A phase reconstruction based artifact suppression method

Abstract

The application discloses a kind of based on phase reconstruction's artifact suppression method, comprising: first, the synthetic aperture radar image containing artifact is transformed to distance frequency domain-azimuth Doppler domain by two-dimensional Fourier transform, then sub-image is constructed from distance frequency domain-azimuth Doppler domain, and dictionary is constructed based on sub-image;After low rank and sparse decomposition are carried out to dictionary, the low rank matrix solved is split and reorganized to construct sub visual low rank matrix, then initial sub phase hologram is constructed in distance frequency domain-azimuth Doppler domain, and sub image of phase in time domain is converted, the modulus of this sub image is calculated with the root mean square error of corresponding sub visual low rank matrix and this step is repeatedly executed until it is considered that the phase of sub image is correct after meeting the requirement;Finally, all sub images are spliced and inverse Fourier transform is carried out to carry out the reconstruction of characterization matrix, and the synthetic aperture radar image without artifact is obtained using the reconstructed characterization matrix.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing, and in particular to an artifact suppression method based on phase reconstruction. Background Art

[0002] In strip mode, a type of artifact appears in the range-Doppler domain of synthetic aperture radar images as oblique signals with large azimuth modulation rates distributed on both sides of the Doppler spectrum. Because this feature is similar to that of fast-moving targets, the Doppler sub-view decomposition method for moving target detection can also be used to detect and suppress this type of artifact. The specific operations of this method are: 1) The artifact image is transformed into the range-Doppler domain through Fourier transform. 2) The range-Doppler domain is sub-band divided and inverse Fourier transformed to obtain sub-images. 3) The sub-images are spliced ​​into a dictionary, the sub-image phases are retained, and the sub-image moduli are subjected to low-rank and sparse decomposition. 4) The low-rank parts are sub-band spliced ​​through Fourier transform, and the final suppressed image is obtained through inverse Fourier transform.

[0003] However, narrow pulse artifact suppression methods based on Doppler sub-view decomposition only suppress the modulus of each artifact sub-image, while preserving its phase. If the phase is not processed, the final suppressed image reconstructed through sub-image stitching and inverse Fourier transform steps may contain residual artifacts. Summary of the Invention

[0004] The purpose of the present invention is to provide an artifact suppression method based on phase reconstruction, which uses the modulus value of a known clean sub-image to reconstruct the phase of the sub-image, so that the final suppressed image reconstructed through the sub-band splicing and inverse Fourier transform steps will not have any artifact residues.

[0005] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0006] A phase reconstruction-based artifact suppression method comprises: firstly, transforming an artifact-containing synthetic aperture radar image into a range-frequency domain-azimuth Doppler domain through a two-dimensional Fourier transform, then constructing a sub-image from the range-frequency domain-azimuth Doppler domain, and constructing a dictionary based on the sub-image; after performing low-rank and sparse decomposition on the dictionary, splitting and recombining the solved low-rank matrix to construct a sub-view low-rank matrix; then constructing an initial sub-phase hologram in the range-frequency domain-azimuth Doppler domain, converting it to obtain a sub-image with a phase in the time domain, calculating the root mean square error between the modulus of the sub-image and the corresponding sub-view low-rank matrix, and repeatedly performing this step until the phase of the sub-image is considered correct after meeting the requirements; finally, splicing and inverse Fourier transforming all the sub-images to reconstruct the representation matrix, and using the reconstructed representation matrix to obtain an artifact-free synthetic aperture radar image.

[0007] Furthermore, constructing a sub-image from the range frequency domain-azimuth Doppler domain and constructing a dictionary based on the sub-image includes:

[0008] According to the representation matrix X(f τ ,f η ) The number of sampling points N corresponding to the known Doppler width Ba The number of sampling points N corresponding to the signal bandwidth Br , using the initial phase θ to characterize the matrix X(f τ ,f η ) is evenly divided into M×N sub-matrices, so that the number of distance frequency domain sampling points of each sub-matrix is The number of sampling points in the azimuth Doppler domain is

[0009] Perform two-dimensional inverse Fourier transform on the M×N sub-matrices to obtain the corresponding M×N sub-images; the m×nth sub-image is X mn (τ, η), m=1,2,...,M; n=1,2,...,N;

[0010] For M×N sub-images X mn (τ,η) is straightened, and the sub-image X mn (τ,η) will be stacked in the direction and arranged into a After straightening all sub-images, we get M×N column vectors; then, we concatenate the M×N column vectors to form a matrix, and use this matrix as the dictionary D.

[0011] Furthermore, performing low-rank and sparse decomposition on the dictionary includes:

[0012] Decompose the dictionary D into a low-rank matrix and a sparse matrix, expressed as:

[0013]

[0014] Where L is a low-rank matrix, S represents a sparse matrix, rank(·) represents a rank operation, r represents an estimated rank, card(·) represents a sparsity statistical operation, k represents the estimated number of artifact points, ||·|| F represents the F norm of the matrix;

[0015] This problem can be broken down into the following two sub-problems:

[0016]

[0017] Among them, L t and S t Respectively represent the low-rank matrix and sparse matrix obtained at the t-th iteration;

[0018] By alternately iteratively solving the above two sub-problems, we can obtain the low-rank matrix L corresponding to the joint sub-view matrix. When the rank of the low-rank matrix L is equal to the estimated rank r, the iteration is stopped:

[0019]

[0020] where svd(·) represents singular value decomposition, U and V are orthogonal matrices, Λ is a diagonal matrix, superscript T represents the transpose operation, and λ i represents the i-th largest singular value in the diagonal matrix, U i and V i is the column vector corresponding to the i-th largest singular value in the orthogonal matrices U and V; P Ω (·) represents hard threshold decision, Ω is DL t The first k largest non-zero subsets of Ω, i,j represents the element at the (i,j)th position in Ω.

[0021] Furthermore, the step of splitting and recombining the solved low-rank matrix to construct a sub-view low-rank matrix includes:

[0022] The low-rank matrix L obtained by the solution is subjected to the inverse operation of straightening and splicing. First, the low-rank matrix L is split according to the direction to obtain M×N matrices of size Column vector, and then the M×N column vectors are all arranged in the direction of length Cyclic reorganization along the azimuth direction finally results in an M×N sub-view low-rank matrix L mn (τ,η).

[0023] Furthermore, the initial sub-phase hologram is constructed in the range frequency domain-azimuth Doppler domain and converted into a sub-image with a phase in the time domain, including:

[0024] The initial phase θ is divided into M×N parts, so that the number of range frequency domain sampling points of each part is The number of sampling points in the azimuth Doppler domain is Thus, M×N initial sub-phase holograms in the range frequency domain and azimuth Doppler domain are obtained.

[0025] For each initial sub-phase hologram Perform two-dimensional inverse Fourier transform respectively to obtain sub-images of phase in the time domain reserve Phase As the sub-view low-rank matrix L mn Initial phase distribution of (τ,η).

[0026] Furthermore, the step of calculating the root mean square error between the modulus of the sub-image and the corresponding sub-view low-rank matrix and repeatedly performing this step until the phase of the sub-image is considered correct after the requirements are met includes:

[0027] Sub-image of phase in time domain The modulus value is replaced by the sub-view low-rank matrix L mn (τ,η) to form a new complex matrix

[0028] The new complex matrix Perform a two-dimensional Fourier transform to obtain the expression of range frequency domain-azimuth Doppler domain While retaining Phase, M×N sub-phase holograms in the range frequency domain-azimuth Doppler domain are obtained for the next iteration cycle;

[0029] Iterate repeatedly until the defined root mean square error (SSE) reaches the set accuracy or the set maximum number of iterations is reached.

[0030] Furthermore, the step of stitching all sub-images and performing inverse Fourier transform to reconstruct a representation matrix, and obtaining an artifact-free synthetic aperture radar image using the reconstructed representation matrix, includes:

[0031] For the M×N phase sub-images in the time domain that meet the accuracy requirements or the maximum number of iterations, Perform two-dimensional Fourier transform respectively to obtain its expression in the two-dimensional frequency domain right According to the representation matrix X(f τ ,f η ) is divided into sub-matrices, and the M×N Place it in the corresponding position and get the reconstructed representation matrix X′(f τ ,f η );

[0032] The reconstructed representation matrix X′(f τ ,f η ) performs a two-dimensional inverse Fourier transform to obtain the reconstructed artifact-free synthetic aperture radar image X F (τ,η).

[0033] A terminal device comprises a memory, a processor and a computer program stored in the memory and runnable on the processor; when the processor executes the computer program, the artifact suppression method based on phase reconstruction is implemented.

[0034] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the phase reconstruction-based artifact suppression method is implemented.

[0035] Compared with the prior art, the present invention has the following technical features:

[0036] This paper addresses the problem of artifact suppression methods based on Doppler sub-view decomposition not properly processing the phase. Based on the Doppler sub-view decomposition method, this method continuously updates the phase of the clean sub-image using the angular spectrum iteration principle based on the modulus value of the clean sub-image. This effectively solves the problem of residual artifacts caused by the Doppler sub-view decomposition method that retains the phase of the original artifact image. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of a specific implementation flow of the method of the present invention;

[0038] Figure 2 (a) is the original artifact image in the embodiment of the present invention, (b) is the final image obtained by using the 8-view Doppler sub-view decomposition method for (a), and (c) is the image reconstructed by using this method for (a). DETAILED DESCRIPTION

[0039] The present invention provides an artifact suppression method based on phase reconstruction, comprising: first, transforming an artifact-containing synthetic aperture radar image into a range-frequency domain-azimuth Doppler domain through a two-dimensional Fourier transform, then constructing sub-images from the range-frequency domain-azimuth Doppler domain, and constructing a dictionary based on the sub-images; after performing low-rank and sparse decomposition on the dictionary, splitting and recombining the solved low-rank matrix to construct a sub-view low-rank matrix; then constructing an initial sub-phase hologram in the range-frequency domain-azimuth Doppler domain, and converting it to obtain a sub-image with a phase in the time domain; calculating the root mean square error between the modulus of the sub-image and the corresponding sub-view low-rank matrix, and repeatedly performing this step until the phase of the sub-image is considered correct after meeting the requirements; finally, splicing all the sub-images and performing an inverse Fourier transform to reconstruct the representation matrix, and using the reconstructed representation matrix to obtain an artifact-free synthetic aperture radar image. The specific implementation process of the present invention is further described in detail below with reference to the accompanying drawings.

[0040] Step 1: Perform a two-dimensional Fourier transform on the synthetic aperture radar image X(τ,η) containing artifacts to obtain the image representation matrix X(f τ ,f η ) and the corresponding initial phase θ, expressed as:

[0041]

[0042] The representation matrix τ is the fast time in distance dimension, η is the slow time in azimuth dimension, f τ is the distance dimension frequency, f η is the azimuthal Doppler frequency, e is a natural constant, j is an imaginary unit, The number of sampling points in the orientation dimension of the matrix is ​​N a , the number of sampling points in the distance dimension is Nr , Represents complex space.

[0043] At the same time, record the representation matrix X(f τ ,f η ) is expressed as:

[0044]

[0045] Where real(·) and imag(·) represent the real and imaginary part operations respectively.

[0046] Step 2: According to the representation matrix X(f τ ,f η ) The number of sampling points corresponding to the known Doppler width The number of sampling points corresponding to the signal bandwidth Using the initial phase θ, the representation matrix X(f τ ,f η ) is evenly divided into M×N sub-matrices, so that the number of distance frequency domain sampling points of each sub-matrix is The number of sampling points in the azimuth Doppler domain is The values ​​of M and N can be set arbitrarily according to requirements.

[0047]

[0048] in, Denotes the representation matrix X(f τ ,f η ) in (N r ,N a ) is the submatrix at .

[0049] Step 3: Perform two-dimensional inverse Fourier transform on each of the M×N sub-matrices to obtain the corresponding M×N sub-images, where the m×nth sub-image is Where m = 1, 2, ..., M; n = 1, 2, ..., N.

[0050] Step 4: For the M×N sub-images X mn The modulus of (τ,η) Perform straightening operation, sub-image X mn (τ,η) will be stacked in the direction and arranged into a After straightening all sub-images, we get M×N column vectors; then, we concatenate the M×N column vectors to form a matrix, and use this matrix as the dictionary The straightening and splicing can be expressed as:

[0051]

[0052] Step 5, decompose the dictionary D into a low-rank matrix and a sparse matrix. The problem can be expressed as:

[0053]

[0054] in, is a low-rank matrix, represents a sparse matrix, rank(·) represents a rank operation, r represents the estimated rank, card(·) represents a sparsity statistical operation, k represents the estimated number of artifact points, ||·|| F Represents the F-norm of the matrix.

[0055] This problem can be broken down into the following two sub-problems:

[0056]

[0057] Among them, L t and S t represent the low-rank matrix and sparse matrix obtained at the t-th iteration respectively.

[0058] By alternately iteratively solving the above two sub-problems, we can obtain the low-rank matrix L corresponding to the joint sub-view matrix. When the rank of the low-rank matrix L is equal to the estimated rank r, the iteration is stopped:

[0059]

[0060] where svd(·) represents singular value decomposition, U and V are orthogonal matrices, Λ is a diagonal matrix, superscript T represents the transpose operation, and λ i represents the i-th largest singular value in the diagonal matrix, U i and V i is the column vector corresponding to the i-th largest singular value in the orthogonal matrices U and V; P Ω (·) represents hard threshold decision, Ω is DL t The first k largest non-zero subsets of , i, j represents the element at the (i, j)th position in Ω; in the tth iteration, DL t The value of Ω that does not meet the decision condition is set to 0.

[0061] Step 6: Perform the inverse operation of straightening and splicing in step 4 on the low-rank matrix L obtained in step 5, that is, first split the low-rank matrix L according to the direction to obtain M×N matrices of size Column vector, and then the M×N column vectors are all arranged in the direction of length Cyclic reorganization along the azimuth direction finally results in an M×N sub-view low-rank matrix

[0062] Step 7, according to the representation matrix X(f τ ,f η) The number of sampling points corresponding to the known Doppler width The number of sampling points corresponding to the signal bandwidth Divide the initial phase θ determined in step 1 into M×N parts, so that the number of range-frequency sampling points in each part is The number of sampling points in the azimuth Doppler domain is Thus, M×N initial sub-phase holograms in the range frequency domain and azimuth Doppler domain are obtained.

[0063] Step 8, start iteration: for each initial sub-phase hologram Perform two-dimensional inverse Fourier transform respectively to obtain sub-images of phase in the time domain It can be specifically expressed as:

[0064]

[0065] in represents the sub-phase hologram in the range-frequency domain and azimuth Doppler domain of the lth iteration, It represents the expression of the m×nth child phase diagram in the time domain at the lth iteration, where m=1,2,...,M; n=1,2,...,N.

[0066] Step 9, keep Phase As the sub-view low-rank matrix L mn The initial phase distribution of (τ,η) can be specifically expressed as:

[0067]

[0068] The sub-image in step 3 is a complex matrix with artifacts, and the sub-view low-rank matrix obtained in step 6 is a real matrix without artifacts. Therefore, the phase needs to be added in step 9 to obtain a complex matrix so that the entire image can be restored through a two-dimensional inverse Fourier transform. The initial phase distribution here still has artifacts, which will affect the final image result, so it is necessary to use the artifact-free real matrix information to reconstruct it into an artifact-free phase through phase iteration.

[0069] Sub-image of phase in time domain The modulus value Replace it with the sub-view low-rank matrix L obtained in step 6 mn (τ,η) to form a new complex matrix It can be specifically expressed as:

[0070]

[0071] Step 10: Transform the new complex matrix Perform a two-dimensional Fourier transform to obtain the expression of range frequency domain-azimuth Doppler domain While retaining The phase of the image is obtained by multiplying the image by 1 and 2 to obtain M×N sub-phase holograms in the range frequency domain-azimuth Doppler domain for the next iteration cycle.

[0072] Step 11: Repeat steps 8 to 10 until the defined root mean square error (SSE) reaches the set accuracy or reaches the set maximum number of iterations. The root mean square error (SSE) is defined as:

[0073]

[0074] Step 12: for the M×N phase sub-images in the time domain that meet the accuracy requirement or the maximum number of iterations, Perform two-dimensional Fourier transform respectively to obtain its expression in the two-dimensional frequency domain right According to the positions divided in step 2, Place it in the corresponding position and get the reconstructed representation matrix X′(f τ ,f η ); the reconstructed matrix can be expressed as:

[0075]

[0076] Step 13: Reconstruct the representation matrix X′(f τ ,f η ) performs a two-dimensional inverse Fourier transform to obtain the reconstructed artifact-free synthetic aperture radar image X F (τ,η).

[0077] Example:

[0078] In one embodiment of the present invention, Figure 2 As shown in (a), when using this method, the specific configuration is: the number of azimuth sampling points N a is 14259, the number of distance dimension sampling points N r is 27438, the number of sampling points corresponding to the known Doppler width is N Ba The number of sampling points N corresponding to the signal bandwidth is 21954. Br The representation matrix is ​​divided into 8 equal parts, so that the number of sampling points in the range frequency domain and the number of sampling points in the azimuth Doppler domain are 3173 and 10977, respectively. The final reconstructed image is as follows: Figure 2 As shown in (c).

[0079] exist Figure 2 In (b), we can see that there is an obvious artifact horizontal line at the position of the red arrow; Figure 2There is no residual artifact in (c), which verifies the effectiveness of the present invention.

[0080] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A method for suppressing artifacts based on phase reconstruction, characterized in that: include: First, the synthetic aperture radar image containing artifacts is transformed into the range-frequency domain-azimuth Doppler domain through two-dimensional Fourier transform, and then a sub-image is constructed from the range-frequency domain-azimuth Doppler domain, and a dictionary is constructed based on the sub-image; after the dictionary is low-rank and sparse decomposition, the solved low-rank matrix is ​​split and reorganized to construct a sub-view low-rank matrix, and then the initial sub-phase hologram is constructed in the range-frequency domain-azimuth Doppler domain, and the phase sub-image in the time domain is converted. The root mean square error between the modulus of the sub-image and the corresponding sub-view low-rank matrix is ​​calculated and the step is repeated until the phase of the sub-image is considered correct after meeting the requirements; finally, all sub-images are spliced ​​and inverse Fourier transformed to reconstruct the representation matrix, and the reconstructed representation matrix is ​​used to obtain the synthetic aperture radar image without artifacts; The step of constructing a sub-image from the range-frequency domain-azimuth Doppler domain and constructing a dictionary based on the sub-image includes: According to the representation matrix The number of sampling points corresponding to the known Doppler width The number of sampling points corresponding to the signal bandwidth , using the initial phase The representation matrix The average score is The molecular matrix makes the number of sampling points in the distance frequency domain of each molecular matrix be , the number of sampling points in the azimuth Doppler domain is ; right The molecular matrix performs two-dimensional inverse Fourier transform to obtain the corresponding sub-images; The sub-image is , ; right sub-images The model is straightened, and the sub-image They will be stacked in the direction and arranged into a After straightening all sub-images, we get column vectors; then, Column vectors are concatenated to form a matrix, and this matrix is ​​used as a dictionary .

2. The artifact suppression method based on phase reconstruction according to claim 1, characterized in that: The low-rank and sparse decomposition of the dictionary includes: The dictionary Decomposed into low-rank matrix and sparse matrix, expressed as: in, is a low-rank matrix, represents a sparse matrix, represents the rank operation, represents the estimated rank, represents the sparsity statistical operation, Indicates the estimated number of artifact points, represents the F norm of the matrix; The dictionary Decomposition into low-rank matrix and sparse matrix problem is decomposed into the following two sub-problems: in, and Respectively represent t The low-rank matrix and sparse matrix obtained by the iteration; Alternately iterate and solve the above two sub-problems to obtain the low-rank matrix corresponding to the joint sub-view matrix , when the low-rank matrix The rank of is equal to the estimated rank Stop the iteration when: in represents singular value decomposition, U and V are orthogonal matrices, It is a diagonal matrix, and the superscript T indicates the transpose operation. represents the i-th largest singular value in the diagonal matrix, and is the column vector corresponding to the i-th largest singular value in the orthogonal matrices U and V; represents the hard threshold decision, Ω is Before k The largest non-zero subset of elements, represents the first The element at position.

3. The artifact suppression method based on phase reconstruction according to claim 1, characterized in that: The step of splitting and recombining the solved low-rank matrix to construct a sub-view low-rank matrix includes: The low-rank matrix obtained by solving Perform the inverse operation of straightening and splicing, first transform the low-rank matrix L Split by direction, we get The size is Column vector of The column vectors are all arranged in the direction of length Cyclic reorganization along the azimuth direction finally yields Subview low-rank matrix .

4. The artifact suppression method based on phase reconstruction according to claim 1, characterized in that: The initial sub-phase hologram is constructed in the range frequency domain-azimuth Doppler domain and converted into a sub-image of the phase in the time domain, including: The initial phase The average score is The number of distance frequency domain sampling points in each portion is , the number of sampling points in the azimuth Doppler domain is , thus obtaining An initial sub-phase hologram in the range frequency domain-azimuth Doppler domain ; For each initial sub-phase hologram Perform two-dimensional inverse Fourier transform respectively to obtain sub-images of phase in the time domain ;reserve Phase , as the sub-view low-rank matrix The initial phase distribution.

5. The artifact suppression method based on phase reconstruction according to claim 1, characterized in that: The step of calculating the root mean square error between the modulus of the sub-image and the corresponding sub-view low-rank matrix and repeatedly performing this step until the phase of the sub-image is considered correct after the requirements are met includes: Sub-image of phase in time domain The modulus value is replaced by the sub-view low-rank matrix To form a new complex matrix ; The new complex matrix Perform a two-dimensional Fourier transform to obtain the expression of range frequency domain-azimuth Doppler domain , while retaining The phase of A sub-phase hologram in the range frequency domain-azimuth Doppler domain is generated for the next iteration cycle; Repeat the iteration until the defined root mean square error SSE Achieve the set accuracy or reach the set maximum number of iterations.

6. The artifact suppression method based on phase reconstruction according to claim 1, characterized in that: The method of stitching all sub-images and performing inverse Fourier transform to reconstruct a representation matrix, and obtaining an artifact-free synthetic aperture radar image using the reconstructed representation matrix, includes: To achieve the accuracy requirement or the maximum number of iterations Phase sub-image in the time domain Perform two-dimensional Fourier transform respectively to obtain its expression in the two-dimensional frequency domain ,right According to the representation matrix The positions of the divided sub-matrices are indivual Place it in the corresponding position to obtain the reconstructed representation matrix ; The reconstructed representation matrix Perform a two-dimensional inverse Fourier transform to obtain the reconstructed artifact-free synthetic aperture radar image .

7. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; When the processor executes the computer program, the artifact suppression method based on phase reconstruction according to any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium storing a computer program; wherein: When the computer program is executed by a processor, the artifact suppression method based on phase reconstruction according to any one of claims 1 to 6 is implemented.

Images