SAR (Synthetic Aperture Radar) echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition

Abstract

The invention discloses an SAR echo suppression interference suppression method, and belongs to the technical field of SAR signal processing. The method comprises the following steps of: firstly, preprocessing an SAR echo matrix according to a sequence of zero equalization and energy normalization, and then symmetrically filling zero for each frame of interfered echo of the preprocessed matrix forwards at the head end and backwards at the tail end and then performing short-time Fourier transform to obtain corresponding time-frequency domain representation; decomposing a time-frequency domain into a structural component representing interference and a texture component representing pure echoes, performing inverse short-time Fourier transform on the texture component, intercepting an effective signal segment without mutation according to a previous zero padding position, and storing the effective signal segment in an interference suppression echo matrix; according to the method, the structural component and the texture component of the SAR echo time-frequency domain are separated, the damage of suppressing interference to the echo quality is suppressed, and then the detail features of the SAR echo after interference suppression are enhanced through wavelet detail coefficient reconstruction.

Description

Technical Field

[0001] This invention relates to the field of synthetic aperture radar (SAR) signal processing technology, and more specifically to a SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition. Background Technology

[0002] As an important member of the radar family, Synthetic Aperture Radar (SAR) has been widely used due to its all-weather, all-day operation capability. However, without any anti-jamming protection measures, electromagnetic interference can weaken or even paralyze SAR's information acquisition capabilities. Therefore, anti-jamming has always been a research hotspot in the SAR academic community. Electromagnetic interference can be divided into suppression interference and deception interference according to different mechanisms: the former has stronger energy than the echo and hinders SAR's detection of specific signals, while the latter has characteristics similar to the echo and increases the false alarm rate of SAR in detecting specific signals. Since suppression interference can form bright stripes or strong noise on SAR images, or even saturate the images, it can seriously damage the quality of SAR images. Therefore, this invention specifically innovates a suppression method for suppression interference.

[0003] Based on their bandwidth, interference suppression can be categorized into narrow-band interference (NBI) and wide-band interference (WBI): the former typically has a bandwidth less than 1% of the echo bandwidth, while the latter typically has a bandwidth greater than 10% of the echo bandwidth. Generally, narrow-band interference (NBI) is modeled as a superposition of sinusoidal signals, while wide-band interference (WBI) is further divided into chirp-modulated wide-band interference (CMWBI) and sinusoidal-modulated wide-band interference (SMWBI).

[0004] In existing technologies, Huang Yan et al., in their paper "A Review of Synthetic Aperture Radar Anti-jamming Technology" (Journal of Radar, Vol. 9, No. 1, February 2020), categorized SAR jamming suppression techniques into three types: non-parametric methods, parametric methods, and semi-parametric methods. Non-parametric methods construct jamming filters based on the differences between the jamming and the echo in the time, frequency, or time-frequency domains, or separate the jamming from the echo by projecting the echo onto the jamming subspace. Least mean square filtering, notch filtering, and eigenspace projection are classic non-parametric methods. Generally, non-parametric methods are suitable for situations with strong jamming, but they only consider the characteristics of the jamming and lack protection for the echo. Parametric methods establish mathematical models based on the statistical characteristics or spatial distribution of the jamming and the echo, achieving the purpose of separating the jamming by estimating the jamming model parameters. Least square fitting, iterative adaptive methods, and high-order fuzzy function estimation are classic parametric methods. Generally, parametric methods are suitable for isolated and single-type jamming, but pulse-by-pulse estimation of jamming parameters leads to high computational complexity, and they also lack protection for the echo. Semi-parametric methods have been a hot technology in SAR anti-jamming for over a decade. Based on sparse representation, low-rank recovery, matrix factorization, and tensor factorization, they transform interference separation into an optimization problem, enabling the suppression of interference while protecting the echo. However, the performance of semi-parametric methods in suppressing interference depends heavily on the hyperparameter values ​​and the selection of the optimization model; moreover, the computational complexity per pulse is even higher than that of parametric methods. In recent years, deep learning theory has also begun to be applied to SAR anti-jamming, resulting in interference suppression methods based on convolutional neural networks, generative adversarial networks, and gated recurrent units. However, the effectiveness of these methods depends heavily on the number of samples and the network size, making them prone to overfitting or underfitting, and the intermediate results have poor interpretability, hindering algorithm evaluation and improvement.

[0005] Therefore, how to better suppress narrowband and broadband interference remains a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition to overcome the shortcomings of current SAR echo suppression interference suppression technology.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A time-frequency domain structure-texture decomposition-guided SAR echo suppression interference reduction method includes the following steps:

[0009] Obtain the SAR echo matrix and preprocess the echo matrix in the order of first zero-mean normalization and then energy normalization;

[0010] Interference detection is performed on the transmitted pulse echoes of each frame in the preprocessed echo matrix to obtain the interfered echo matrix and the interference-free echo matrix.

[0011] After padding each frame of the interfered echo in the interfered echo matrix with zeros at both ends, a short-time Fourier transform is performed to obtain the corresponding time-frequency domain representation.

[0012] The time-frequency domain representation of each frame of disturbed echo is decomposed into structural components representing the disturbance and texture components representing the pure echo by using a weighted curvature-driven energy functional minimization model.

[0013] Perform an inverse short-time Fourier transform on the texture components and extract the effective signal segments without abrupt changes based on the previously padded zero positions, and save them into the interference suppression echo matrix.

[0014] Further, the SAR echo matrix is ​​obtained, and the echo matrix is ​​preprocessed in the order of first zero-mean normalization and then energy normalization, specifically including:

[0015] Record the echoes of each transmitted pulse of the SAR in a matrix. In the diagram, is the echo matrix of the SAR, where matrix R is... e a line of This represents a frame of transmitted pulse echo;

[0016] For the echo matrix R e Each frame of transmitted pulse echo is zero-mean processed to obtain a zero-mean matrix. in, i∈[1,M] represents the average value of the transmitted pulse echo in the i-th frame;

[0017] For zero-mean matrix D c Each row is processed to normalize energy, resulting in an energy normalization matrix.

[0018] in,

[0019] σ represents the energy value of the transmitted pulse echo in the i-th frame, and σ represents the set uniform energy value.

[0020] Furthermore, interference detection is performed on the transmitted pulse echoes of each frame in the preprocessed echo matrix to obtain the interfered echo matrix and the interference-free echo matrix, specifically including:

[0021] Interference detection is performed on the transmitted pulse echo of each frame in the preprocessed echo matrix;

[0022] The detected interference echoes are recorded in the first complex matrix to obtain the interference echo matrix;

[0023] The undisturbed echoes are recorded into the second complex matrix to obtain the undisturbed echo matrix.

[0024] Furthermore, after padding each frame of the interfered echo in the interfered echo matrix with zeros at both ends, a short-time Fourier transform is performed to obtain the corresponding time-frequency domain representation, specifically including:

[0025] Each row in the interference echo matrix represents the interference echo of each frame.

[0026] For the disturbed echo matrix R j Each frame of the interference echo Leading zeros are padded at the beginning of each line, and for each line... Zeros are padded to the end of the signal to obtain the zero-padded interference echo signal. The number of zeros used for forward and backward padding is the same, and the total number of zeros is taken from the number of interference echoes in each frame. 10% to 50% of the length;

[0027] The zero-padding interference echo signal Performing a short-time Fourier transform yields the corresponding time-frequency domain representation, STFT0:

[0028]

[0029] Where N is the signal The length of the discrete Fourier transform can also be viewed as the total number of points in the discrete Fourier transform. Indicates signal The k-th element, k = 1, 2, ..., N; ω * (km) represents the signal The discrete-time window function, where m is the center index of the window function, m = 1, 2, ..., N, and k represents the area of ​​the signal covered by the window function. The k-th element; n is the discrete frequency index, n = 1, 2, ..., N, and * is the complex conjugate symbol.

[0030] Furthermore, a weighted curvature-driven energy functional minimization model is used to decompose the time-frequency domain representation of a frame of disturbed echoes into structural components representing the disturbance and texture components representing the pure echoes, specifically including:

[0031] Establish a weighted curvature-driven energy functional minimization model for the time-frequency domain representation matrix STFT0 of a frame of interfered echo:

[0032]

[0033] In this equation, the first integral term is called the linear regularization term, the second integral term is called the weighted curvature term, and the third integral term is called the approximation term; λ is a positive constant. It is the gradient operator, Δ-1 It is the inverse Laplace operator;

[0034] It is a weighted function, where β is a positive constant and G is a negative constant. σ The Gaussian kernel function is represented by the asterisk (*), and the asterisk (*) indicates convolution. h(x) = 1 + t·c(x)·e(x) is the exponential function, where t is a positive constant. It is a deviation function. Sup is the Butterworth low-pass filter result of the horizontal set curvature of matrix STFT0, where sup represents the supremum of the set; K is the Butterworth low-pass filter result of the horizontal set curvature of the structural component U of matrix STFT0; V is the texture component of matrix STFT0. It is a gradient-dependent nonconvex function of matrix STFT0; is the weighted level set curvature of matrix STFT0, div is the divergence; Q is the zero-div vector field, divQ=0;

[0035] The optimal separation of U and V is obtained by solving the weighted curvature-driven energy functional minimization model, where U represents the structural component of the disturbance and V represents the texture component of the pure echo.

[0036] Furthermore, the structural components representing interference include one or more of narrowband interference (NBI), frequency modulation broadband interference (CMWBI), and sinusoidal modulation broadband interference (SMWB).

[0037] Furthermore, an inverse short-time Fourier transform is performed on the texture components, and effective signal segments without abrupt changes are extracted based on the previously padded zero positions and stored in the interference suppression echo matrix. Specifically, this includes:

[0038] Performing an inverse short-time Fourier transform on the composite texture component X yields the transformed texture signal:

[0039]

[0040] Where N is the length of the original signal, which can also be regarded as the total number of points of the discrete Fourier transform of the original signal; M is the total number of short-time signal segments into which the original signal is divided; ω(km) represents the additional discrete-time window function after performing the inverse Fourier transform on the spectrum X(m,n) of the m-th short-time signal segment, m=1,2,…,M, k represents the k-th element of the original signal, k=1,2,…,N; n is the discrete frequency index, n=1,2,…,N.

[0041] Based on the interference echo signal after zeroing at the beginning and end The number of zeros padded at the beginning and end is used to truncate the beginning of x(k) backward and the end forward, respectively, and the middle part is retained to obtain the effective signal y(t) without abrupt changes, that is, to obtain the interference suppression echo;

[0042] All interference suppression echoes are combined to form the interference suppression echo matrix Y.

[0043] Furthermore, the above method also includes,

[0044] Five-level wavelet transform is used to enhance the details of the interference suppression echo matrix.

[0045] Furthermore, the details of the interference suppression echo matrix are enhanced using a five-level wavelet transform, specifically including:

[0046] Calculate the amplitude matrix and phase angle matrix of the interference suppression echo matrix Y;

[0047] Based on wavelet transform, the amplitude matrix M and the phase angle matrix A are reconstructed to obtain the corresponding reconstruction matrix M. r and A r ;

[0048] Reconstruct matrix M r and A r Linearly map to the interference-free echo matrix R as follows: c Amplitude matrix M c and phase angle matrix A c Within the range of values, matrix M is obtained. m and A m :

[0049]

[0050] For matrix M m and A m Perform global zero-mean normalization as follows to obtain matrix M. z and A z

[0051]

[0052] in, It is matrix M m average value It is matrix A m The average value;

[0053] matrix M z and A z According to Z=M z ·exp(A z The interference suppression echo matrix is ​​reorganized into a complex matrix in a manner that enhances the details.

[0054] Furthermore, the amplitude matrix M is reconstructed based on wavelet transform to obtain the corresponding reconstruction matrix Mreconstructed. r Specifically, it includes:

[0055] Perform a five-level wavelet transform on the amplitude value matrix M, and extract the horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix of the amplitude value matrix M obtained after the first level transform;

[0056] The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix of the amplitude value matrix M are all upsampled in both row and column dimensions to obtain coefficient matrices of the same size as the amplitude value matrix M.

[0057] By using a synthesized filter bank that is orthogonal to the forward wavelet transform analysis filter bank, the row-dimensional and column-dimensional convolutions are performed on the upsampled horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix, respectively, to obtain the corresponding inverse filter coefficient matrix;

[0058] The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix after inverse filtering are superimposed and merged to obtain a reconstruction matrix that contains only high-frequency information of the amplitude value matrix M.

[0059] Furthermore, the phase angle matrix A is reconstructed based on wavelet transform to obtain the corresponding reconstructed matrix A. r Specifically, it includes:

[0060] Perform a five-level wavelet transform on the phase angle matrix A, and extract the horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix of the phase angle matrix A obtained after the first level transform;

[0061] The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix of the phase angle matrix A are all upsampled in both row and column dimensions to obtain coefficient matrices of the same size as the phase angle matrix A.

[0062] By using a synthesized filter bank that is orthogonal to the forward wavelet transform analysis filter bank, the row-dimensional and column-dimensional convolutions are performed on the upsampled horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix, respectively, to obtain the corresponding inverse filter coefficient matrix;

[0063] The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix after inverse filtering are superimposed and merged to obtain a reconstruction matrix that contains only high-frequency information of the phase angle matrix A.

[0064] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition, which has the following beneficial effects:

[0065] The method disclosed in this invention focuses on the time-frequency domain structure-texture decomposition of the interfered echo. Interference suppression is achieved by inversely transforming the texture components to the time domain, followed by detail enhancement of the interference-suppressed echo based on a five-level wavelet transform. Compared to existing methods, the method disclosed in this invention can effectively suppress narrowband and broadband interference without relying on sample selection and hyperparameter values, while also possessing a certain echo protection capability and appropriate computational complexity. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0067] Figure 1 The overall flow of the SAR echo suppression interference suppression method disclosed in this invention is shown.

[0068] Figure 2 The detailed process of the time-frequency domain structure-texture decomposition method used in this invention is shown.

[0069] Figure 3 The figure illustrates the key principle of the time-frequency domain structure-texture decomposition method used in this invention. Right now Right now

[0070] Figure 4 The detailed process of the interference suppression echo matrix detail enhancement method disclosed in this invention is shown.

[0071] Figure 5 This invention demonstrates the key principles of the interference suppression echo matrix detail enhancement method disclosed in this invention.

[0072] Figure 6 The data flow of the SAR echo suppression interference reduction method disclosed in this invention is shown. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only one embodiment of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] The time-frequency domain structure-texture decomposition-guided SAR echo suppression interference suppression method disclosed in this invention can achieve simultaneous suppression of narrowband and broadband interference under certain computational complexity, and can reduce echo distortion caused by the interference suppression process. The overall process of this method is as follows: Figure 1 As shown:

[0075] First, the SAR echo matrix is ​​acquired and preprocessed. Then, interference detection is performed on the preprocessed echoes, storing the interfered echoes in the interfered echo matrix and the undisturbed echoes in the non-interference echo matrix. Next, a short-time Fourier transform is performed on each frame of interfered echoes, followed by time-frequency domain structure-texture decomposition. Subsequently, an inverse short-time Fourier transform is performed on the texture components to obtain the interference-suppressed echoes. Finally, detail enhancement is performed on the interference-suppressed echo matrix. In this embodiment of the invention, it is assumed that the presence of interference has already been determined based on the shape of the echoes in the time domain, frequency domain, or time-frequency domain. Therefore, in this embodiment of the invention, the interference detection of echoes to obtain interfered and undisturbed echoes is a known prior art, and this invention will not describe it in detail.

[0076] The embodiments of the present invention may specifically include the following steps:

[0077] Step S1: Echo matrix preprocessing.

[0078] Record the echoes of each transmitted pulse of the SAR in a matrix. Each row of the matrix That is, the average value of a frame of transmitted pulse echo is To eliminate the DC component of each frame echo and reflect the fluctuating nature of that frame echo, this invention proposes to optimize the R... e Each frame of echo is zero-mean processed to obtain a matrix.

[0079]

[0080] Each row of the matrix Possesses energy To unify the energy of each row to the σ value, this invention proposes to adjust D... c Each row is energy normalized to obtain the matrix.

[0081]

[0082] Step S2: Preprocessing echo interference detection.

[0083] matrix E u The distorted echo and the undisturbed echo are respectively recorded in the first complex matrix R j The second complex matrix R c In this context, the first complex matrix is ​​specifically represented as follows: The second complex matrix Rc Represented as

[0084] Calculate matrix R c Amplitude matrix M c =|R c | and phase angle matrix Used for linear mapping calculation in step six.

[0085] Step S3: Short-time Fourier transform of the disturbed echo.

[0086] For matrix R j All industries Perform a short-time Fourier transform to obtain the time-frequency domain representation.

[0087]

[0088] Where N is the signal The length of the discrete Fourier transform can also be viewed as the total number of points in the discrete Fourier transform. Indicates signal The k-th element, k = 1, 2, ..., N; ω * (km) represents the signal The discrete-time window function, where m is the center index of the window function, m = 1, 2, ..., N, and k represents the area of ​​the signal covered by the window function. The k-th element; n is the discrete frequency index, n = 1, 2, ..., N, and * denotes the complex conjugate sign. In actual calculations, the line signal is often... The system is divided into several segments of length K, with L points overlapping between the segments. Then, a Discrete Fourier Transform is performed on each segment. In the resulting time-frequency domain, NBI, CMWBI, and SMWBI appear as horizontal straight lines, sloping straight lines, and sine curves, respectively, nested within the chaotic texture patterns formed by the pure echo.

[0089] To avoid abrupt changes in texture morphology after the inverse short-time Fourier transform, we can first... The beginning and end of the array are symmetrically padded with zeros forward and backward, respectively. Furthermore, to mitigate boundary effects without excessively increasing computational burden, the total number of zeros can be set to a certain value. 10% to 50% of the length, then fill with zeros for the signal. The transformation yields the matrix STFT0, which does not alter the time-frequency domain form of the interference.

[0090] Step S4: Time-frequency domain structure-texture decomposition.

[0091] The time-frequency domain morphologies of NBI, CMWBI, and SMWBI can be viewed as the skeleton of matrix STFT0, also known as the time-frequency domain structural component, denoted by the letter U; simultaneously, the time-frequency domain morphology of the pure echo can be viewed as the background of matrix STFT0, also known as the time-frequency domain texture component, denoted by the letter V, thus STFT0 = U + V. To optimally separate U and V, this embodiment establishes a weighted curvature-driven energy functional minimization model for matrix STFT0.

[0092]

[0093] In this equation, the first integral term is called the linear regularization term, the second integral term is called the weighted curvature term, and the third integral term is called the approximation term; moreover, λ is a positive constant. It is the gradient operator, Δ -1 It is the inverse Laplace operator. (Using...) Figure 2 The process shown demonstrates how to solve this model to obtain the optimal separation of U and V. The key principle of the solution is as follows: Figure 3 As shown. Among them, for the linear regularization term, a weighting function is defined. In the formula, β is a positive constant, G σ It is the Gaussian kernel function, and * represents the convolution operation; at the same time, the exponential function h(x) is defined as 1 + t·c(x)·e(x), where t is a positive constant. It is a deviation function. K and are the level set curvatures of matrices STFT0 and U, respectively. and The Butterworth low-pass filtering results and sup denote the supremacy of the set. Under the action of c(x) and h(x), (S4-1) has better time-frequency domain edge and contour feature preservation capabilities. For the weighted curvature term, the gradient-dependent non-convex function of STFT0 is defined. and weighted level set curvature Multiplying these two factors makes the structural components solved by model (S4-1) closer to a piecewise uniform image separated by edges. Hodge decomposition is used for the approximation term. Where P is a scalar function and Q is a divergence vector field divQ = 0. The use of Q can make the separation results of U and V more accurate.

[0094] By employing the Alternate Direction Method of Multipliers (ADMM) and introducing auxiliary variables W, Z, and G, the unconstrained optimization problem of finding the minimum value of model (S4-1) can be transformed into the following constrained optimization problem.

[0095]

[0096] At this point, setting the solution set {U,W,Z,Q,G}=0 and introducing intermediate variables B1, B2, B3, and B4 with given initial values, the solution to model (S4-2) can be obtained by solving an unconstrained optimization problem of the following iterative form.

[0097]

[0098] Where the superscript k represents the k-th iteration, the superscript k+1 represents the (k+1)-th iteration, and Γ={U,W,Z,Q,G}, The coefficients μ1, μ2, μ3, and γ are positive constants, and Δ is the Laplace operator. By alternately updating certain elements of the set {U, W, Z, Q, G} while keeping other elements unchanged, model (S4-3) can be separated into the following five unconstrained optimization subproblems.

[0099]

[0100] For these five subproblems, we can use the soft thresholding method when the integrand is non-differentiable, and variational solutions when the integrand is differentiable, by having the derivative equal to zero. Thus, the solution to subproblem (S4-4) is...

[0101]

[0102] in, It is a soft thresholding function, and when |x|=0 In addition, the subscript is specified. The solution to subproblem (S4-5) is the Euler-Lagrange equation.

[0103] The solution. Among them, gradient operator adjacency operator, I is the unit operator. The solution to subproblem (S4-6) is the Euler-Lagrange equation.

[0104] The solution to subproblem (S4-7) is the Euler-Lagrange equation.

[0105] The solution to subproblem (S4-8) is...

[0106]

[0107] Wherein, function S r The definition of (x) is the same as in (S4-9), but the subscript is specified. Solve equations (S4-9) to (S4-13) sequentially until the condition is satisfied. The iteration stops when ε is reached. Here, ε is the stopping condition, and V = divG is the texture component. To improve the solution efficiency, the Fast Fourier Transform (FFT) and its inverse FFT (IFFT) can be used to accelerate the solution of equations (S4-10) to (S4-12). Taking equation (S4-11) as an example, the accelerated solution method is as follows:

[0108]

[0109] Step S5: Inverse short-time Fourier transform of texture components.

[0110] Summing the auxiliary variable G calculated from step S4 and the texture component V = divG calculated from G, we obtain the composite texture component X = V + G. Performing an inverse short-time Fourier transform on X yields...

[0111]

[0112] Where N is the length of the original signal, which can also be seen as the total number of points in the discrete Fourier transform of the original signal; M is the total number of short-time signal segments into which the original signal is divided; ω(km) represents the discrete-time window function added after performing an inverse Fourier transform on the spectrum X(m,n) of the m-th short-time signal segment, m = 1, 2, ..., M; k represents the k-th element of the original signal, k = 1, 2, ..., N; and n is the discrete frequency index, n = 1, 2, ..., N. At this point, based on the number of zeros padded forward and backward at the beginning and end of the interfered echo during the short-time Fourier transform, the beginning of x(k) is truncated backward and the end forward, retaining the effective signal segment y(t) without abrupt changes, thus obtaining the interference-suppressed echo. After all the interfered echoes have undergone this step, the interference-suppressed echo matrix Y is formed.

[0113] Step S6: Interference suppression echo matrix detail enhancement.

[0114] Calculate the magnitude matrix M = |Y| and phase angle matrix of matrix Y. Five-level wavelet transforms are performed on matrices M and A respectively. The horizontal detail coefficient matrix LH1, vertical detail coefficient matrix HL1, and diagonal detail coefficient matrix HH1 obtained after the first-level transform are extracted and upsampled in both row and column dimensions. Then, a synthesis filter bank that is biorthogonal to the wavelet transform analysis filter bank is used to convolve the upsampled results in both row and column dimensions. The convolution results are superimposed and merged to obtain the reconstructed matrix M containing only the high-frequency information of matrices M and A. r and A r Next, matrix M... r and A r Linearly map to the interference-free echo matrix R as follows: c Amplitude matrix Mc and phase angle matrix A c Within the range of values, matrix M is obtained. m and A m

[0115]

[0116] Finally, for matrix M m and A m Perform global zero-mean normalization as follows to obtain matrix M. z and A z

[0117]

[0118] in, It is matrix M m average value It is matrix A m The average value. At this point, matrix M... z and A z According to Z=M z ·exp(A z The interference suppression echo matrix is ​​reorganized into a complex matrix in a manner that enhances the details. Figure 4 The process of enhancing the details of the interference suppression echo matrix described above is demonstrated. Figure 5 This demonstrates the key principles underlying the processing flow.

[0119] For a SAR echo frame affected by NBI, CMWBI, and SMWBI, Figure 6 The data flow after preprocessing, including interference suppression and echo detail enhancement, is demonstrated. It can be seen that the method disclosed in this invention can effectively suppress interference and also possesses a certain degree of echo detail preservation capability.

[0120] It should be noted that although this invention provides a weighted curvature-driven method for the time-frequency domain structure-texture decomposition problem, those skilled in the art should understand that improvements or other structure-texture decomposition methods should also be covered within the scope of the claims of this invention. Similarly, although this invention only provides five-layer wavelet transform and corresponding post-processing methods for the interference suppression echo matrix detail enhancement problem, those skilled in the art should understand that improvements or other detail enhancement methods should also be covered within the scope of the claims of this invention.

[0121] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to the embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A time-frequency domain structure-texture decomposition-guided SAR echo suppression interference reduction method, characterized in that, Includes the following steps: Obtain the SAR echo matrix and preprocess the echo matrix in the order of first zero-mean normalization and then energy normalization; Interference detection is performed on the transmitted pulse echoes of each frame of the preprocessed echo matrix to obtain the interfered echo matrix and the interference-free echo matrix. After padding each frame of the interfered echo matrix with zeros at both ends, a short-time Fourier transform is performed to obtain the corresponding time-frequency domain representation. The time-frequency domain representation of each frame of disturbed echo is decomposed into structural components representing the disturbance and texture components representing the pure echo by using a weighted curvature-driven energy functional minimization model. Perform an inverse short-time Fourier transform on the texture components and extract the effective signal segments without abrupt changes based on the previously padded zero positions, and save them into the interference suppression echo matrix.

2. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 1, characterized in that, Obtain the SAR echo matrix, and preprocess the echo matrix according to the order of first zero-mean normalization and then energy normalization, specifically including: Record the echoes of each transmitted pulse of the SAR in a matrix. In the diagram, the echo matrix of the SAR is represented by R, where the echo matrix R is... e each line This represents a frame of transmitted pulse echo; For the echo matrix R e Each frame of transmitted pulse echo is zero-mean processed to obtain a zero-mean matrix. in, i∈[1,M] represents the average value of the transmitted pulse echo in the i-th frame; For zero-mean matrix D c Each row is processed to normalize energy, resulting in an energy normalization matrix. in, i∈[1,M] represents the energy value of the transmitted pulse echo in the i-th frame, and σ represents the set uniform energy value.

3. The SAR echo suppression interference reduction method guided by time-frequency domain structure-texture decomposition according to claim 1, characterized in that, Interference detection is performed on the transmitted pulse echoes of each frame in the preprocessed echo matrix to obtain the interfered echo matrix and the interference-free echo matrix, specifically including: Interference detection is performed on the transmitted pulse echo of each frame in the preprocessed echo matrix; The detected interference echoes are recorded in the first complex matrix to obtain the interference echo matrix; The undisturbed echoes are recorded into the second complex matrix to obtain the undisturbed echo matrix.

4. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 1, characterized in that, After padding each frame of the interfered echo in the interfered echo matrix with zeros at both ends, a short-time Fourier transform is performed to obtain the corresponding time-frequency domain representation, which specifically includes: Each row in the interference echo matrix represents the interference echo of each frame. For the disturbed echo matrix R j Each frame of the interference echo The leading zeros are padded forward, and the... Zeros are padded to the end of the signal to obtain the zero-padded interference echo signal. The number of zeros used for forward and backward padding is the same, and the total number of zeros is taken from the number of interference echoes in each frame. 10% to 50% of the length; The zero-padding interference echo signal Performing a short-time Fourier transform yields the corresponding time-frequency domain representation, STFT0: Where N is the signal The length of the discrete Fourier transform can also be viewed as the total number of points in the discrete Fourier transform. Indicates signal The k-th element, k = 1, 2, ..., N; ω * (km) represents the signal The discrete-time window function, where m is the center index of the window function, m = 1, 2, ..., N, and k represents the area of ​​the signal covered by the window function. The k-th element; n is the discrete frequency index, n = 1, 2, ..., N, and * is the complex conjugate symbol.

5. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 1, characterized in that, The time-frequency domain representation of each frame of disturbed echo is decomposed into structural components representing the disturbance and texture components representing the clean echo using a weighted curvature-driven energy functional minimization model. Specifically, this includes: Establish a weighted curvature-driven energy functional minimization model for the time-frequency domain representation matrix STFT0 of a frame of interfered echo: In this equation, the first integral term is called the linear regularization term, the second integral term is called the weighted curvature term, and the third integral term is called the approximation term; λ is a positive constant, ▽ is the gradient operator, and Δ... -1 It is the inverse Laplace operator; It is a weighted function, where β is a positive constant and G is a negative constant. σ The Gaussian kernel function is represented by the asterisk (*), and the asterisk (*) indicates convolution. h(x) = 1 + t·c(x)·e(x) is the exponential function, where t is a positive constant. It is a deviation function. Sup is the Butterworth low-pass filter result of the horizontal set curvature of matrix STFT0, where sup represents the supremum of the set; K is the Butterworth low-pass filter result of the horizontal set curvature of the structural component U of matrix STFT0; V is the texture component of matrix STFT0. It is a gradient-dependent nonconvex function of matrix STFT0; STFT0 is the weighted level set curvature, div is the divergence; Q is the zero-div vector field divQ=0; The optimal separation of U and V is obtained by solving the weighted curvature-driven energy functional minimization model, where U represents the structural component of the disturbance and V represents the texture component of the pure echo.

6. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 4, characterized in that, The structural components representing interference include one or more of narrowband interference (NBI), frequency modulation broadband interference (CMWBI), and sinusoidal modulation broadband interference (SMWB).

7. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 1, characterized in that, Perform an inverse short-time Fourier transform on the texture components and extract the effective signal segments without abrupt changes based on the previously padded zero positions. Save these segments to the interference suppression echo matrix. Specifically, this includes: Performing an inverse short-time Fourier transform on the composite texture component X yields the transformed texture signal: Where N is the length of the original signal, which can also be regarded as the total number of points of the discrete Fourier transform of the original signal; M is the total number of short-time signal segments into which the original signal is divided; ω(km) represents the additional discrete-time window function after performing the inverse Fourier transform on the spectrum X(m,n) of the m-th short-time signal segment, m=1,2,…,M, k represents the k-th element of the original signal, k=1,2,…,N; n is the discrete frequency index, n=1,2,…,N. Based on the interference echo signal after zeroing at the beginning and end The number of zeros padded at the beginning and end of x(k) is used to truncate the beginning of x(k) backward and the end forward, respectively, and the middle part is retained to obtain the effective signal y(t) without abrupt changes, which is the interference suppression echo. All interference suppression echoes are combined to form the interference suppression echo matrix Y.

8. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 1, characterized in that, It also includes, Five-level wavelet transform is used to enhance the details of the interference suppression echo matrix.

9. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 8, characterized in that, The details of the interference suppression echo matrix are enhanced using a five-level wavelet transform, specifically including: Calculate the amplitude matrix and phase angle matrix of the interference suppression echo matrix Y; Based on wavelet transform, the amplitude matrix M and the phase angle matrix A are reconstructed to obtain the corresponding reconstruction matrix M. r and A r ; Reconstruct matrix M r and A r Linearly map to the interference-free echo matrix R as follows: c Amplitude matrix M c and phase angle matrix A c Within the range of values, matrix M is obtained. m and A m : For matrix M m and A m Perform global zero-mean normalization as follows to obtain matrix M. z and A z in, It is matrix M m average value It is matrix A m The average value; matrix M z and A z According to Z=M z ·exp(A z The interference suppression echo matrix is ​​obtained by recombining the complex matrix in the manner of detail enhancement.

10. The SAR echo suppression interference suppression method guided by time-frequency domain structure-texture decomposition according to claim 8, characterized in that, Based on wavelet transform, the amplitude matrix M and the phase angle matrix A are reconstructed to obtain the corresponding reconstruction matrix M. r and A r Specifically, it includes: Perform a five-level wavelet transform on the amplitude value matrix M, and extract the horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix of the amplitude value matrix M obtained after the first level transform; The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix of the amplitude value matrix M are all upsampled in both row and column dimensions to obtain coefficient matrices of the same size as the amplitude value matrix M. By using a synthesized filter bank that is orthogonal to the forward wavelet transform analysis filter bank, the row-dimensional and column-dimensional convolutions are performed on the upsampled horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix, respectively, to obtain the corresponding inverse filter coefficient matrix; The horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix after inverse filtering are superimposed and merged to obtain a reconstruction matrix that contains only high-frequency information of the amplitude value matrix M. Perform a five-level wavelet transform on the phase angle matrix A, and extract the horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix of the phase angle matrix A obtained after the first level transform; The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix of the phase angle matrix A are all upsampled in both row and column dimensions to obtain coefficient matrices of the same size as the phase angle matrix A. By using a synthesized filter bank that is orthogonal to the forward wavelet transform analysis filter bank, the row-dimensional and column-dimensional convolutions are performed on the upsampled horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix, respectively, to obtain the corresponding inverse filter coefficient matrix; The horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix after inverse filtering are superimposed and merged to obtain a reconstruction matrix that contains only high-frequency information of the phase angle matrix A.

Images