A SAR image radio frequency interference suppression method based on spectrum outlier detection

Abstract

The present application relates to a kind of SAR image radio frequency interference suppression method based on spectrum outlier detection, it is related to signal processing field.KL divergence is used to measure the difference degree of power spectrum along the direction of orientation and Rayleigh distribution fitting, and interference is detected by outlier, and finally through adaptive iteration and filter, the interference suppression of SAR image in range frequency domain is realized.Compared with the traditional notch filter interference suppression method, the present application can quickly identify and suppress some weak energy interference, and the useful signal loss in SLC image after interference suppression is less, the precision of interference suppression is higher, and the image quality after interference suppression can be greatly improved.

Description

Technical Field

[0001] This invention relates to the field of signal processing, and in particular to a method for suppressing SAR radio frequency interference under frequency domain outlier detection, which is applicable to the suppression of interference with clustering characteristics in the range frequency domain. Background Technology

[0002] Synthetic Aperture Radar (SAR), as an all-weather, all-day imaging radar, is capable of acquiring high-resolution scene images and is an important reconnaissance tool. However, with the increasing complexity of the space electromagnetic environment, services operating in the same frequency band often cause radio frequency interference (RFI) to SAR. RFI distorts the amplitude, phase, frequency, and polarization characteristics of the echo signal, causing SAR images to become defocused, blurred, or exhibit artifacts. How to effectively suppress interference and improve SAR image quality is a research hotspot in the field of radar signal processing.

[0003] Interference can be categorized into narrow-band interference (NBI) and wide-band interference (WBI) based on its bandwidth. NBI has a bandwidth less than 1% of the useful signal bandwidth, and radar echo modeling is relatively simple. Therefore, effective interference suppression can be achieved through time-domain or frequency-domain analysis. Traditional interference suppression methods, such as notch filters, set nulls in strong narrow-band interference regions exceeding a set spectral or image energy threshold. While notch filters are relatively simple to implement, in complex real-world electromagnetic environments, various types of interference are received. When the energy of the interference signal is similar to that of the useful signal or is very weak, it is difficult to set the threshold for the notch filter, making it difficult to effectively suppress the interference. Such methods have low robustness in suppressing interference and are prone to causing significant missed detections and signal loss. Summary of the Invention

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

[0005] To overcome the shortcomings of existing interference suppression techniques, this invention provides a SAR interference suppression method under spectrum outlier detection.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A method for suppressing radio frequency interference in SAR images based on spectrum outlier detection, characterized by comprising:

[0008] The SAR SLC image data is subjected to windowing processing and converted into a two-dimensional spectrum in the range frequency domain and azimuth time domain.

[0009] The power spectrum of the two-dimensional spectrum of SLC image is fitted with Rayleigh distribution to obtain KL divergence, which is then used to detect outliers to obtain a one-dimensional mask for localization interference.

[0010] The signal-to-noise ratio (SNR) is calculated by classifying the spectrum along the range direction using a one-dimensional interference mask, and an adaptive iterative threshold is set based on the SNR.

[0011] Based on the set adaptive iteration threshold, a frequency domain notch filter is used to predict two-dimensional interference masking based on the two-dimensional spectrum of the SLC image according to the number of iterations.

[0012] Frequency domain notch filtering is performed based on the mask, interference suppression is completed, windowing is applied, and then the image is converted to the time domain to obtain the final image.

[0013] A further technical solution of the present invention: the windowing process performed on the SAR SLC image data to convert it into a two-dimensional spectrum in the range frequency domain and azimuth time domain is as follows:

[0014] Receive SAR single-look complex image data, read the parameters according to the corresponding parameter file, and perform windowing processing within the corresponding frequency band as required. Use Fast Fourier Transform to obtain the range spectrum and two-dimensional spectrum in the range frequency domain and azimuth time domain of the SLC image data.

[0015] A further technical solution of the present invention: The power spectrum of the two-dimensional spectrum of the SLC image is fitted with a Rayleigh distribution to obtain the KL divergence, and the outlier value is detected using this divergence to obtain a one-dimensional mask for locating interference. Specifically:

[0016] The power spectrum range mean is calculated using the two-dimensional spectrum of SLC imagery, and the power spectrum range mean is fitted to a Rayleigh distribution.

[0017] In the frequency band, the difference between the power spectrum and the fitted Rayleigh distribution is measured by calculating the KL divergence along the azimuth direction. The smoothed KL divergence is used to detect outliers to locate interference, resulting in a one-dimensional interference mask Mask1D.

[0018] A further technical solution of the present invention includes: in order to prevent abnormal values ​​of DC component contained in the interference mask, the intersection of the two-dimensional spectrum distance-direction intermediate region and the mask Mask1D is taken. If the entire distance-direction intermediate region is in Mask1D, then such abnormal values ​​of DC component are removed from Mask1D.

[0019] A further technical solution of the present invention: the one-dimensional interference mask calculates the signal-to-noise ratio (SNR) by classifying the spectrum along the range direction, and sets an adaptive iterative threshold based on the SNR, specifically as follows:

[0020] The two-dimensional spectrum is divided into interference spectrum and signal spectrum along the distance direction according to Mask1D. The interference energy N and useful signal energy S are calculated by taking the modulus of each spectrum. The maximum value of the interference energy and the average value of the useful signal energy are taken to obtain the signal-to-noise ratio SNR.

[0021] Set the number of adaptive filtering iterations k, and when the difference in signal-to-noise ratio before and after filtering is |SNR k -SNR k-1 The iteration ends when the value is less than the set iteration threshold ε.

[0022] A further technical solution of the present invention: The step of using a frequency domain notch filter to predict two-dimensional interference masks based on the two-dimensional spectrum of the SLC image according to the set adaptive iteration threshold and the number of iterations is specifically as follows:

[0023] By setting up block processing range frequency points, the average value of each range frequency point within the block is taken along the azimuth dimension; RFI forms peaks in the range spectrum, and background clutter whose power deviates significantly from the Gaussian distribution will be detected by the filter; the frequency domain filter extracts the feature of whether it deviates from the Gaussian distribution without calculating the DC component, and calculates the Z score for the frequency points within the block;

[0024] The confidence level v for whether the frequency point deviates from the Gaussian distribution is calculated using the Z-score and compared with the set filter threshold to determine whether the frequency point contains interference.

[0025] The design incorporates an adaptive number of iterations. After multiple iterations, a mask for each block is obtained. The masks are then aggregated to obtain the final mask and filled.

[0026] A further technical solution of the present invention: the step of performing frequency domain notch filtering based on the mask, completing interference suppression, windowing processing, and converting to the time domain to obtain the final image, specifically involves:

[0027] Frequency domain notch filtering is applied to the final mask, and the positions where the interference mask is 1 are set to zero in the spectrum. The two-dimensional spectrum after interference suppression is windowed according to the parameter requirements, and then transformed into the time domain through inverse Fourier transform to obtain the final interference-suppressed SAR image.

[0028] 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.

[0029] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0030] The beneficial effects of this invention are as follows:

[0031] This invention provides a SAR image radio frequency interference suppression method based on spectral outlier detection. It utilizes KL divergence to measure the difference between the power spectrum along the azimuth direction and the Rayleigh distribution, detects interference by identifying outliers, and finally achieves interference suppression in the range-frequency domain of SAR images through adaptive iteration and filters. Compared to traditional notch filtering methods, this invention can quickly identify and suppress some weak-energy interference, and the loss of useful signals in the SLC image after interference suppression is less, resulting in higher accuracy and significantly improved image quality. Attached Figure Description

[0032] 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.

[0033] Figure 1 This is a flowchart of the SAR interference suppression method under spectrum outlier detection.

[0034] Figure 2 This is an example of an L-band SAR satellite image before interference suppression.

[0035] Figure 3 Here is an example of the range spectrum of SAR measured SLC image data, where Figure 3 (a) is the distance-dimensional frequency spectrum, Figure 3 (b) is a two-dimensional spectrum diagram in the range frequency domain and azimuth time domain.

[0036] Figure 4 The KL divergence is calculated for the spectrum, where Figure 4 (a) represents the KL divergence, Figure 4 (b) is the KL divergence after smoothing.

[0037] Figure 5 This is the final interference mask obtained using the method of the present invention.

[0038] Figure 6 The method of this invention is used to analyze the spectrum after masking and notching.

[0039] Figure 7 This is a SAR image after interference suppression using the method of the present invention. Detailed Implementation

[0040] 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.

[0041] This invention provides a SAR image radio frequency interference suppression method based on spectral outlier detection. It calculates the KL divergence as a frequency domain outlier detection operator to extract interference. The absolute difference in signal-to-noise ratio is used as a limitation on the number of iterations to prevent the filter from detecting too much or too little interference, achieving separation of the interference signal from the useful signal within multiple pulses. Due to its flexible and effective detection mechanism, compared with traditional frequency domain and time-frequency domain notch filtering methods, this invention can effectively preserve the useful signal while suppressing interference, significantly improving the image quality after interference suppression.

[0042] Figure 1 The diagram shows the flowchart of the SAR interference suppression method under spectrum outlier detection of the present invention, which includes the following steps:

[0043] Step 1: Receive SAR Single Look Complex (SLC) image data, read the parameters according to the corresponding parameter file, and perform windowing processing within the corresponding frequency band as required. Use Fast Fourier Transform (FFT) to obtain the range spectrum and two-dimensional spectrum of the SLC image data in the range frequency domain and azimuth time domain.

[0044] Step 2: Using the two-dimensional spectrum of the SLC image obtained in Step 1, calculate the range mean of the power spectrum, and fit the range mean of the power spectrum to a Rayleigh distribution. The probability density of the Rayleigh distribution is expressed as:

[0045]

[0046] Where x is the value of the power spectrum upwards, and σ is the standard deviation.

[0047] In this study, the difference between the power spectrum and the fitted Rayleigh distribution is measured along the azimuth direction using KL divergence in the frequency band. The smoothed KL divergence is then used to detect outliers and locate interference, resulting in a one-dimensional interference mask, Mask1D. KL divergence D... KL The calculation is expressed as:

[0048]

[0049] Where p is the azimuth power spectrum and q is the result of fitting the range power spectrum according to the Rayleigh distribution.

[0050] To prevent the interference mask from containing outliers with DC components, the intersection of the two-dimensional spectrum's mid-range region and the mask Mask1D is taken. If the entire mid-range region is within Mask1D, then such outliers with DC components are removed from Mask1D.

[0051] Step 3: Based on the one-dimensional interference mask Mask1D obtained in Step 2, the two-dimensional spectrum is divided into interference spectrum and signal spectrum along the distance direction according to Mask1D. The interference energy N and useful signal energy S are calculated by taking the modulus of each spectrum. The maximum value of the interference energy is taken, and the average value of the useful signal energy is taken to obtain the signal-to-noise ratio SNR. SNR can be expressed as:

[0052]

[0053] Set the number of adaptive filtering iterations k, and when the difference in signal-to-noise ratio before and after filtering is |SNR k -SNR k-1 The iteration ends when the value is less than the set iteration threshold ε.

[0054] Step 4: The frequency domain notch filter performs two-dimensional interference mask prediction based on the number of iterations. By setting up block processing range frequencies, the average value of each range frequency within the block is taken along the azimuth dimension. The RFI forms peaks in the range spectrum; when background clutter with power significantly deviates from the Gaussian distribution, it will be detected by the filter. Without calculating the DC component, the frequency domain filter extracts the feature of whether it deviates from the Gaussian distribution and calculates the Z-score for the frequency points within the block. The Z-score is expressed as:

[0055]

[0056] Where data represents the in-band data, μ is its mean, and ζ is its unbiased estimate.

[0057] The confidence level v for whether a frequency point deviates from a Gaussian distribution is calculated using Z, and compared with the set filter threshold to determine whether the frequency point contains interference. The confidence level v is expressed as:

[0058]

[0059] Based on the iteration number requirement obtained in step 3, after multiple iterations, the mask for each block is obtained, and after summarizing, the final mask is obtained and then filled.

[0060] Step 5: Perform frequency domain notch filtering based on the final mask obtained in Step 4. Set the positions where the interference mask is 1 to zero in the spectrum. Window the two-dimensional spectrum after interference suppression according to the parameters and transform it to the time domain through inverse Fourier transform to obtain the final interference-suppressed SAR image.

[0061] Example 1:

[0062] Step 1: Receive measured SLC image data from L-band SAR; the imaging results are as follows. Figure 2As shown, based on the corresponding parameter file, the parameters are read and windowing processing within the appropriate frequency band is performed as required. The in-band truncation interval is [chop+1, Nr-chop], where Nr is the range length. B w For bandwidth, F s The sampling frequency.

[0063] The distance spectrum of SLC image data is obtained using Fast Fourier Transform (FFT) as follows: Figure 3 As shown in (a), the two-dimensional spectrum in the range frequency domain and the orientation time domain is as follows: Figure 3 As shown in (b);

[0064] Step 2: Using the two-dimensional spectrum of the SLC image obtained in Step 1, calculate the range mean of the power spectrum, and fit the range mean of the power spectrum to a Rayleigh distribution. The probability density of the Rayleigh distribution is expressed as:

[0065]

[0066] Where x is the value of the power spectrum upwards, and σ is the standard deviation.

[0067] In the frequency band, the difference between the power spectrum and the fitted Rayleigh distribution is measured by calculating the KL divergence along the azimuth direction. Figure 4 As shown in (a). KL divergence D KL The calculation is expressed as:

[0068]

[0069] Where p is the power spectrum in the azimuth direction, and q is the power spectrum in the range direction fitted by Rayleigh distribution.

[0070] KL divergence measures the similarity between two functions. In this embodiment, a Gaussian weighted moving average filter with a window length of 100 is used to smooth the KL divergence. The smoothed KL divergence is then used to detect outliers to locate interference. Figure 4 As shown in (b), a one-dimensional interference mask is obtained. To prevent the interference mask from containing outliers with DC components, the intersection of the two-dimensional spectrum's mid-range region and the mask Mask1D is taken. If the entire mid-range region is within Mask1D, then such DC component outliers are removed from Mask1D. In this embodiment, the 11 frequency points in the middle of the flipped two-dimensional spectrum are used as DC components for detection. The outliers are checked to see if they contain DC component outliers. If the length of the included frequency point is less than 10, it is removed from the outliers.

[0071] Step 3: Based on the one-dimensional interference mask Mask1D obtained in Step 2, the two-dimensional spectrum is divided into interference spectrum and signal spectrum along the distance direction according to Mask1D. The interference energy N and useful signal energy S are calculated by taking the modulus of each spectrum. The maximum value of the interference energy is taken, and the average value of the useful signal energy is taken to obtain the signal-to-noise ratio SNR. SNR can be expressed as:

[0072]

[0073] Set the number of adaptive filtering iterations k, and when the difference in signal-to-noise ratio before and after filtering is |SNR k -SNR k-1 When the value is less than the set iteration threshold ε, the iteration ends. In this embodiment, the threshold is set to 0.1.

[0074] Step 4: The frequency domain notch filter predicts the two-dimensional interference mask based on the number of iterations. It processes range frequencies in blocks, averaging the values ​​of each range frequency within the block along the azimuth dimension. In this embodiment, the block size is set to 16×N. r , where N r For the range direction, the RFI forms a peak in the range spectrum. When the background clutter power deviates significantly from the Gaussian distribution, it will be detected by the filter. The frequency domain filter extracts the feature of whether it deviates from the Gaussian distribution without calculating the DC component, and calculates the Z-score for the frequency points within the block. The Z-score is expressed as:

[0075]

[0076] Where data represents the in-band data, μ is its mean, and ζ is its unbiased estimate.

[0077] The confidence level v for whether a frequency point deviates from a Gaussian distribution is calculated using Z, and compared with a set filter threshold to determine whether interference is present. In this embodiment, the filter threshold is set to 0.075. The confidence level v is calculated as follows:

[0078]

[0079] Based on the iteration count requirement obtained in step 3, after multiple iterations, the mask for each block is obtained. These are then aggregated to obtain the final mask, which is then filled. Figure 5 As shown;

[0080] Step 5: Perform frequency domain notch filtering based on the final mask obtained in Step 4, setting the positions where the interference mask is 1 to zero in the frequency spectrum, such as... Figure 6 The image shows the two-dimensional spectrum after interference suppression. The spectrum after interference suppression requires windowing based on the parameters, and is then transformed into the time domain using an inverse Fourier transform to obtain the final interference-suppressed echo data, as shown below. Figure 7 The image shown is the one after interference suppression.

[0081] Comparing the images and two-dimensional spectra before and after interference suppression, it can be found that the SAR radio frequency interference suppression method under spectrum outlier detection results in less loss of useful echo signal after interference suppression, higher accuracy of interference suppression, and can significantly improve the image quality after interference suppression.

[0082] 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 suppressing radio frequency interference in SAR images based on spectrum outlier detection, characterized in that... include: S1: Perform windowing processing on the SAR SLC image data to convert it into a two-dimensional spectrum in the range frequency domain and azimuth time domain; S2: The power spectrum of the two-dimensional spectrum of the SLC image is fitted with a Rayleigh distribution to obtain the KL divergence, which is then used to detect outliers to obtain a one-dimensional mask for localization interference; specifically: The power spectrum range mean is calculated using the two-dimensional spectrum of SLC imagery, and the power spectrum range mean is fitted to a Rayleigh distribution. In this study, the difference between the power spectrum and the fitted Rayleigh distribution is measured along the azimuth direction using KL divergence in the frequency band. The smoothed KL divergence is then used to detect outliers and locate interference, resulting in a one-dimensional interference mask. ; To prevent interference with outliers containing DC components in the mask, the two-dimensional spectral distance is adjusted towards the center region relative to the mask. Find the intersection; if the distance to the middle region is all within the intersection... In the middle, then from Remove this abnormal DC component value from the data; S3: Calculate the signal-to-noise ratio (SNR) for spectrum classification along the range direction using a one-dimensional interference mask, and set an adaptive iterative threshold based on the SNR; specifically: For the two-dimensional spectrum along the distance direction The interference spectrum and the signal spectrum are divided into interference spectrum and signal spectrum, and the interference energy is calculated by taking the modulus of each. and useful signal energy The signal-to-noise ratio is obtained by taking the maximum value of the interference energy and the average value of the useful signal energy. ; Set the adaptive iteration number of the filter. The difference in signal-to-noise ratio before and after filtering Less than the set iteration threshold When the iteration ends; S4: Based on the set adaptive iteration threshold, a frequency domain notch filter is used to predict two-dimensional interference masks based on the two-dimensional spectrum of the SLC image according to the number of iterations; specifically: By setting up block processing range frequencies, the average value of each range frequency within the block is taken along the azimuth dimension; RFI forms peaks in the range spectrum, and background clutter whose power deviates significantly from the Gaussian distribution will be detected by the filter; The frequency domain filter extracts the feature of whether it deviates from the Gaussian distribution without calculating the DC component, and calculates the Z score for the frequency points within the block; use Calculate the confidence level for whether the score deviates from a Gaussian distribution. The frequency point is compared with the set filter threshold to determine whether it contains interference. An adaptive iteration count is designed, and after multiple iterations, the mask for each block is obtained. The final mask is obtained by summing the results. And then perform the filling process; S5: Perform frequency domain notch filtering based on the mask, complete interference suppression, window the image, and convert it to the time domain to obtain the final image.

2. The SAR image radio frequency interference suppression method based on spectrum outlier detection according to claim 1, characterized in that... S1 specifically refers to: Receive SAR single-look complex image data, read the parameters according to the corresponding parameter file, and perform windowing processing within the corresponding frequency band as required. Use Fast Fourier Transform to obtain the range spectrum and two-dimensional spectrum in the range frequency domain and azimuth time domain of the SLC image data.

3. The SAR image radio frequency interference suppression method based on spectrum outlier detection according to claim 1, characterized in that... S5 specifically refers to: Frequency domain notch filtering is applied to the final mask, and the positions where the interference mask is 1 are set to zero in the spectrum. The two-dimensional spectrum after interference suppression is windowed according to the parameter requirements, and then transformed into the time domain through inverse Fourier transform to obtain the final interference-suppressed SAR image.

4. A computer system, characterized in that... include: 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 cause the one or more processors to implement the method of claim 1.

5. A computer-readable storage medium, characterized in that... The device stores computer-executable instructions, which, when executed, are used to implement the method of claim 1.

Images