A method for suppressing weak interference in SAR images based on block-based iterative elimination

Abstract

This invention relates to a method for suppressing weak interference in SAR images based on block-based iterative elimination, belonging to the field of radar signal processing. The method includes performing a range-based FFT on the original data to obtain a matrix, calculating the kurtosis K of each column of the matrix, detecting outliers in K using the MAD method to obtain Mask1D, statistically analyzing the connected components of Mask1D, and setting labels to achieve interference block segmentation; varying the ergodic of the interference sub-blocks, setting the average amplitude of the signal portion except for the current sub-block, and obtaining a mask for each detection; merging each mask to obtain a final mask, using the final mask to perform notch filtering on the matrix, and performing an inverse Fourier transform on the notched matrix to obtain the interference-suppressed image. This invention overcomes the problem of difficulty in effectively distinguishing features between different types of signals. Through multiple block processing, it can eliminate the influence of strong interference on the detection results of weak interference, thereby achieving effective suppression of SLC images with interference sub-blocks having large energy differences in the range-frequency domain.

Description

Technical Field

[0001] This invention belongs to the field of radar signal processing, specifically relating to a SAR image weak interference suppression method based on block iterative elimination, which is suitable for suppressing radio frequency interference in synthetic aperture radar single-look complex (SLC) images. Background Technology

[0002] Synthetic Aperture Radar (SAR), with its all-weather, high-resolution characteristics, plays a crucial role in military, civilian, and scientific research fields. However, with the widespread application of modern radio technology, electromagnetic spectrum resources are becoming increasingly scarce, which may increase the likelihood of radio frequency interference (RFI). RFI mainly originates from ground-based radars, meteorological observation systems, and communication equipment operating in the same frequency band. When a SAR system receives RFI, it affects the impulse response of the SAR signal, degrades the quality of the SAR image, and further hinders subsequent image interpretation applications such as target detection and interferometry.

[0003] Currently, the main radio frequency interference (RFI) suppression methods are categorized into parametric, semi-parametric, and non-parametric methods. Parametric methods estimate RFI parameters by establishing an RFI parameterized model, thereby suppressing RFI in echo data. However, their performance in estimating broadband interference parameters is limited. Semi-parametric methods transform the RFI suppression problem into a hyperparameter optimization problem, such as using low-rank sparse features to establish an optimization model for RFI suppression. Compared to parametric methods, semi-parametric methods have the advantage of not requiring accurate prior knowledge of the interference signal and achieving good interference suppression performance with relatively simple models. However, the performance of semi-parametric methods is highly dependent on the selection of hyperparameters and involves significant computational cost. Non-parametric methods effectively suppress interference in the transform domain by utilizing the amplitude characteristic differences between interference and the true echo, with relatively low computational cost. However, methods that can effectively distinguish between interference and echo signals and employ appropriate mapping transformations are not readily available. Most of the above methods process echoes, but echo data is difficult for remote sensing data users to obtain. Therefore, it is necessary to research effective image domain interference suppression methods. Summary of the Invention

[0004] The technical problem to be solved by this invention is:

[0005] Non-parametric methods analyze the characteristic differences between interference and data, amplifying the differences between interference and SAR echoes through a mapping transformation, thereby enabling adaptive filtering of the interference. Currently, non-parametric methods typically perform statistical tests on amplitude characteristics to quantify the characteristic differences between the signal and interference. This method performs well in detecting strong interference with significant amplitude differences from the original echo, but it cannot effectively suppress situations where both strong and weak interference coexist.

[0006] To address the problems existing in non-parametric interference suppression techniques, this invention provides a SAR image weak interference suppression method based on block-based iterative elimination. The interference is coarsely divided into blocks according to its location in the range-frequency domain, and processed separately in descending order of amplitude. During processing, the amplitude of other sub-blocks is set to the average of the signal amplitude. This invention aims to eliminate the influence of the relative energy of different interference blocks on the statistical test results, thereby achieving effective suppression of SLC images with interference exhibiting significant energy differences.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A method for suppressing weak interference in SAR images based on block-based iterative elimination, characterized by the following steps:

[0009] Step 1: Perform a Fourier transform on the synthetic aperture radar (SLC) image along the range dimension to obtain the range-frequency domain-azimuth time domain matrix. X(i,j) represents the element in the i-th row and j-th column of the distance frequency domain-azimuth time domain matrix, where 1≤i≤Na and 1≤j≤Nr, i represents slow time discrete sampling points, j represents fast time sampling points, Na represents the total number of slow time discrete sampling points, and Nr represents the total number of fast time discrete sampling points.

[0010] Step 2: The range-frequency domain-azimuth time domain matrix X is represented by the useful signal matrix S and N interference matrices {I1,I2,...,I...} N The superposition of}:

[0011] X(i,j)=S(i,j)+I1(i,j)+I2(i,j)+...+I N (i,j)(1)

[0012] Where S(i,j),I N (i,j) represent the i-th row and j-th column elements of the useful signal matrix and the N-th interference matrix, respectively;

[0013] Step 3: Calculate the kurtosis of each column of the range-frequency domain-azimuth-time domain matrix X along the azimuth dimension. This quantifies the degree of deviation of the interference signal from the statistical distribution of the real part of the useful signal, i.e., the normal distribution, thereby determining the location of the interference in the range dimension. The kurtosis K of X is represented as {K1, K2, ..., K...} Nr}, where K j The calculation formula is as follows:

[0014]

[0015] Where X(j) represents the j-th column of the range frequency domain-azimuth time domain matrix X;

[0016] Step 4: After smoothing the kurtosis K, detect kurtosis outliers using the median absolute difference method, i.e., the fast-time sampling points corresponding to the interference; calculate the median absolute deviation (MAD) of the kurtosis K obtained in Step 3, defined as:

[0017] MAD = c·median(|K) j -median(K)|) (3)

[0018] Where median(·) is the operation to calculate the median of the observation points, and c is the conversion factor, calculated as follows:

[0019]

[0020] Where erfcinv is the inverse complement error function; the distance d from the center of all observation points to the kurtosis K based on MAD is calculated as follows:

[0021]

[0022] Points where d > Th are selected as the fast-time sampling points corresponding to the final detected interference locations; finally, a one-dimensional interference mask Mask1D is obtained. 1×Nr ∈{0,1} 1×Nr Set the sampling point with interference to 1, and the sampling point without interference to 0.

[0023] Step 5: Count the connected components of Mask1D and assign labels to each component to achieve interference segmentation; where 0 corresponds to the useful signal, and 1, 2, ..., N correspond to the interference strips; let the current sub-block be I. m Then the magnitude of the corresponding sub-blocks is reset to:

[0024] I n (i,j)=mean(X Mask1D＝0 ),n≠m (6)

[0025] Where mean(·) is used to calculate the mean of a two-dimensional matrix, X Mask1D＝0This indicates the column of X corresponding to Mask1D being equal to 0; ultimately, a new range-frequency domain-azimuth time domain matrix X' is obtained;

[0026] Step 6: Design a frequency domain notch filter for X' on I m Two-dimensional interference mask prediction is performed on the interference sub-block; the Z-value of X' is calculated. score The difference between quantization interference and the original echo:

[0027]

[0028] Where std(·) is the operation for calculating the standard deviation of a 2D matrix; and Z is calculated. score The confidence interval is determined by setting the confidence level to obtain I. m Final detection mask m :

[0029]

[0030] Where normcdf(·) is the cumulative standard normal distribution, and α is the confidence level; traversing the interference sub-blocks, the mask tensor obtained after N detections is χ=[Mask1,Mask2,...,Mask... N Perform an OR operation on all sub-masks to obtain the final detection mask M:

[0031]

[0032] Step 7: After notching the original two-dimensional time-frequency domain representation matrix X at position M, perform an inverse Fourier transform to obtain the interference-suppressed SLC image:

[0033]

[0034] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0035] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0036] The beneficial effects of this invention are as follows:

[0037] This invention provides a SAR image weak interference suppression method based on the idea of ​​block iterative elimination. Compared with traditional non-parametric interference suppression methods, this invention can overcome the problem that it is difficult to find effective distinctions between features of different types of signals. Through multiple block processing, the influence of strong interference on the detection results of weak interference can be eliminated, thereby achieving effective suppression of SLC images with interference sub-blocks with large energy differences in the range-frequency domain. Attached Figure Description

[0038] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0039] Figure 1 This is a flowchart of a SAR image weak interference suppression method based on the block iterative elimination idea.

[0040] Figure 2 This is the two-dimensional azimuth time-domain distance-frequency domain matrix corresponding to the SLC image.

[0041] Figure 3 The result image is for kurtosis detection and outlier extraction.

[0042] Figure 4 This is the result of segmenting interference after statistical analysis of connected components.

[0043] Figure 5 shows the interference detection mask and notch filter results in the two-dimensional matrix of azimuth and azimuth in the range-frequency domain: (a) is the mask obtained by final detection, and (b) is the azimuth-time domain and range-frequency domain matrix after interference suppression.

[0044] Figure 6 shows the results of weak interference suppression based on the block-based iterative elimination method: (a) is the SLC image before interference suppression, and (b) is the SLC image after interference suppression. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0046] Example 1:

[0047] Step 1: The synthetic aperture radar (SLC) image is shown in Figure 6(a). A Fourier transform is performed along the range dimension to obtain the range-frequency domain-azimuth time domain matrix. like Figure 2, where X(i,j) represents the element in the i-th row and j-th column of the two-dimensional time-frequency characterization matrix, where 1≤i≤Na and 1≤j≤Nr, i represents slow-time discrete sampling points, j represents fast-time sampling points, Na represents the total number of slow-time discrete sampling points (10000 in this example), and Nr represents the total number of fast-time discrete sampling points (14352 in this example).

[0048] Step 2: The range-frequency domain-azimuth time domain matrix X is represented as the superposition of the useful signal matrix S and N interference matrices {I1,I2,I3}.

[0049] X(i,j)=S(i,j)+I1(i,j)+I2(i,j)+I3(i,j) (11)

[0050] Where S(i,j),I N (i,j) represent the i-th row and j-th column elements of the useful signal matrix and the N-th interference matrix, respectively.

[0051] Step 3: Calculate the kurtosis of each column of the range-frequency domain-azimuth-time domain matrix X along the azimuth dimension. This quantifies the degree of deviation of the interference signal from the statistical distribution (i.e., the normal distribution) of the real part of the useful signal, thereby determining the location of the interference in the range dimension. The kurtosis K of X is represented as {K1, K2, ..., K...} 14352}, where K j The calculation formula is as follows:

[0052]

[0053] Here, is represented by the j-th column of the range-frequency-azimuth time-domain matrix X.

[0054] Step 4: After smoothing the kurtosis K, the median absolute deviation (MAD) method is used to detect kurtosis outliers, i.e., the fast-time sampling points corresponding to the interference. The MAD is calculated based on the kurtosis K obtained in Step 3 and is defined as follows:

[0055] MAD = c·median(|K) j -median(K)|) (3)

[0056] Where median(·) is the operation to calculate the median of the observation points, and c is the conversion factor, calculated as follows:

[0057]

[0058] Where erfcinv is the inverse complement error function. The distance d from the center of all observation points to the kurtosis K based on MAD is calculated as follows:

[0059]

[0060] Points where d > Th are taken as the fast-time sampling points corresponding to the final detected interference locations. The detection threshold Th = 4 is set. The kurtosis and outlier detection results are as follows: Figure 3 The final result is a one-dimensional interference mask, Mask1D. 1×Nr ∈{0,1} 1×Nr Set the sampling point with interference to 1, and the sampling point without interference to 0.

[0061] Step 5: Count the connected components of Mask1D and assign labels to each component to achieve interference block segmentation. Figure 4 Different colored sub-blocks represent different categories. 0 corresponds to the useful signal, and 1, 2, and 3 correspond to interference stripes. Let the sub-block being processed be I. m Then the amplitude of the other sub-blocks is reset to

[0062] I n (i,j)=mean(X Mask1D＝0 ),n≠m (6)

[0063] Where mean(·) is used to calculate the mean of a two-dimensional matrix, X Mask1D＝0 This indicates the column of X where Mask1D equals 0. Ultimately, a new range-frequency domain-azimuth time domain matrix X' is obtained.

[0064] Step 6: Design a frequency domain notch filter for X' on I m Two-dimensional interference mask prediction is performed on the interference sub-block. The Z-value of X' is calculated. score The difference between quantization interference and the original echo:

[0065]

[0066] std(·) is the operation for calculating the standard deviation of a 2D matrix. Calculate Z score The confidence interval is determined by setting the confidence level to obtain I. m Final detection mask m :

[0067]

[0068] Where normcdf(·) is the cumulative standard normal distribution, and α is the confidence level, which is set to 0.15 in this example. Traversing the interference sub-blocks, the mask tensor for N detections is obtained as χ=[Mask1,Mask2,Mask3]. Performing an OR logic operation on all sub-masks yields the final detection mask M as shown in Figure 5(a):

[0069]

[0070] Step 7: Notch the position of the original two-dimensional time-frequency domain characterization matrix X at M as shown in Figure 5(b), and perform inverse Fourier transform on the notch result to obtain the interference-suppressed SLC image as shown in Figure 6(b).

[0071]

[0072] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for weak jammer suppression in SAR images based on block-wise iterative cancellation, characterized in that The steps are as follows: Step 1: Fourier transform the synthetic aperture radar (SAR) SLC image along the range dimension to obtain a range-frequency domain-azimuth-time domain matrix X(i,j) is expressed as the i-th row and j-th column element of the range-frequency domain-azimuth-time domain matrix, wherein 1≤i≤Na and 1≤j≤Nr, i represents a slow-time discrete sampling point, j represents a fast-time sampling point, Na represents the total number of slow-time discrete sampling points, and Nr represents the total number of fast-time discrete sampling points; Step 2: The range-frequency - azimuth-time matrix X is represented as a superposition of the useful signal matrix S and the N block interference matrices {I1, I2,..., IN}: N} X(i,j) = S(i,j) + I1(i,j) + I2(i,j) +... + I N (i,j) (1) where S(i,j), I N (i,j) represent the ith row, jth column element of the useful signal matrix, the Nth block interference matrix, respectively. Step 3: Calculate the kurtosis of each column of the range-dimension frequency-range time matrix X along the azimuth dimension, which quantifies the deviation of the statistical distribution of the interference signal relative to the real part of the useful signal, i.e., the normal distribution, thereby determining the position of the interference in the range dimension; the kurtosis K of X is represented as {K1, K2,..., K Nr}, wherein K j The calculation formula of K is as follows: Wherein, X(j) represents the jth column of the distance frequency domain-azimuth time domain matrix X; Step 4: The kurtosis K is smoothed, and the kurtosis outliers corresponding to the fast time sampling points of the interference are detected by using the absolute median deviation method; the median absolute deviation MAD of the kurtosis K obtained in step 3 is calculated, and is defined as: MAD = c median(|K j -median(K)|) (3) Wherein median(·) is the operation of calculating the median of the observation points, and c is a conversion coefficient, and the calculation formula is as follows: Wherein erfcinv is the inverse complementary error function; the distance value d of all observation points of the kurtosis K from the center based on the MAD is calculated as: The point of d>Th is the fast time sampling point corresponding to the final detected interference position; and a one-dimensional interference mask Mask1D is finally obtained 1×Nr ∈{0,1} 1×Nr The sampling point with interference is set to 1, and otherwise 0. Step 5: Statistics of the connected components of Mask1D, and set labels respectively, to realize interference blocking operation; wherein 0 corresponds to useful signal, 1, 2, …, N correspond to interference strips; let the current processing sub-block be I m Then the amplitude size corresponding to other sub-blocks is reset as: I n (i,j) = mean(X Mask1D＝0 ), n≠m (6) where mean(•) is the mean of a two-dimensional matrix, X Mask1D＝0 represents that Mask1D equals 0 corresponds to the column of X; finally a new range-azimuth matrix X' is obtained; Step 6: Design a frequency-domain notch filter to X' to remove the I m interference sub-blocks from the 2D interference mask prediction; calculate Z score quantify the difference between the interference and the original echo: where std( ) is a 2-dimensional matrix standard deviation operation; calculate Z score confidence interval, set confidence degree decision to get I m final detection mask Mask m : Where normcdf(·) is the cumulative standard normal distribution, and α is the confidence level; traversing the interference sub-blocks, the mask tensor obtained after N detections is χ=[Mask1,Mask2,...,Mask... N Perform an OR operation on all sub-masks to obtain the final detection mask M: Step 7: After the wave of the original two-dimensional time domain-frequency domain representation matrix X at the position of M is suppressed, the SLC image after the interference suppression is obtained by inverse Fourier transform:

2. A computer system, characterized by Comprise: One or more processors, a computer readable storage medium, for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method of claim 1.

3. A computer-readable storage medium, characterized in that Computer executable instructions are stored, which instructions are executed to implement the method of claim 1.

Images