum of the data fidelity term and the sparse constraint term; wherein the data fidelity term is the cumulative value of the L2 norm square of the expression of the word typical vector of the radar receiver noise in several sub-pixel displacement frame echo convolution models; the sparse constraint term is the product of the L1 norm of the scattering coefficient matrix of the target and the regularization parameter controlling the sparse term;

[0014] Super-resolution imaging result output steps: The final target scattering coefficient matrix is ​​obtained by iteratively solving the gradient of the multi-frame joint super-resolution imaging objective function, and the radar forward-looking super-resolution imaging result is output according to the final target scattering coefficient matrix.

[0015] The present invention utilizes the redundant complementary information extended in the spatial dimension by radar multi-frame scanning and the suppression characteristics of the target scattering coefficient matrix and receiver noise fluctuation in the temporal dimension, combines sparse prior constraints to construct the imaging objective function, and adopts an iterative threshold shrinkage algorithm for optimization solution.

[0016] The beneficial effect of the present invention is that it can break through the limitation of the information carried by single-frame echo data of the existing method, accurately estimate the sub-pixel displacement movement between multiple frames of echo data, complete the reorganization of multiple frames of redundant complementary information from the two dimensions of time and space, and effectively improve the imaging quality of airborne forward-looking scanning radar. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the method provided by the present invention.

[0018] Figure 2 This is the original scene required for echo generation in the specific implementation manner of the present invention.

[0019] Figure 3 These are three consecutive frames of original echo signals in a specific implementation manner of the present invention; (a) is the first frame; (b) is the second frame; and (c) is the third frame.

[0020] Figure 4 It is the result of three consecutive frames after pulse compression in a specific implementation mode of the present invention; (a) is the first frame; (b) is the second frame; (c) is the third frame.

[0021] Figure 5 It is the result of three consecutive frames after migration correction in a specific implementation mode of the present invention; (a) is the first frame; (b) is the second frame; (c) is the third frame.

[0022] Figure 6 The super-resolution results after being processed by the method of the present invention and the existing method; (a) is the super-resolution result of the sparse regularization method; (b) is the super-resolution result of the method of the present invention. DETAILED DESCRIPTION

[0023] The present invention utilizes the redundant complementary information extended in the spatial dimension of radar multi-frame scanning and the suppression characteristics of the target scattering coefficient matrix and receiver noise fluctuations in the temporal dimension, combines sparse prior constraints to construct the imaging objective function, and then uses the iterative threshold shrinkage method to optimize the objective function to finally obtain high-quality radar images.

[0024] The method of the present invention, such as Figure 1 As shown, including:

[0025] Step 1: Perform range-dimensional pulse compression on the real beam scanning echo:

[0026] It is known that the pulse width of the radar transmission signal is , the signal bandwidth is , emission signal wavelength , radar transmit signal center frequency and FM slope ;

[0027] The radar transmits a linear frequency modulation signal, the antenna beam scans the imaging area, and the received real beam scanning echo for:

[0028] (1)

[0029] in, Represents the time variable of radar sampling in the distance dimension, with a total of M sampling points. is the time variable sampled by the radar in the azimuth dimension, and the total number of sampling points is N. and They are and The integral variable, represents the target echo scattering coefficient, represents the antenna pattern of the scanning antenna, is the speed of light, is the target slant distance, is an exponential function.

[0030] The above real beam scanning echo is subjected to range-dimensional pulse compression, that is, the real beam scanning echo is first subjected to Fourier transform along the range dimension; then a range-dimensional pulse compression reference function is constructed, and then the real beam scanning echo after Fourier transform is multiplied by the conjugate of the range-dimensional pulse compression reference function, and the multiplication result is inversely transformed into the time domain to obtain the echo after pulse compression , and finally the echo after pulse compression can be obtained expression:

[0031] (2)

[0032] in, is the Sinque function.

[0033] Step 2: Perform motion correction on the echo after pulse compression:

[0034] In a platform with fast scanning time but slow movement speed, ,in, is the initial distance of the target from the radar, For the under-the-radar perspective, is the projection of the platform motion speed in the distance dimension. According to this approximate formula, the echo after pulse compression can be Written as:

[0035] (3)

[0036] Perform linear migration correction on it to obtain the echo after migration correction for:

[0037] (4)

[0038] in, is the target scattering coefficient matrix after phase weighting.

[0039] Step 3: Multi-frame sub-pixel displacement estimation:

[0040] That is, we get a frame of data without super-resolution processing, and after image display, we can get an echo image, and its actual physical resolution in azimuth is determined by the azimuth beam width of the radar. When the airborne platform moves, there is a displacement between the next frame of data and the previous frame of data. This displacement is mostly sub-pixel level, that is, the movement between the two frames is not exactly at the position of the whole pixel. The next frame of data collects the information between the sampling points of the previous frame, thereby obtaining additional spatial redundant information, which makes it possible to further improve the azimuth resolution. However, in order to make reasonable use of this additional spatial redundant information, it is necessary to accurately estimate the sub-pixel displacement between multiple frames of echoes, which is what needs to be implemented in this step.

[0041] The sub-pixel displacement between multiple frames of observation data can be equivalent to: the whole pixel displacement between different original high-resolution image frames is formed by downsampling, that is, the images that appear to have sub-pixel (i.e. less than one pixel) displacement between multiple frames are actually displaced by integer pixel values ​​at the beginning, but due to the subsequent downsampling process, these displacements are reduced to sub-pixel level. In this patent radar, this downsampling corresponds to the fast time sampling in distance and the slow time sampling in azimuth, which are finally obtained after step one and step two. Next we will start with the analysis of the normalized cross-power spectra of the two original high-resolution data images that have not been downsampled by the radar.

[0042] Assume two frames of original resolution image data and There is a full pixel displacement between .in , are the distance and azimuth variables of the original resolution image data, and They are respectively the integer pixel level displacement in the distance dimension and the integer pixel level displacement in the azimuth dimension between two frames of original data images.

[0043] The Fourier transforms of the two image data are and ,in and Respectively represent , The frequency domain variable. Then the Fourier transform after F times and G times downsampling along the corresponding radar range dimension and azimuth is shown as follows:

[0044] (5)

[0045] (6)

[0046] Get the cross-correlation power spectrum of the two downsampled data images As shown below:

[0047] (7)

[0048] in, As shown below:

[0049] (8)

[0050] According to the above results, the normalized cross-power spectrum of two downsampled images is a multiphase decomposition of a filtered unit pulse, that is, the cross-correlation power spectrum is composed of the sum of all multiphase components of the filtered unit pulse. Therefore, the cross-correlation power spectrum can be regarded as a downsampled version of a unit pulse processed by a rectangular frequency response filter, that is, a downsampled version of a two-dimensional Dirac kernel is formed, which has been shown to be very close to a two-dimensional Sinker function.

[0051] Using formula (7), we can estimate the sum of the multiphase components in the normalized cross-correlation power spectrum, which is all the information needed to estimate the sub-pixel displacement. Performing inverse Fourier transform yields the following formula:

[0052] (9)

[0053] Where W and H are the two-dimensional lengths of the data at the original image resolution, and the above formula is approximated to obtain:

[0054] (10)

[0055] is the center point of the two-dimensional Dirac kernel function. The center point can be represented by the peak point (usually The specific derivation process is as follows:

[0056] Will Substituting into the formula we get :

[0057] (11)

[0058] Will Substituting into the formula we get :

[0059] (12)

[0060] The above derivation yields Sub-pixel distance displacement after F-fold downsampling for:

[0061] (13)

[0062] The same can be said Sub-pixel azimuth displacement after G-fold downsampling for:

[0063] (14)

[0064] From equations (13) and (14), we can see that solving the sub-pixel distance displacement and sub-pixel azimuth displacement It is not necessary to know the downsampling ratio between the known echo image and the unknown original resolution image in advance. and .

[0065] Back to the actual operation, what we actually get is the image data after downsampling of two frames of original resolution images, that is, the image data after range migration correction. and Perform Fourier transform on the two frames of data to obtain and , , They are , In the frequency domain, solving the normalized cross-correlation spectrum gives a direct expression of formula (7):

[0066] (15)

[0067] in, represents conjugation, Perform inverse Fourier transform to get , substituting into formula (13) and (14) to obtain the sub-pixel distance displacement and sub-pixel azimuth displacement .

[0068] Step 4: Construction of sub-pixel displacement frame echo convolution model

[0069] The present invention mainly performs super-resolution in the azimuth direction. The direction signal Extract it:

[0070] (16)

[0071] is the convolution of the scattering coefficient of the target and the antenna pattern of the scanning antenna:

[0072] (17)

[0073] in is the convolution operation. Considering the influence of noise, the azimuth signal expression is rewritten as

[0074] (18)

[0075] in is the radar receiver noise.

[0076] Discretize formula (17) and express it in vector form to obtain the azimuth convolution signal vector As a reference frame echo convolution model:

[0077] (19)

[0078] in, , and for dimensional column vector, is the discretized radar receiver noise, is the total sampling points of the two-dimensional echo matrix in the azimuth dimension, The convolution matrix determined for the antenna pattern is:

[0079] (20)

[0080] in, The vector sampling points, is the total number of sampling points of the antenna pattern.

[0081] The process of generating sub-pixel displacement frame echo is to scan and observe again on the basis of the reference frame through platform movement. The displacement has sub-pixel characteristics. The echo can also be regarded as an original high-resolution image formed by integer-level displacement downsampling. Therefore, the sub-pixel displacement frame echo convolution model is as follows:

[0082] (twenty one)

[0083] in For the The typical vector of the frame radar echo after range migration correction is composed of data, and the vector length is , is the total number of azimuth slow time sampling points, is the total number of sampling points in the distance dimension fast time. For the The typical vector of the word composed of the frame radar receiver noise, the vector length is .

[0084] Sub-pixel distance displacement and sub-pixel azimuth displacement The conventional displacement matrix constructed represents the The displacement operation between the frame radar echo and the reference frame, the matrix size is , and They are the downsampling multiples of the distance dimension and the azimuth, respectively, which depend on the compromise between the accuracy of sub-pixel displacement estimation and the increase in computational complexity caused by super-resolution reconstruction. In layman's terms, if it is too high, the computing power consumption will be too great, and if it is too low, the super-resolution effect will be limited. Usually, it does not exceed 1 order of magnitude. The specific formula is as follows:

[0085] (twenty two)

[0086] in The size is The zero matrix of indivual matrix, It is before A diagonal matrix with zero rows is as follows:

[0087] (twenty three)

[0088] Depend on conduct After interpolation Constructed, the matrix size is , as shown below:

[0089] (twenty four)

[0090] is the conventional downsampling matrix or sampling matrix, and Determine that the matrix size is .

[0091] Formula (21) can be used to represent both sub-pixel displacement frames and reference frames, that is, , the displacement is (0,0) as the base frame.

[0092] Step 4: Construction of multi-frame joint super-resolution imaging objective function:

[0093] In order to solve the pathological problem of deconvolution super-resolution, the embodiment adopts regularization technology, and aims at the limitation of single-frame super-resolution information of existing algorithms, utilizes the spatial redundant information between multiple frames and the temporal smoothing characteristics of the target scattering coefficient matrix and noise, and jointly constructs the multi-frame joint super-resolution imaging objective function with sparse constraints as follows:

[0094] (25)

[0095] in, is the L2 norm, is the L1 norm, is the multi-frame joint data fidelity term, is a sparse constraint term, is the regularization parameter that controls the sparsity constraint. The objective function in the regularization method combines the data fidelity term and the regularization term to incorporate prior knowledge about the desired solution into the optimization problem.

[0096] Step 5: Output of super-resolution imaging results:

[0097] To optimize the above objective function and find the extreme solution that minimizes the objective function, it is obviously quite difficult to solve it directly, so we can use an iterative method to gradually approach the true solution. This paper adopts the iterative threshold shrinkage IST method. Because the L1 norm is not differentiable, to differentiate the objective function (25), we first need to introduce a very small constant , used to solve the non-differentiability of the L1 norm and rewrite the objective function for:

[0098] (26)

[0099] Then solve the gradient of the modified objective function:

[0100] (27)

[0101] in, for No. values, To construct a diagonal matrix function.

[0102] Then the basic iteration formula of the IST method is as follows:

[0103] (28)

[0104] in, and is the result of two consecutive iterations, is the iteration step length, which should satisfy the following conditions to ensure the convergence of the algorithm: , is the iterative threshold shrinkage operator, the intermediate quantity ;

[0105] (29)

[0106] in, is the upper threshold limit, which is usually taken as an estimated value smaller than the standard deviation of the noise. The regularization parameter can be determined by the L curve method, that is, when the regularization parameter is located at the best position of the data fidelity term and the regularization term, it is taken as the optimal value.

[0107] After the iterative solution is completed, the target scattering coefficient matrix is ​​output After obtaining the target scattering coefficient matrix distribution, the radar forward-looking super-resolution imaging result can be output.

[0108] Example

[0109] The present invention is mainly verified by simulation experiment, and all steps and conclusions are verified to be correct on Matlab 2020. The implementation process of the embodiment is further described in detail below.

[0110] The simulation parameters of the radar system include: carrier frequency 10GHz, platform speed 200m / s, transmission signal bandwidth 5MHz, transmission signal pulse width 2us, sampling rate 10MHz, pulse repetition frequency 500Hz, first zero point azimuth main lobe beam width 1°, antenna scanning speed , Antenna scanning range , the average signal-to-noise ratio of the scene after superimposed noise (after pulse compression) is 20dB and the initial scene distance range is 5000m-8000m.

[0111] Step 1: Radar original echo image simulation. Use the above radar system simulation parameters to select Figure 2 The original scene () shown in the figure is subjected to real beam scanning echo simulation, and 3 frames of original real beam scanning echo simulation data with sub-pixel displacement are generated, as shown in FIG. Figure 3 As shown in (a), (b) and (c).

[0112] Step 2: Perform Fourier transform on the real beam scanning echo data along the distance dimension. Construct the distance dimension pulse compression reference function ,in Then the real beam scanning echo after Fourier transformation is compared with Multiply, and then transform the multiplication result back to the time domain to complete the distance dimension pulse compression of the simulated echo, and the echoes after three consecutive frames of pulse compression are obtained. , , ,like Figure 4 As shown in (a), (b) and (c). represents the distance dimension frequency domain variable, Indicates conjugation.

[0113] Step 3: Perform range migration correction on the data in step 2. First, perform Fourier transform on the echo after pulse compression along the range dimension, and record it as Then construct the distance migration correction reference function ,in Scan time . Will echo and distance migration correction reference function Multiply, and then inversely transform the multiplication result to the time domain to complete the range motion correction of the simulated echo. The three consecutive frames of data after the motion correction are recorded as , , ,like Figure 5 As shown in (a), (b) and (c).

[0114] Step 4: Perform sub-pixel displacement estimation on the data in step 3. The first frame of data is defined as the reference frame. First, Fourier transform the first frame and the second frame of data to obtain and , and then the normalized cross-correlation spectrum is obtained according to formula (14): , and then perform inverse Fourier transform to obtain the two-dimensional Dirac kernel function Finally, according to formulas (13) and (14), the sub-pixel displacement of the second frame relative to the first frame is (-5.623, 0.312). Repeat the above steps to obtain the sub-pixel displacement of the third frame relative to the first frame as (-10.843, -0.531). Considering the increase in calculation amount caused by displacement estimation accuracy and interpolation multiples, the super-resolution reconstruction multiple is 4, that is, the distance dimension and azimuth are downsampled by and Both are 4, so the sub-pixel displacements of the second and third frames relative to the reference frame are readjusted to (-5.5, 0.25) and (-10.75, -0.5) respectively according to the displacement estimation accuracy of 0.25 pixels.

[0115] Step 5: Construct multi-frame joint super-resolution imaging objective function.

[0116] First, three consecutive frames of data after range migration correction , , They are represented by typical word vectors, respectively. , , ; The sub-pixel displacement values ​​(0,0), (-5.5, 0.25), and (-10.75, -0.5) after precision adjustment are constructed into three-frame displacement matrices according to formulas (21) and (22) respectively. , , ;

[0117] Then obtain the antenna pattern convolution matrix, and calculate the main lobe beam width between the first zero points of the antenna pattern. , pulse repetition frequency and scanning speed The scanning antenna pattern is discretized by The number of discretized antenna pattern points is 10, which can be constructed according to formulas (20) and (24): .

[0118] Finally, according to the downsampling multiple and Constructing a regular downsampling matrix , and construct the multi-frame joint super-resolution imaging objective function according to formula (24).

[0119] Step 6: Obtain the objective function gradient according to expression (28). First, Assign to , , the regularization parameter The L curve method is used to determine. The optimal solution is obtained by iterative solution using formula (29): Take 1, It can be estimated based on the noise standard deviation. When the results of two iterations satisfy the following formula

[0120] (30)

[0121] The result converges, that is, the target solution is obtained, where the convergence threshold The empirical value is set to 0.015. If formula (30) is not satisfied, update Continue iterating until the iteration termination condition is met and the target solution is obtained. The final output imaging result is as follows: Figure 6 (b) shown.

[0122] As a comparison, the echo signal is super-resolved using the existing sparse regularization method, and the imaging results are as follows: Figure 6 (a). Comparison Figure 6As can be seen from Figures 6(a) and 6(b), the imaging quality of the method provided by the present invention is higher, which is specifically manifested in the following aspects: 1) the method of the present invention shows more detailed contours of ships in the water; 2) the method of the present invention has higher resolution for 5 buildings in the area of ​​7°~10° and 6400m~6600m; 3) the prior art method will have pathological characteristics caused by noise at the edge of the image, which is manifested as striped curling and undulation. The method of the present invention has smooth characteristics for noise, and this phenomenon is alleviated to a certain extent.

Claims

1. A multi-frame super-resolution imaging method for airborne forward-looking scanning radar, characterized in that: Includes steps: Range dimension pulse compression step: performing pulse compression on the real beam scanning echo along the range dimension to obtain a pulse compressed echo; Motion correction step: performing range motion correction on the echo after pulse compression along the range dimension to obtain a motion-corrected echo; Multi-frame sub-pixel displacement estimation steps: Fourier transform the two frames of echo data after motion correction respectively, multiply the conjugate of the two Fourier transform results and divide them by the absolute value of the multiplication result to obtain a normalized cross-correlation spectrum, then perform inverse Fourier transform on the normalized cross-correlation spectrum to obtain a two-dimensional Dirac kernel function, and solve the sub-pixel distance displacement and sub-pixel azimuth displacement between the two frames of echo data according to the peak point of the two-dimensional Dirac kernel function and the sampling values ​​around it; The steps of constructing the sub-pixel displacement frame echo convolution model are as follows: the convolution matrix is ​​obtained from the antenna pattern; the displacement matrix is ​​constructed using the sub-pixel range displacement and the sub-pixel azimuth displacement; the word-typical vector of each frame echo data after motion correction is expressed as the sum of the product of the downsampling matrix, the convolution matrix, the displacement matrix and the target scattering coefficient matrix and the word-typical vector of the radar receiver noise; Among them, the sub-pixel displacement frame echo convolution model is specifically: Among them, j is the frame number, x j is the typical vector of the j-th frame echo data after motion correction, the vector length is MN×1, N is the total number of azimuth slow time sampling points, M is the total number of distance dimension fast time sampling points, D is the downsampling matrix, To obtain the convolution matrix from the antenna pattern, F j is the displacement matrix of the j-th frame echo data, σ is the target scattering coefficient matrix, is the typical vector of radar receiver noise; F j It is a conventional displacement matrix constructed by sub-pixel range displacement Δτ and sub-pixel azimuth displacement Δt, representing the displacement operation between the j-th frame radar echo and the reference frame. The matrix size is L1L2MN×L1L2MN, where L1 and L2 are the downsampling multiples of the range dimension and azimuth, respectively. It depends on the compromise between the accuracy of sub-pixel displacement estimation and the increase in computational complexity caused by super-resolution reconstruction multiples. j The specific formula is as follows: Where 0′ is a zero matrix of size L1L2MN×ΔtL2M, there are L2N-L2Δt F′ matrices on the right diagonal, and F′ is a diagonal matrix with zeros in the first ΔτL1 rows, as follows: The convolution matrix H determined by the antenna pattern is interpolated by L2 times and is constructed as H′. The matrix size is L1L2MN×L1L2MN, as shown below: Among them, the convolution matrix H determined by the antenna pattern is: Among them, h i , i=1,2…l is the i-th sampling point of the vector, l is the total number of sampling points of the antenna pattern; The steps of constructing the objective function of multi-frame joint super-resolution imaging are as follows: the objective function of multi-frame joint super-resolution imaging is constructed based on the sub-pixel displacement frame echo convolution model to solve the minimum value of the sum of the data fidelity term and the sparse constraint term; wherein the data fidelity term is the cumulative value of the L2 norm square of the expression of the word typical vector of the radar receiver noise in several sub-pixel displacement frame echo convolution models; the sparse constraint term is the product of the L1 norm of the scattering coefficient matrix of the target and the regularization parameter controlling the sparse term; Super-resolution imaging result output steps: The final target scattering coefficient matrix is ​​obtained by iteratively solving the gradient of the multi-frame joint super-resolution imaging objective function, and the radar forward-looking super-resolution imaging result is output according to the final target scattering coefficient matrix.

2. The method according to claim 1, characterized in that: The specific steps of range-dimensional pulse compression are as follows: perform Fourier transform on the real beam scanning echo along the range dimension, then construct a range-dimensional pulse compression reference function, multiply the real beam scanning echo after Fourier transform by the conjugate of the range-dimensional pulse compression reference function, and inversely transform the multiplication result into the time domain to obtain the echo after pulse compression.

3. The method according to claim 1, characterized in that: The specific steps of motion correction are: perform Fourier transform on the echo after pulse compression along the distance dimension, then construct a distance motion correction reference function, multiply the echo after pulse compression after Fourier transform with the distance motion correction reference function, and then inversely transform the multiplication result to the time domain to complete the distance motion correction of the echo and obtain the echo after motion correction.

4. The method according to claim 1, characterized in that: The sub-pixel distance displacement Δτ and sub-pixel azimuth displacement Δt between two frames of echo data are specifically: Where C(0,0), C(0,1), and C(1,0) are the values ​​of the two-dimensional Dirac kernel function C(τ,t) at positions (0,0), (1,0), and (0,1), respectively. The two-dimensional Dirac kernel function C(τ,t) is the normalized mutual power spectrum function between the two frames. Obtained by inverse Fourier transform; τ is the variable sampled by the radar in the distance dimension, t is the variable sampled by the radar in the azimuth dimension, is the variable of τ in the frequency domain, is the variable of t in the frequency domain.

5. The method according to claim 1, characterized in that: The specific objective function of multi-frame joint super-resolution imaging is: Among them, ||·||2 is the L2 norm, ||·||1 is the L1 norm, is the data fidelity term, λ||σ||1 is the sparse constraint term, and λ is the regularization parameter that controls the sparse constraint term.

6. The method according to claim 5, characterized in that In the process of iteratively solving the gradient of the multi-frame joint super-resolution imaging objective function to obtain the final target scattering coefficient matrix, the iterative update method of the target scattering coefficient matrix σ each time is: Among them, σ k is the kth iteration result of the target scattering coefficient matrix σ, is the iterative threshold shrinkage operator, is the gradient of the multi-frame joint super-resolution imaging objective function, γ is the iteration step size, T stands for transpose; Intermediate δ is the upper threshold value.

7. The method according to claim 6, characterized in that: Gradient of objective function for multi-frame joint super-resolution imaging Specifically: Among them, diag(·) is the function of constructing a diagonal matrix, σ i is the i value in σ, and ε is a tiny constant introduced to solve the L1 norm non-differentiable problem.

8. The method according to claim 7, characterized in that: The regularization parameter λ that controls the sparsity constraint term is determined using the L-curve method.