A SAR image interference suppression method based on frequency domain signal recovery

Abstract

The present invention discloses a SAR image interference suppression method based on frequency domain signal recovery, comprising: receiving synthetic aperture radar image data containing radio frequency interference, performing Fourier transform on the image data along the range dimension to convert it into a two-dimensional spectrum in the range frequency domain and the azimuth time domain; calculating the kurtosis of the two-dimensional spectrum and setting an outlier detection threshold based on the kurtosis; performing interference detection on the kurtosis, wherein sampling points greater than the threshold are considered to be interference, and constructing a one-dimensional mask; constructing an effective data selection matrix based on the one-dimensional mask; constructing an amplitude steering matrix; extracting the frequency domain pulses of each row in the two-dimensional spectrum, constructing an amplitude spectrum and an effective data covariance matrix using the frequency domain pulses, the amplitude steering matrix, and the effective data selection matrix; reconstructing the time domain pulses corresponding to the frequency domain pulses; and then the reconstructed time domain pulse combination corresponding to all the frequency domain pulses is the interference suppressed SAR image. The present invention achieves refined and effective suppression of SLC images with interference.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing, and in particular to a SAR image interference suppression method based on frequency domain signal recovery. Background Art

[0002] Spaceborne Synthetic Aperture Radar (SAR) offers all-day, all-weather, and high-resolution Earth observation capabilities. However, radio frequency interference (RFI) received by spaceborne SAR is becoming increasingly severe, degrading SAR image quality and impacting subsequent image interpretation. Frequency-domain notching, which exploits the unidirectional nature of RFI and its high energy, and sets an energy threshold for interference suppression, is an efficient and robust RFI mitigation method and is a popular application in engineering practice.

[0003] The lack of prior information in energy threshold selection can easily lead to residual interference or severe loss of useful signals. Partially zeroing the frequency domain signal increases the sidelobes at strong scattering points in the image, significantly reducing image quality at these points. Currently, the main method for missing signal recovery is the Missing-Data Iterative Adaptive Approach (MIAA). This method primarily targets time-domain signal recovery, but due to the large amount of computation required and the need for multiple iterations, it is difficult to achieve efficient interference suppression given the massive amount of data from spaceborne SAR. Summary of the Invention

[0004] The purpose of the present invention is to provide a SAR image interference suppression method based on frequency domain signal recovery, aiming to eliminate the sidelobe lifting phenomenon of strong scattering points after frequency domain notching, thereby achieving effective and refined suppression of SLC images with interference.

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

[0006] A SAR image interference suppression method based on frequency domain signal recovery, comprising:

[0007] Receive synthetic aperture radar image data containing radio frequency interference, perform Fourier transform on the image data along the range dimension to obtain a two-dimensional spectrum in the range frequency domain and the azimuth time domain;

[0008] Calculating the kurtosis of the two-dimensional spectrum and setting an outlier detection threshold based on the kurtosis; performing interference detection on the kurtosis, wherein sampling points greater than the threshold are considered to be interference, and constructing a one-dimensional mask;

[0009] constructing a valid data selection matrix based on the one-dimensional mask;

[0010] Construct an amplitude-steering matrix;

[0011] Extract the frequency domain pulse of each row in the two-dimensional spectrum, and use the frequency domain pulse, amplitude steering matrix, and effective data selection matrix to construct the amplitude spectrum and effective data covariance matrix:

[0012] Based on the spectrum amplitude and the effective data covariance matrix, the time domain pulses corresponding to the frequency domain pulses are reconstructed; then the reconstructed time domain pulse combination corresponding to all frequency domain pulses is the SAR image after interference suppression.

[0013] Furthermore, calculating the kurtosis of the two-dimensional spectrum and setting the outlier detection threshold based on the kurtosis include:

[0014] For the two-dimensional spectrum X(f,η), calculate the kurtosis K of X(f,η) along the distance direction:

[0015]

[0016] Among them, X(f) is the vector of X(f,η) column-wise mean, is the mean of X(f), is the variance of X(f), x i ∈X(f), n is the length of X(f); f represents the range frequency, η represents the azimuth slow time, and E{·} represents the expectation;

[0017] The threshold Thr is expressed as:

[0018] Thr=μ K -3σ X (2)

[0019] in, is the mean of the kurtosis K, m is the length of the kurtosis K.

[0020] Furthermore, the construction of a one-dimensional mask is expressed as:

[0021] Interference detection is performed on the kurtosis K, where the sampling point x greater than the threshold Thr i It is considered to be interference; the interference one-dimensional mask is Mask1D, the sampling point where the interference is located is set to 1, and the other positions are 0. Mask1D is expressed as:

[0022]

[0023] Furthermore, constructing a valid data selection matrix based on the one-dimensional mask includes:

[0024] Convert the one-dimensional mask Mask1D to a size of N g ×N effective data selection matrix S g ; where Ng is the number of Mask1D values ​​0, and N is the number of pulse sampling points;

[0025] Effective data selection matrix S g Expressed as:

[0026]

[0027] Wherein, Mask1D[k] represents the k-th element in the one-dimensional mask Mask1D, {k|Mask1D[k]=0} represents the vector composed of the positions where all elements in Mask1D are 0, and {i} represents the i-th value of the vector.

[0028] Furthermore, the amplitude steering matrix is ​​constructed as follows:

[0029] A H =[α(ω1) α(ω2) ... α(ω K )] (6)

[0030]

[0031] in,(·) H represents the conjugate transpose, ω k =(2π / K)k, k=1,2,...,K, ωk is the created grid frequency, K is the number of amplitude spectrum resolution points, j is the imaginary unit, and e is a natural constant.

[0032] Furthermore, the frequency domain pulse, amplitude steering matrix and effective data selection matrix are used to construct the amplitude spectrum and effective data covariance matrix, which can be expressed as:

[0033] P=(A H y·(A H y) H ) / N (8)

[0034] R g =A g H PA g (9)

[0035] Among them, A g =S g A is a valid data-oriented matrix.

[0036] Furthermore, the time domain pulse X after frequency domain recovery is expressed as:

[0037]

[0038] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, the SAR image interference suppression method based on frequency domain signal recovery is implemented.

[0039] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the SAR image interference suppression method based on frequency domain signal recovery is implemented.

[0040] Compared with the prior art, the present invention has the following technical features:

[0041] The method of the present invention can achieve refined interference suppression of SAR level 1 images containing radio frequency interference; after suppressing the radio frequency interference component using kurtosis and outlier detection in the range frequency domain, the useful signal frequency domain component is restored, effectively avoiding the problem of enhanced sidelobes of strong scattering points in the image caused by the traditional frequency domain notch method. The image after frequency domain signal recovery obtains a significant improvement in the indicators of equivalent view number and image entropy, and the image quality is greatly improved, which plays an important role in subsequent applications such as SAR image interpretation and target detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the process of the present invention;

[0043] Figure 2 is an SLC image with interference in an embodiment of the present invention;

[0044] Figure 3 A two-dimensional range frequency domain-azimuth time domain diagram in an embodiment of the present invention;

[0045] Figure 4 This is a graph showing the frequency domain kurtosis detection result in an embodiment of the present invention;

[0046] Figure 5 is a one-dimensional interference mask in an embodiment of the present invention;

[0047] Figure 6 A two-dimensional range frequency domain-azimuth time domain diagram after interference suppression and signal recovery in an embodiment of the present invention;

[0048] Figure 7 This is an SLC image after interference suppression and signal recovery in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention provides a method for suppressing interference in SAR images based on frequency-domain signal recovery. Based on the unidirectional propagation and high energy characteristics of radio frequency interference, KL divergence and outlier detection are used to filter interference pulses in the range-frequency domain. The MIAA algorithm is improved using a missing signal model, and the sparsity of SAR images in the time domain is exploited to achieve frequency-domain signal recovery with zeroing in the frequency domain. The specific steps of the present method are as follows:

[0050] Step 1: Receive synthetic aperture radar (Single Look Complex, SLC) image data X(t,η) containing radio frequency interference, where t and η represent the fast time of range and the slow time of azimuth, respectively; perform Fourier transform on the image data X(t,η) along the range dimension to obtain a two-dimensional spectrum X(f,η) in the range frequency domain and the azimuth time domain, where f represents the range frequency.

[0051] Step 2: Calculate the kurtosis K for the two-dimensional spectrum X(f,η), set the outlier detection threshold based on the kurtosis K; perform interference detection on the kurtosis K, wherein sampling points greater than the threshold are considered to be interference, and construct a one-dimensional mask Mask1D.

[0052] For the two-dimensional spectrum X(f,η) obtained in step 1, calculate the kurtosis K of X(f,η) along the distance direction:

[0053]

[0054] Among them, X(f) is the vector of X(f,η) column-wise mean, is the mean of X(f), is the variance of X(f), x i ∈X(f), n is the length of X(f).

[0055] Kurtosis is a statistical characteristic. Interference signals often cause the distribution of received signals to deviate from the normal distribution. Outlier detection is performed on the kurtosis K to locate the interference location, obtaining the sampling point location of the interference in the range frequency domain. Outlier detection is performed using a threshold. The threshold Thr can be expressed as:

[0056] Thr=μ K -3σ X (2)

[0057] in, is the mean of the kurtosis K, m is the length of the kurtosis K.

[0058] Interference detection is performed on the kurtosis K, where sampling points greater than the threshold are considered to be interference; the interference one-dimensional mask is Mask1D, where the sampling points where the interference occurs are set to 1 and the rest are set to 0. Mask1D can be expressed as:

[0059]

[0060] Step 3: Construct a valid data selection matrix S based on the one-dimensional mask g .

[0061] Convert the interference one-dimensional mask Mask1D obtained in step 2 to a size of N g ×N effective data selection matrix S g ; where N g is the number of Mask1D values ​​0, and N is the number of pulse sampling points.

[0062] Effective data selection matrix S g It can be expressed as:

[0063]

[0064] Wherein, Mask1D[k] represents the k-th element in the one-dimensional mask Mask1D, {k|Mask1D[k]=0} represents the vector composed of the positions where all elements in Mask1D are 0, and {i} represents the i-th value of the vector.

[0065] Step 4: Construct the amplitude steering matrix A.

[0066] A H =[α(ω1) α(ω2) ... α(ω K )] (6)

[0067]

[0068] in,(·) H represents the conjugate transpose, ω k =(2π / K)k, k=1,2,...,K, ωk is the grid frequency created, K is the number of amplitude spectrum resolution points, j is the imaginary unit, and e is a natural constant. H is the frequency grid created, and ω k This is the frequency corresponding to the grid.

[0069] Step 5: Extract the frequency domain pulse y of each row in the two-dimensional spectrum X(f,η), and use the frequency domain pulse y, the steering matrix A and the effective data selection matrix S g Construct the amplitude spectrum P and the effective data covariance matrix R g :

[0070] P=(A H y·(A H y) H ) / N (8)

[0071] R g =A gH PA g (9)

[0072] Among them, A g =S g A is a valid data-oriented matrix.

[0073] Step 6: Based on the spectrum amplitude P and the effective data covariance matrix R g , reconstruct the time domain pulse X corresponding to the frequency domain pulse y; then the reconstructed time domain pulse X combination corresponding to all frequency domain pulses is the SAR image X after interference suppression m (t,η).

[0074] Among them, starting from step 4, the processing is performed pulse by pulse. After each frequency domain pulse y is processed through steps 4 to 6, the time domain pulse X after frequency domain recovery can be obtained:

[0075]

[0076] Example:

[0077] Step 1: Figure 1 FIG2 is a flow chart of a method for frequency domain outlier detection and recovery of SAR image interference suppression according to an embodiment of the present invention. Receive the measured SLC image data X(t,η) of SAR containing radio frequency interference, as shown in FIG2. Figure 2 As shown, the size of the SAR measured SLC image selected in this embodiment is 2252×2556, t and η represent the fast time of range and the slow time of azimuth respectively, and Fourier transform is performed along the range dimension to obtain the range spectrum and the two-dimensional spectrum X(f,η) of the range frequency domain and the azimuth time domain, where f represents the range frequency domain, as shown in FIG. Figure 3 shown.

[0078] Step 2: For the two-dimensional spectrum X(f,η) obtained in step 1, calculate the kurtosis K of X(f,η) along the distance direction.

[0079]

[0080] Among them, X(f) is the vector of X(f,η) column-wise mean, is the mean of X(f), is the variance of X(f), x i ∈X(f), n is the length of X(f), and in this embodiment, n is 2252. Kurtosis is a statistical feature. Interference signals usually cause the distribution of received signals to deviate from the normal distribution, such as Figure 4 As shown. The kurtosis K is used to detect the location of the interference and obtain the sampling point location of the interference in the range frequency domain. The threshold is used for outlier detection. The threshold Thr can be expressed as:

[0081] Thr=μ X-3σ X (2)

[0082] in, is the mean of the kurtosis K, m is the length of the kurtosis K, which is 2556 in this embodiment. The interference one-dimensional mask is Mask1D, where the interference sampling point is 1 and the rest of the positions are 0, as shown in the following example: Figure 5 As shown, Mask1D can be expressed as:

[0083]

[0084] Step 3: Convert the interference one-dimensional mask Mask1D obtained in step 2 to a size of N g ×N effective data selection matrix S g and size N m ×N missing data selection matrix S m , N g The number of Mask1D values ​​​​being 0, in this embodiment N g =2413. Valid data selection matrix S g It can be expressed as

[0085]

[0086] Wherein, Mask1D[k] represents the k-th element in the one-dimensional mask Mask1D, {k|Mask1D[k]=0} represents the vector composed of the positions where all elements in Mask1D are 0, and {i} represents the i-th value of the vector.

[0087] Step 4: Process pulse by pulse along the azimuth direction to convert the frequency domain pulse y into interference-free data y g =S g y. And construct the magnitude steering vector α(ω K ) and the steering matrix A, the calculation expression is:

[0088] A H =[α(ω1) α(ω2) ... α(ω K )] (6)

[0089]

[0090] in,(·) H represents the conjugate transpose, ω k =(2π / K)k, k=1,2,...,K, ωk is the created grid frequency, K is the number of amplitude spectrum resolution points, N is the number of pulse sampling points, and in this embodiment, K=2556. The higher the value of K, the more accurate the signal recovery, but it will increase the algorithm running time.

[0091] Step 5: Construct the amplitude spectrum P and the effective data covariance matrix R according to the frequency domain pulse y and the steering matrix A processed pulse by pulse in step 4. g

[0092] P=(A H y·(A H y) H ) / N (8)

[0093] R g =A g H PA g (9)

[0094] Among them, A g =S g A is the effective data-oriented matrix, and N is the number of pulse sampling points.

[0095] Step 6: Use the spectrum amplitude P and effective data covariance R obtained in step 5 g , the time domain pulse X corresponding to the reconstructed data:

[0096]

[0097] Starting from step 4, pulse by pulse is processed. After each pulse frequency domain y is processed through steps 4 to 6, the time domain pulse X after frequency domain recovery can be obtained, and finally the SAR image X after interference suppression can be obtained. m (t,η). In this embodiment, the two-dimensional range frequency domain-azimuth time domain after interference suppression and signal recovery is as follows: Figure 6 As shown, the SAR image after interference suppression and signal recovery is as follows Figure 7 shown.

[0098] After frequency domain signal recovery, the strong scattering point slices achieve significant improvements in both equivalent view count and image entropy. As shown in Table 1, this substantial improvement in image quality plays a crucial role in subsequent applications such as SAR image interpretation and target detection. In this example, the original interfered SAR image has an equivalent view count of 0.3285 and an image entropy of 0.0077. After interference suppression and signal recovery, the equivalent view count of the SAR image is 0.1725 and the image entropy is 0.0930.

[0099] Table 1 SAR image evaluation indicators before and after interference suppression

[0100] Image quality assessment Equivalent visual number Image entropy Original interference SAR image 0.3285 0.0077 SAR image after interference suppression 0.1725 0.0930

[0101] 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 SAR image interference suppression method based on frequency domain signal recovery, characterized in that: include: Receive synthetic aperture radar image data containing radio frequency interference, perform Fourier transform on the image data along the range dimension to obtain a two-dimensional spectrum in the range frequency domain and the azimuth time domain; calculating a kurtosis for the two-dimensional spectrum, and setting an outlier detection threshold based on the kurtosis; Interference detection is performed on the kurtosis, where sampling points greater than the threshold are considered to be interference, and a one-dimensional mask is constructed; Constructing a valid data selection matrix based on the one-dimensional mask, comprising: The one-dimensional mask Convert to size The effective data selection matrix ;in for The number of values ​​0, is the number of pulse sampling points; Valid data selection matrix Expressed as: ； in, Represents a one-dimensional mask Middle k elements, express The vector consisting of all positions where the elements in are 0, The vector i values; Construct the amplitude steering matrix, expressed as: ； ； in, represents the conjugate transpose, , is the frequency of the created grid, is the number of amplitude spectrum resolution points, j is the imaginary unit, e is a natural constant; Extract the frequency domain pulse of each row in the two-dimensional spectrum, and use the frequency domain pulse, amplitude steering matrix, and effective data selection matrix to construct the amplitude spectrum and effective data covariance matrix: Based on the spectrum amplitude and the effective data covariance matrix, the time domain pulses corresponding to the frequency domain pulses are reconstructed; then the reconstructed time domain pulse combination corresponding to all frequency domain pulses is the SAR image after interference suppression.

2. The SAR image interference suppression method based on frequency domain signal recovery according to claim 1, characterized in that: The calculating the kurtosis of the two-dimensional spectrum and setting the outlier detection threshold based on the kurtosis includes: For two-dimensional spectrum , calculated along the distance Kurtosis : ； in, for A vector of column-wise means, for The mean of for The variance of , for length; represents the distance frequency, Represents the slow time of direction, Expressing hope; Threshold Expressed as: ； in, Kurtosis The mean of , , Kurtosis length.

3. The SAR image interference suppression method based on frequency domain signal recovery according to claim 1, characterized in that: The construction of a one-dimensional mask is expressed as: Kurtosis K Interference detection is performed, where the value is greater than the threshold Sampling points It is considered as interference; the interference one-dimensional mask is , the sampling point where the interference occurs is set to 1, and the rest of the locations are 0, Expressed as: 。 4. The SAR image interference suppression method based on frequency domain signal recovery according to claim 1, characterized in that: Using frequency domain pulse , amplitude-steering matrix And the effective data selection matrix Constructing the amplitude spectrum and the effective data covariance matrix , expressed as: ； ； in, is a valid data-oriented matrix.

5. The SAR image interference suppression method based on frequency domain signal recovery according to claim 4, characterized in that: Time domain pulse after frequency domain recovery Expressed as: 。 6. A terminal device comprising a processor, a memory, and a computer program stored in the memory; characterized in that: When the processor executes the computer program, the SAR image interference suppression method based on frequency domain signal recovery according to any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium storing a computer program; wherein: When the computer program is executed by a processor, the SAR image interference suppression method based on frequency domain signal recovery according to any one of claims 1 to 5 is implemented.

Images