SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering

Abstract

The invention provides an SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering. The SAR image LFM interference suppression method comprises the steps that a single-view complex SAR image is divided into a plurality of image blocks; estimating the distance direction modulation frequency of each image block by adopting a multi-stage phase gradient method; transposing the image blocks, and estimating azimuth modulation frequency by adopting a multi-stage phase gradient method; sPECAN filtering processing is carried out on each image block, and an original phase is recovered; and splicing all the image blocks after the SPECAN filtering processing to obtain a complete SAR image after interference suppression. According to the method, the multi-stage phase gradient method is innovatively introduced, and the frequency modulation rate of LFM interference can be quickly and accurately estimated without global search through iteration from small step length to large step length and a multi-stage phase unwrapping mechanism.

Description

Technical Field

[0001] This invention belongs to the field of radar data processing technology, specifically a method for suppressing LFM (linear frequency modulation) interference in SAR images based on phase gradient estimation and SPECAN (spectrum analysis) filtering. Background Technology

[0002] Synthetic Aperture Radar (SAR) is a core sensor for remote sensing. However, SAR systems are highly susceptible to interference from linear frequency modulated (LFM) signals from radiation sources such as ground-based radars in complex electromagnetic environments, leading to large areas of haze or stripes in images and severely obscuring real ground targets. Existing interference suppression methods are mainly divided into preprocessing and post-processing. Among post-processing methods, Block Subspace Filtering (BSF) and its improved algorithms separate interference through matrix decomposition, but suffer from high computational complexity and sensitivity to strong scattering points; Notch Filtering is simple but damages the signal; Time-Frequency Analysis methods are computationally intensive and suffer from cross-term interference. In recent years, the SPECAN filtering method based on frequency modulation estimation has been proposed, but traditional methods rely on prior knowledge or use search methods for frequency modulation estimation, resulting in low computational efficiency and difficulty in meeting practical engineering needs. Summary of the Invention

[0003] This invention proposes a SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering, aiming to solve the problems of high computational complexity, low frequency modulation estimation efficiency, and easy damage to effective signal in existing single-look complex SAR image LFM interference suppression methods.

[0004] The technical solution to achieve the objective of this invention is as follows: A method for suppressing LFM interference in SAR images based on phase gradient estimation and SPECAN filtering, comprising the following steps:

[0005] Step 1: Divide the single-look complex SAR image into several image blocks;

[0006] Step 2: Estimate the range-modulated frequency of each image block using the multi-level phase gradient method;

[0007] Step 3: Transpose the image patch and estimate the azimuth modulation frequency using the multi-level phase gradient method;

[0008] Step 4: Perform SPECAN filtering on each image block to restore the original phase;

[0009] Step 5: Stitch together all the image patches after SPECAN filtering to obtain the complete SAR image after interference suppression.

[0010] Compared with the prior art, the significant advantages of this invention are:

[0011] 1. Accurate and efficient frequency modulation estimation, effectively overcoming phase ambiguity and high computational complexity constraints: This invention innovatively introduces a multi-level phase gradient method. Through iteration from small to large step sizes and a multi-level phase dewinding mechanism, the frequency modulation of LFM interference can be estimated quickly and accurately without global search. This avoids the phase ambiguity problem easily caused by long delay step sizes and greatly reduces the computational complexity of the algorithm.

[0012] 2. Minimal signal impairment and high image fidelity: This invention utilizes a precisely estimated frequency modulation to construct a descrambling function, perfectly compressing broadband LFM interference into an extremely narrow bright spot in the frequency domain. At this point, extremely narrow notch filtering is performed, and combined with inverse Fourier transform and conjugate phase compensation of the descrambling function, the interference can be completely eliminated while preserving the original amplitude and phase information of the SAR image to the maximum extent, with minimal loss of image information.

[0013] 3. Wider scene adaptability and strong generalization ability: This invention achieves accurate local estimation of non-uniformly distributed interference through image block processing, eliminates block boundary effects through overlapping area design, and provides clear parameter adjustment rules for phase ambiguity scenes caused by strong interference. It can adapt to SAR images with different resolutions, sizes and interference intensities, and can achieve stable suppression effects for both single-point strong interference and distributed broadband interference.

[0014] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention.

[0016] Figure 2 This is a detailed flowchart of the multi-stage phase gradient method.

[0017] Figure 3 This is a diagram showing the effect of processing SAR image data before the present invention.

[0018] Figure 4 This is a diagram showing the effect of processing SAR image data according to the present invention. Detailed Implementation

[0019] A method for suppressing LFM interference in SAR images based on phase gradient estimation and SPECAN filtering is proposed. This method accurately estimates the frequency modulation (FMC) using the phase gradient method and combines it with SPECAN filtering to achieve efficient interference suppression. The method includes the following steps:

[0020] Step 1: Image segmentation:

[0021] The single-look complex SAR image is divided into several image blocks, each block being of size [size missing]. ,in This represents the number of sampling points in the azimuth direction. This represents the number of sampling points in the distance direction.

[0022] Step 2: Range-directed frequency estimation:

[0023] For each image patch, the range modulation frequency is estimated using a multi-level phase gradient method. The specific implementation method is as follows:

[0024] For each image patch, perform the following operations. Assume... The current iteration number is indicated by a superscript representing the iteration number.

[0025] Step 2.1: Preprocessing and parameter initialization:

[0026] The image patch data is normalized by amplitude to obtain the normalized data. The specific formula is as follows:

[0027]

[0028] In the formula, This is a normalized matrix; This represents the mean of the magnitudes of all elements in the matrix; To prevent extremely small constants with a denominator of zero, This represents image block data.

[0029] Set the target delay step and initialize the parameters, specifically including:

[0030]

[0031]

[0032] In the formula, Delay the step size for the target; This is a floor function; This represents the number of iterations. This is the step size for the first iteration.

[0033] Step 2.2: Calculate the difference matrix:

[0034] In the In the next iteration, the current step size is used. Calculate the first-order and second-order difference matrices:

[0035]

[0036]

[0037] In the formula, and The first The first and second order difference matrices of the next iteration; It represents the Hadamardi (or Hadama) stack; Represents the conjugate operation for complex numbers; and Indicates step size For normalized matrix The right and left submatrices are obtained by slicing along the distance direction. The right submatrix is... Extract the original matrix from the first... The submatrix consisting of all elements up to the last column, the left submatrix Extract the original matrix from the first column to the... A subarray consisting of all the elements of a column; and Indicates step size For matrix The right and left submatrices are obtained by slicing along the distance.

[0038] Step 2.3: Extract the entanglement phase:

[0039] Summing the second-order difference matrix to suppress noise and extracting the entangled phase:

[0040]

[0041] In the formula, This is the extracted entanglement phase.

[0042] Step 2.4: Multi-stage phase unwinding and frequency modulation update:

[0043] like Calculate the first one directly Next iteration frequency modulation estimate: ;

[0044] like Using the frequency modulation estimate from the previous iteration Unwinding:

[0045]

[0046] Based on the unwinding fuzzy number Calculate the first Next iteration frequency modulation estimate:

[0047]

[0048]

[0049] In the formula, This is the true phase; This is the updated frequency modulation.

[0050] Step 2.5: Step size increment and iteration termination judgment:

[0051] Update the next delay step proportionally:

[0052]

[0053] Step 2.6: Repeat steps 2.2 to 2.5 until... The frequency modulation estimate of the current iteration is used as the range frequency modulation.

[0054] Step 3: Azimuth Chromatography Frequency Estimation: Transpose each image block and estimate the azimuth modulation frequency using the multi-level phase gradient method from Step 2. .

[0055] The image patch is transposed, and steps 2.1 to 2.5 above are repeated to finally output the azimuth frequency modulation. .

[0056] Step 4: SPECAN Filtering: Perform SPECAN filtering on each image block to restore the original phase. The specific steps are as follows:

[0057] Step 4.1: Construct a two-dimensional de-skew reference function

[0058] Based on the estimated range-modulated frequency and azimuth frequency Construct a two-dimensional deslope function:

[0059]

[0060] In the formula, and These represent the time coordinates for the azimuth and range directions, respectively. This is a two-dimensional complex deslanting reference function.

[0061] Step 4.2: Two-dimensional de-skewing process:

[0062] Image patch data Multiplying with the deslope function completes time-domain deslope removal:

[0063]

[0064] In the formula, It is a two-dimensional complex data matrix after deskewing.

[0065] Step 4.3: Two-dimensional Discrete Fourier Transform

[0066] Perform two-dimensional FFT and zero-frequency shift operations on the de-skewed complex data matrix to transform it into the two-dimensional frequency domain:

[0067]

[0068] In the formula, This represents a two-dimensional Fourier transform operation; This is the spectrum matrix.

[0069] Step 4.4: Frequency Domain Interference Detection and Notch Screening

[0070] A constant false alarm rate (CFAR) algorithm is used for adaptive detection of strong interference points in the two-dimensional frequency domain. The specific process is as follows:

[0071] A sliding window is used in the two-dimensional spectrum, with the current frequency to be measured as the center, and the amplitude of neighboring frequency points within the surrounding reference window is used to estimate the local background noise level. Combined with the set false alarm probability, an adaptive threshold corresponding to the frequency to be measured is dynamically calculated.

[0072] (1) Local background power estimation:

[0073] The local background noise level is characterized by the average amplitude of pixels within a reference window surrounding the frequency point to be measured.

[0074]

[0075] In the formula, To use the current frequency to be measured The reference window area centered on; This represents the total number of pixels within the reference window.

[0076] (2) Threshold factor calculation:

[0077] Based on the preset false alarm probability Calculate the threshold factor :

[0078]

[0079] In practical engineering, if the background undulation is small, it can be simplified to a fixed coefficient. .

[0080] (3) Generate adaptive threshold:

[0081] Multiplying the local background mean by the threshold factor yields the dynamic threshold for that frequency point:

[0082]

[0083] Subsequently, the spectral amplitude of the frequency point to be measured was... The signal is compared with its corresponding adaptive threshold: if the spectral amplitude of the frequency point is greater than the adaptive threshold, it is determined that the strong interference condition is met, and it is regarded as a high-energy strong point of LFM interference convergence, and its value in the spectrum is forcibly set to zero; if it is less than or equal to the threshold, it is regarded as a valid signal or normal background and is retained. The specific notch filter formula is as follows:

[0084]

[0085] In the formula, This is the spectral matrix after notch filtering; This indicates the spectral amplitude of the current frequency point being measured; This is the adaptive detection threshold corresponding to this frequency point, calculated using the CFAR algorithm. It is a variable jointly determined by the mean of local background noise and the threshold multiplier, and it changes dynamically with the movement of spatial frequency position, thus reflecting the adaptability of the threshold.

[0086] Step 4.5: Inverse Transformation and Phase Compensation

[0087] The spectral matrix after notch filtering is recovered to the time domain by performing a two-dimensional inverse FFT, and then multiplied by the conjugate descrambling function to recover the original phase:

[0088]

[0089] In the formula, This indicates the inverse two-dimensional Fourier transform operation; The complex conjugate of the deslant reference function; This is to complete the complex data matrix of the image blocks after interference suppression.

[0090] Step 5: Image Reconstruction

[0091] All the SPECAN-filtered image patches are stitched together to obtain a complete SAR image with interference suppression.

[0092] Example

[0093] The following details the implementation steps of the present invention:

[0094] A method for suppressing LFM interference in SAR images based on phase gradient estimation and SPECAN filtering includes:

[0095] Step 1: Data Input and Preprocessing

[0096] Input single-look complex SAR image data and divide it into Image blocks of pixels (block size can be flexibly adjusted according to interference characteristics). To avoid boundary effects in subsequent processing, a certain proportion of overlap area can be set between adjacent image blocks.

[0097] Step 2: Range-directed frequency estimation

[0098] For each image patch, perform the following operations. Assume... The following parameters are all superscripted to indicate the iteration number, representing the current iteration number:

[0099] Step 2.1: Preprocessing and Parameter Initialization

[0100] Extracting image patch data Calculate the mean amplitude and normalize it; set the target delay step size and initialize the parameters:

[0101]

[0102]

[0103]

[0104] In the formula, This is the normalized data matrix; This represents the mean of the magnitudes of all elements in the matrix; Delay the step size for the target; This is a floor function; This is the delay step size for the first iteration; This is the initial estimate for frequency modulation.

[0105] Step 2.2: Calculate the difference matrix

[0106] In the In the next iteration, the current step size is used. Calculate the first-order and second-order difference matrices:

[0107]

[0108]

[0109] In the formula, and The first The first and second order difference matrices of the next iteration; It represents the Hadamardi (or Hadama) stack; Represents the conjugate operation for complex numbers; and Indicates step size The right and left submatrices obtained by slicing a matrix.

[0110] Step 2.3: Extract the entanglement phase

[0111] Global summation of the second-order difference matrix to suppress noise, and extraction of the entangled phase:

[0112]

[0113] Step 2.4: Multi-stage phase unwinding and frequency modulation update

[0114] like Calculate the temporary frequency directly: ;

[0115] like Calculate the desired phase And perform unwinding calculation to determine the true phase:

[0116]

[0117]

[0118]

[0119] In the formula, To unwind the fuzzy numbers; This is the true phase.

[0120] The frequency modulation estimate was then updated: .

[0121] Step 2.5: Increasing Step Size and Determining Iteration Termination

[0122] Update the next delay step proportionally:

[0123]

[0124] like Then let Repeat steps 2.2 to 2.5; if Then exit the loop and output the distance-directed frequency modulation. .

[0125] Step 3: Azimuth frequency estimation

[0126] Perform a matrix transpose on the image patch and repeat steps 2.1 to 2.5 above to obtain the azimuth frequency modulation. .

[0127] Step 4: SPECAN Filtering

[0128] Step 4.1: Construct a two-dimensional de-skew reference function

[0129] Construct distance-time coordinates and azimuth time coordinate And based on the estimated and Construct a two-dimensional descrambling function:

[0130]

[0131] In the formula, This is a two-dimensional complex deslanting reference function.

[0132] Step 4.2: Two-dimensional skew removal

[0133] The input raw image patch complex data matrix Multiplying by the deslope function completes the time-domain deslope removal:

[0134]

[0135] In the formula, It is a two-dimensional complex data matrix after deskewing.

[0136] Step 4.3: Two-dimensional Fourier Transform

[0137] Perform a two-dimensional discrete Fourier transform and zero-frequency shift operation on the descrambled signal to convert it to the two-dimensional frequency domain:

[0138]

[0139] In the formula, This represents a two-dimensional fast Fourier transform and a zero-frequency shift operation of the spectrum; This is the two-dimensional spectrum matrix of the deskewing data; and These are the azimuth and range frequencies, respectively.

[0140] Step 4.4: Frequency Domain Interference Detection and Notch Screening

[0141] A constant false alarm rate (CFAR) detection algorithm is used for adaptive strong interference point detection in the two-dimensional frequency domain. Adaptive thresholds corresponding to each frequency point are extracted. And force the spectrum corresponding to the strong interference frequency point to be set to zero:

[0142]

[0143] In the formula, This is the frequency spectrum matrix after notch filtering in the frequency domain; This is the adaptive detection threshold estimated based on local background noise.

[0144] Step 4.5: Inverse Transformation and Phase Compensation

[0145] The spectrum after notch filtering is restored to the time domain by inverse zero-frequency shift and two-dimensional inverse discrete Fourier transform, and then multiplied by the conjugate descrambling function to recover the original phase of the effective signal.

[0146]

[0147]

[0148] In the formula, This represents the inverse two-dimensional fast Fourier transform and the inverse zero-frequency translation operation; The complex conjugate of the deslant reference function; This is the complex data matrix of the image patch recovered after interference suppression.

[0149] Step 5: Image Reconstruction

[0150] The processing results of all image patches are stitched together according to their original spatial locations, and redundant parts in overlapping areas are removed to obtain a complete SAR image after interference suppression.

Claims

1. A method for suppressing LFM interference in SAR images based on phase gradient estimation and SPECAN filtering, characterized in that, Includes the following steps: Step 1: Divide the single-look complex SAR image into several image blocks; Step 2: Estimate the range-modulated frequency of each image block using the multi-level phase gradient method; Step 3: Transpose the image patch and estimate the azimuth modulation frequency using the multi-level phase gradient method; Step 4: Perform SPECAN filtering on each image block to restore the original phase; Step 5: Stitch together all the image patches after SPECAN filtering to obtain the complete SAR image after interference suppression.

2. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 1, characterized in that, The specific method for estimating the range-modulated frequency of each image patch using the multi-level phase gradient method is as follows: Step 2.1: Normalize the image patch data to obtain normalized data, set the target delay step size, and initialize the number of iterations. and iteration step size ; Step 2.2: In the first... In the next iteration, the current step size is used. Calculate the first-order and second-order difference matrices; Step 2.3: Summing the second-order difference matrix to suppress noise and extracting the entangled phase. ; Step 2.4: Multi-stage phase unwinding and frequency modulation update: like Calculate the first based on the winding phase Next iteration frequency modulation estimate: ; like Using the frequency modulation estimate from the previous iteration Unwind to obtain the unwinding fuzzy number. ; Based on the unwinding fuzzy number Calculate the first Next iteration frequency modulation estimate: ； ； In the formula, This is the true phase; For the updated frequency modulation; Step 2.5: Update the next delay step proportionally; Step 2.6: Repeat steps 2.2 to 2.5 until... , will the The frequency modulation estimate of the next iteration is used as the range-direction frequency modulation.

3. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 2, characterized in that, Utilize the current step size Calculate the first-order and second-order difference matrices: ； ； In the formula, and The first The first and second order difference matrices of the next iteration; It represents the Hadamardi (or Hadama) stack; Represents the conjugate operation for complex numbers; and Indicates step size For normalized matrix The right and left sub-matrices are obtained by slicing along the distance direction. and Indicates step size For first-order difference matrix The right and left submatrices are obtained by slicing along the distance.

4. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 2, characterized in that, Unwinding fuzzy numbers Specifically: 。 5. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 2, characterized in that, The specific formula for updating the next delay step proportionally is as follows: 。 6. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 2, characterized in that, The specific target delay step size is set as follows: ； In the formula, This represents the number of sampling points along the distance.

7. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 1, characterized in that, The specific method for recovering the original phase by performing SPECAN filtering on each image block is as follows: Based on the estimated range-modulated frequency and azimuth frequency Construct a two-dimensional complex deslope function; Image patch data Multiplying with the deslope function completes time-domain deslope removal; Perform two-dimensional FFT and zero-frequency shift operations on the two-dimensional de-skewed complex data matrix to transform it to the two-dimensional frequency domain; A constant false alarm rate algorithm is used for adaptive detection of strong interference points in the two-dimensional frequency domain; The spectrum matrix after notch filtering is recovered to the time domain by a two-dimensional inverse FFT, and then multiplied by the conjugate descrambling function to recover the original phase.

8. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 7, characterized in that, Constructed two-dimensional complex deslope function: ； In the formula, and These represent the time coordinates for the azimuth and range directions, respectively. This is a two-dimensional complex deslanting reference function.

9. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 7, characterized in that, The specific method for adaptive strong interference point detection in the two-dimensional frequency domain using the constant false alarm rate algorithm is as follows: In the two-dimensional spectrum, a sliding window is used to calculate the amplitude of neighboring frequency points within the surrounding reference window, with the current frequency point to be measured as the center. Dynamically calculate the adaptive threshold corresponding to the frequency to be measured by combining the amplitude of neighboring frequency points. Specifically: ； In the formula, Threshold factor, The average amplitude of pixels within a reference window surrounding the frequency point to be measured; The spectral amplitude of the frequency point to be measured The frequency value is compared with the adaptive threshold corresponding to the frequency to be measured: if the spectral amplitude of the frequency to be measured is greater than the adaptive threshold, it is determined that the strong interference condition is met, and the value of the frequency to be measured in the spectrum is forcibly set to zero; if the spectral amplitude of the frequency to be measured is less than or equal to the adaptive threshold, the spectral amplitude of the frequency to be measured is retained.

10. The SAR image LFM interference suppression method based on phase gradient estimation and SPECAN filtering according to claim 9, characterized in that, The threshold factor is as follows: ； In the formula, The preset false alarm probability, This represents the total number of pixels within the reference window.

Images