A Bayesian estimation-based method for suppressing cross-polarization SAR noise

Abstract

This invention relates to a Bayesian estimation-based method for suppressing cross-polarization SAR noise, comprising: acquiring raw cross-polarization data and preprocessing it; acquiring noise vector information corresponding to the cross-polarization data and stripe boundary information of each sub-band; calculating the noise scaling factor of the sub-band based on the Bayesian method; calculating the power balance factor of the sub-band; and subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the raw cross-polarization data to achieve denoising. The beneficial effects of this invention are: it reads and calculates noise vector information from data products, obtains noise scaling factors through Bayesian estimation, and scales the noise vector to achieve the purpose of suppressing azimuth noise.

Description

Technical Field

[0001] This invention relates to the field of noise suppression technology, and more specifically, to a cross-polarization SAR noise suppression method based on Bayesian estimation. Background Technology

[0002] Synthetic Aperture Radar (SAR), as the core carrier of active microwave remote sensing technology, effectively overcomes the monitoring bottlenecks of traditional optical remote sensing under severe weather conditions such as typhoons and heavy rains, thanks to its meter-level spatial resolution and all-weather, cloud-penetrating observation capabilities. It demonstrates unique advantages in areas such as marine dynamic environment monitoring, shipping safety assurance, and disaster emergency response. The C-band SAR system carried by the Sentinel-1 satellite provides global open data services. Its vertical-horizontal (VH) and horizontal-vertical (HV) cross-polarization signal channels can acquire rich backscattering information, sensitively capturing dynamic features such as sea surface micro-scale roughness, waves, and ocean current structure. This provides a reliable data source for marine environmental monitoring, meeting the need for rapid response to changes in the marine environment. Therefore, acquiring high-quality cross-polarization SAR data is a fundamental condition for improving the effectiveness of marine monitoring.

[0003] However, due to the characteristics of the signal channel, cross-polarized data exhibits significant noise characteristics, primarily additive and multiplicative noise coupled with backscattered signals. Currently, Sentinel-1 cross-polarized data denoising methods mainly include ESA noise vector, Kalman filtering, and reconstructed noise field denoising. ESA noise vectors are provided in Sentinel-1 data products in the form of lookup tables, using bilinear interpolation to generate a full-amplitude noise field for subtraction correction. However, ESA noise vectors cannot accurately describe the actual noise distribution, thus failing to achieve effective denoising of cross-polarized data. While Kalman filtering can effectively remove periodic scallop signals in the azimuth direction, it cannot effectively improve interband noise caused by abrupt power changes in the range direction. Noise field denoising modifies the ESA noise vector to better reflect the actual noise distribution, making it more suitable for calm sea surfaces. However, it struggles to effectively denoise sea surfaces with significant variations in roughness. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a cross-polarization SAR noise suppression method based on Bayesian estimation.

[0005] Firstly, a method for suppressing cross-polarization SAR noise based on Bayesian estimation is provided, including:

[0006] S1. Obtain the raw cross-polarization data and perform preprocessing;

[0007] S2. Obtain the noise vector information corresponding to the cross-polarization data and the strip boundary information of each sub-band;

[0008] S3. Calculate the noise scaling factor of the sub-band based on the Bayesian method;

[0009] S4. Calculate the power balance factor of the sub-band;

[0010] S5. Reconstruct the two-dimensional noise field based on the noise scaling factor and power balance factor, and achieve denoising by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

[0011] Preferably, S3 includes:

[0012] S301. Within each subband, the cross-polarization data is divided into several blocks, and the minimum weighted linear residual sum of squares is calculated for each block.

[0013] S302. Construct the likelihood function and prior distribution function of the noise scaling factor based on the minimum weighted linear minimum residual sum of squares;

[0014] S303. Based on the likelihood function and prior distribution function, the posterior distribution of the noise scaling factor is estimated by Bayesian method, and the value with the highest probability in the posterior distribution is selected as the noise scaling factor for the block.

[0015] S304. Calculate the sum of the scaling factors of all blocks within the sub-band and take the average value, which is used as the noise scaling factor of this sub-band.

[0016] Preferably, S4 includes:

[0017] S401. Determine the power abrupt change region between subbands based on the subband boundary information;

[0018] S402. Calculate the average power mean of the blocks in the power abrupt change region between subbands as the power balance coefficient of the blocks, and then use the average balanced power of all blocks in the region as the power balance coefficient of the subband.

[0019] As a preferred embodiment, in S301, the expression for calculating the minimum weighted linear residual sum of squares is:

[0020]

[0021] in, It is the backscattering coefficient after noise suppression using the initial noise scaling factor within the block; σ 0 It is the cross-original backscattering coefficient; k is the initial scaling factor set within the block, G ds It is the scallop gain read; It is the noise vector; RSS(k) is the least weighted linear residual sum of squares; i is each block, representing the i-th block; ω i These are weights, determined by the absolute gradient of the noise, used to reflect the details of noise variation; It is within the block Fitting results with distance-oriented index.

[0022] As a preferred embodiment, in S302, the calculation expressions for the likelihood function and the prior distribution function are as follows:

[0023]

[0024] Where, L(σ) 0 |k) is the likelihood function, and p(k) is the prior distribution function; Let μ be the variance of the initial noise vector in the block, τ be the prior variance distribution of the initial noise vector in the block, and μ be the variance of the initial noise vector in the block. k The prior mean of the initial scaling factor in the block.

[0025] Preferably, in S303, the expression for calculating the noise scaling factor is as follows:

[0026] p(k|σ 0 )∝L(σ 0 |k)·p(k)

[0027]

[0028] Where K is s,i Calculate the noise scaling factor for each block.

[0029] Preferably, in S402, the formula for calculating the power balance coefficient is as follows:

[0030]

[0031] Among them, start right and start left It refers to the data boundary at the extracted sub-band boundary. It is the result of subtracting the scaled noise vector, where j is the block within the selected area; K is the balance coefficient of the j-th block in the selected region; N is the number of pixels in the selected region; b It is the power balance factor of the sub-band.

[0032] In a second aspect, a cross-polarization SAR noise suppression system based on Bayesian estimation is provided for performing any of the methods described in the first aspect, including:

[0033] The first acquisition module is used to acquire raw cross-polarization data and perform preprocessing.

[0034] The second acquisition module is used to acquire noise vector information corresponding to the cross-polarization data and strip boundary information of each sub-band;

[0035] The first calculation module is used to calculate the noise scaling factor of the sub-band based on the Bayesian method;

[0036] The second calculation module is used to calculate the power balance coefficient of the sub-band;

[0037] The denoising module is used to reconstruct a two-dimensional noise field based on the noise scaling factor and power balance factor. Denoising is achieved by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

[0038] Thirdly, a computer storage medium is provided, wherein a computer program is stored therein; when the computer program is run on a computer, the computer causes the computer to perform any of the methods described in the first aspect.

[0039] Fourthly, an electronic device is provided, comprising:

[0040] Memory, used to store computer programs;

[0041] A processor for executing the computer program to implement the method as described in any of the first aspects.

[0042] The beneficial effects of this invention are as follows: Addressing the problem of significant noise in Sentinel-1 cross-polarization data, leading to low data quality and hindering efficient application, this invention first reads and calculates noise vector information from the data product. Then, using Bayesian estimation, it obtains a noise scaling factor to scale the noise vector, thereby suppressing azimuth noise. Simultaneously, for power abrupt changes between range strips, it calculates a power balance factor to suppress these power abrupt changes, effectively mitigating the problem. This invention overcomes the issues of significant noise interference and low signal-to-noise ratio in Sentinel-1 cross-polarization data. Compared to commonly used ESA noise vector methods, this method significantly outperforms ESA noise vector methods, yielding high-quality cross-polarization SAR data, which is of great significance for the widespread application of cross-polarization SAR. Attached Figure Description

[0043] Figure 1 A flowchart of a cross-polarization SAR noise suppression method based on Bayesian estimation provided in this application;

[0044] Figures 2a-2c This is a schematic diagram of the results obtained using the noise scaling factor;

[0045] Figures 3a-3bA schematic diagram comparing the reconstructed two-dimensional noise field with the ESA noise vector;

[0046] Figures 4a-4d A schematic diagram showing the denoising results and signal-to-noise ratio under calm sea conditions;

[0047] Figures 5a-5d The denoising results and signal-to-noise ratio for complex sea conditions. Detailed Implementation

[0048] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0049] Example 1:

[0050] To address the problems of existing technologies, Embodiment 1 of this application provides a method for suppressing cross-polarization SAR noise based on Bayesian estimation, such as... Figure 1 As shown, it includes:

[0051] S1. Obtain the raw cross-polarization data and perform preprocessing.

[0052] Specifically, the raw cross-polarization data is Sentinel-1 cross-polarization data, and the Sentinel-1 cross-polarization data is preprocessed in SNAP software, including orbit correction, radiometric calibration and other operations.

[0053] It should be noted that, in order to make full use of the high spatial resolution and rich backscattering information of cross-polarization SAR data, EW or IW mode data can be selected according to actual needs.

[0054] S2. Obtain the noise vector information corresponding to the cross-polarization data and the strip boundary information of each sub-band.

[0055] Specifically, noise vector information and stripe boundary information for each subband are read from the XML file in the Sentinel-1 data product.

[0056] S3. Calculate the noise scaling factor of the subband based on the Bayesian method.

[0057] S3 includes:

[0058] S301. Within each subband, the cross-polarized data is divided into several blocks, and the minimum weighted linear residual sum of squares (RSS) of each block is calculated.

[0059] For example, Sentinel-1 cross-polarization SAR data is divided into 500×500 blocks, and all blocks are iterated through, with the same processing steps performed on each block. Then, the minimum weighted linear residual sum of squares for each block is calculated based on the range index, and the absolute gradient of the noise vector is used as the weight for fitting.

[0060] In S301, the calculation of the minimum weighted linear residual sum of squares requires setting an initial noise scaling factor within the block. Based on existing experience, the step size is set to 0.01, with a range of 0.8-2, and the minimum weighted linear residual sum of squares is calculated iteratively.

[0061] Specifically, the expression for calculating the least weighted linear residual sum of squares is:

[0062]

[0063] in, It is the backscattering coefficient after noise suppression using the initial noise scaling factor within the block; σ 0 It is the cross-original backscattering coefficient; k is the initial scaling factor set within the block, G ds It is the scallop gain read; It is the noise vector; RSS(k) is the least weighted linear residual sum of squares; i is each block, representing the i-th block; ω i These are weights, determined by the absolute gradient of the noise, used to reflect the details of noise variation; It is within the block The fitting results are compared with the distance-oriented index. Specific steps include:

[0064] (1) First, set the range and step size of the initial noise scaling factor: based on existing experience, set the step size to 0.01 and the range to 0.8-2, and calculate RSS(k) in a loop;

[0065] (2) Next, calculate the absolute gradient of the noise vector as the weight;

[0066] (3) Fit the data according to the distance index and end the loop when RSS(k) reaches its minimum.

[0067] S302. Construct the likelihood function and prior distribution function of the noise scaling factor based on the minimum weighted linear minimum residual sum of squares, make prior assumptions about the noise scaling factor, and constrain and correct the estimation process to optimize the estimation result of the noise scaling factor and calculate the maximum likelihood value.

[0068] In S302, the construction of the prior distribution function, with only one prior information, the noise vector, can be adjusted by setting a large prior variance to represent the uncertainty of the scaling factor.

[0069] Specifically, the expressions for calculating the likelihood function and the prior distribution function are as follows:

[0070]

[0071] Where, L(σ) 0 |k) is the likelihood function, and p(k) is the prior distribution function; Let μ be the variance of the initial noise vector in the block, τ be the prior variance distribution of the initial noise vector in the block, and μ be the variance of the initial noise vector in the block. k The prior mean of the initial scaling factors in the block. Specific steps include:

[0072] (1) Calculate the variance in each block

[0073] (2) The uncertainty of the scaling factor is represented by setting a large prior variance τ.

[0074] S303. Estimate the posterior distribution of the noise scaling factor. Combining the characteristics of the likelihood function and prior information, infer the posterior distribution of the scaling factor using Bayes' theorem, and estimate the possible probability distribution of the factor. Select the maximum value of the posterior distribution probability as the scaling factor for the block.

[0075] In S303, the expression for calculating the noise scaling factor is as follows:

[0076] p(k|σ 0 )∝L(σ 0 |k)·p(k)

[0077]

[0078] Where K is s,i Calculate the noise scaling factor for each block. Specific steps include:

[0079] (1) Combining the prior distribution and the likelihood function, the posterior distribution of the scaling factor is inferred using Bayes' theorem;

[0080] (2) After obtaining the posterior distribution of the scaling factor, select the maximum value of the posterior distribution probability as the scaling factor of the block.

[0081] S304. Calculate the sum of the scaling factors of all blocks within a sub-band and take the average value, which is used as the noise scaling factor for this sub-band. For EW mode sub-bands, the values ​​are 1, 2, 3, 4, and 5; for IW mode data sub-bands, the values ​​are 1, 2, and 3.

[0082] The expression for the subband noise scaling factor is as follows:

[0083]

[0084] After obtaining the noise scaling factors for all sub-bands, the noise vector is scaled to obtain the scaled cross-polarization SAR image, as shown below. Figures 2a-2c As shown, where, Figure 2a This is a cross-polarization SAR image after noise scaling. Figure 2b The trend of the azimuth backscattering coefficient after noise scaling is shown. Figure 2c The trend of the range backscattering coefficient after noise scaling is shown. Figure 2b and Figure 2c In the diagram, the solid blue line represents the normalized radar cross section (NRCS) after decibel conversion, the dashed yellow line represents the noise vector scaled by a noise scaling factor, and the solid red line represents the original NRCS minus the scaled noise vector. It should be noted that the processing was performed before NRCS decibel conversion; the conversion to decibels is for ease of demonstration.

[0085] S4. Calculate the power balance factor of the sub-band.

[0086] S5. Reconstruct the two-dimensional noise field based on the noise scaling factor and power balance factor, and achieve denoising by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

[0087] Example 2:

[0088] Building upon Example 1, Example 2 of this application provides a more specific method for suppressing cross-polarization SAR noise based on Bayesian estimation, including:

[0089] S1. Obtain the raw cross-polarization data and perform preprocessing.

[0090] S2. Obtain the noise vector information corresponding to the cross-polarization data and the strip boundary information of each sub-band.

[0091] S3. Calculate the noise scaling factor of the subband based on the Bayesian method.

[0092] S4. Calculate the power balance factor of the sub-band.

[0093] S4 includes:

[0094] S401. Determine the power abrupt change region between subbands based on the strip boundary information of the subbands.

[0095] Specifically, subband information of SAR data is read from XML files to determine the transition regions between subbands and the regions where the backscattering coefficient changes abruptly.

[0096] In S401, the expression for determining the inter-subband power abrupt region is as follows:

[0097] start left =min(lastRangeSample) sub-1 )+[-20:-1]

[0098] start right =max(firstRangeSample) sub )+[1:20]

[0099] Among them, start right and start left This is the data boundary at the extracted sub-band boundary; sub is the sub-band, which is 2, 3, 4, or 5 in EW mode and 2 or 3 in IW mode; firstRangeSample and lastRangeSample can be read from the XML file in the data product.

[0100] S402. Calculate the average power mean of the blocks within the power abrupt change region between subbands as the power balance coefficient of the block, and then use the average balanced power of all blocks within the region as the power balance coefficient of the subband. Specifically, the subband values ​​for the power balance coefficient of EW mode are 2, 3, 4, and 5; the subband values ​​for the power balance coefficient of IW mode are 2 and 3.

[0101] In S402, the formula for calculating the power balance factor is as follows:

[0102]

[0103] in, It is the result of subtracting the scaled noise vector, where j is the block within the selected area; K is the balance coefficient of the j-th block in the selected region; N is the number of pixels in the selected region; b This is the power balance factor of the sub-band. The specific steps include:

[0104] (1) First, subtract the scaled noise vector to obtain the preliminary denoising result;

[0105] (2) Secondly, the average power within a block is used to obtain the power balance coefficient of a block;

[0106] (3) Finally, the average power balance coefficient of all blocks in the region is used as the power balance coefficient of a sub-band. The sub-bands of EW mode are 2, 3, 4 and 5 respectively; the sub-bands of IW mode are 2 and 3 respectively.

[0107] S5. Reconstruct the two-dimensional noise field based on the noise scaling factor and power balance factor, and achieve denoising by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

[0108] In S5, the noise field reconstructed based on the ESA noise vector and Bayesian estimation is as follows: Figure 3a and Figure 3b As shown, where Figure 3a The noise field calculated for the ESA noise vector. Figure 3b To estimate the reconstructed noise field using Bayesian estimation.

[0109] The expressions for reconstructing and denoising the two-dimensional noise field are as follows:

[0110]

[0111] in, It is a reconstructed two-dimensional noise field; These are the denoised cross-polarized backscattering coefficients. The specific steps include:

[0112] (1) By the noise scaling factor K s and power balance coefficient K s By positively modifying the noise vector, a two-dimensional noise field is reconstructed;

[0113] (2) Noise reduction is achieved by subtracting the reconstructed two-dimensional noise field from the original backscattering coefficient.

[0114] In addition, signal-to-noise ratio (SNR) is often used as a metric to measure data quality. A comparative analysis of the SNR of the reconstructed noise field with that of ESA noise vector denoising and the original data revealed that the SNR of the reconstructed noise field is significantly higher than that of ESA noise vector denoising and the original data.

[0115] In calm sea conditions and complex sea conditions, such as Figures 4a-4d and Figures 5a-5d As shown, the reconstructed noise field exhibits good denoising performance in both the azimuth and range directions. Specifically, Figure 4a The original SAR image under calm sea conditions. Figure 4b The image is a noise-reconstructed image obtained under calm sea conditions. Figure 4c For the azimuth signal-to-noise ratio under calm sea conditions, Figure 4d The range signal-to-noise ratio under calm sea conditions. Figure 5a Raw SAR images under complex sea conditions. Figure 5b This is an image denoised by reconstructing a noise field under complex sea conditions. Figure 5c For the azimuth signal-to-noise ratio under complex sea conditions, Figure 5d The range signal-to-noise ratio under complex sea conditions. Figure 4c , Figure 4d , Figure 5c and Figure 5dIn the diagram, the blue solid line represents the original SNR of the data, the yellow dashed line represents the SNR after ESA noise vector denoising, and the red solid line represents the SNR after noise field denoising reconstructed in this invention.

[0116] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be referred to each other, and will not be repeated in this application.

[0117] Example 3:

[0118] Based on Examples 1 and 2, Example 3 of this application provides a cross-polarization SAR noise suppression system based on Bayesian estimation, comprising:

[0119] The first acquisition module is used to acquire raw cross-polarization data and perform preprocessing.

[0120] The second acquisition module is used to acquire noise vector information corresponding to the cross-polarization data and strip boundary information of each sub-band;

[0121] The first calculation module is used to calculate the noise scaling factor of the sub-band based on the Bayesian method;

[0122] The second calculation module is used to calculate the power balance coefficient of the sub-band;

[0123] The denoising module is used to reconstruct a two-dimensional noise field based on the noise scaling factor and power balance factor. Denoising is achieved by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

[0124] It should be noted that the system provided in this embodiment is the system corresponding to the method provided in Embodiments 1 and 2. Therefore, the parts in this embodiment that are the same as or similar to those in Embodiments 1 and 2 can be referred to each other, and will not be repeated in this application.

Claims

1. A method for suppressing cross-polarization SAR noise based on Bayesian estimation, characterized in that, include: S1. Obtain the raw cross-polarization data and perform preprocessing; S2. Obtain the noise vector information corresponding to the cross-polarization data and the strip boundary information of each sub-band; S3. Calculate the noise scaling factor of the sub-band based on the Bayesian method; S3 includes: S301. Within each subband, the cross-polarization data is divided into several blocks, and the minimum weighted linear residual sum of squares is calculated for each block. In S301, the expression for calculating the minimum weighted linear residual sum of squares is: in, It is the backscattering coefficient after noise suppression using the initial noise scaling factor within the block; σ 0 It is the cross-original backscattering coefficient; k is the initial scaling factor set within the block, G ds It is the scallop gain read; It is the noise vector; RSS(k) is the least weighted linear residual sum of squares; i is each block, representing the i-th block; ω i These are weights, determined by the absolute gradient of the noise, used to reflect the details of noise variation; It is within the block Fitting results with distance-oriented index; S302. Construct the likelihood function and prior distribution function of the noise scaling factor based on the minimum weighted linear residual sum of squares; S303. Based on the likelihood function and prior distribution function, the posterior distribution of the noise scaling factor is estimated by Bayesian method, and the value with the highest probability in the posterior distribution is selected as the noise scaling factor for the block. S304. Calculate the sum of the scaling factors of all blocks within the sub-band and take the average value, which is used as the noise scaling factor of this sub-band. S4. Calculate the power balance factor of the sub-band; S4 includes: S401. Determine the power abrupt change region between subbands based on the subband boundary information; S402. Calculate the average power mean of the blocks in the power abrupt change region between subbands as the power balance coefficient of the blocks, and then use the average balanced power of all blocks in the region as the power balance coefficient of the subband. S5. Reconstruct the two-dimensional noise field based on the noise scaling factor and power balance factor, and achieve denoising by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

2. The cross-polarization SAR noise suppression method based on Bayesian estimation according to claim 1, characterized in that, In S302, the expressions for calculating the likelihood function and the prior distribution function are as follows: Where, L(σ) 0 |k) is the likelihood function, and p(k) is the prior distribution function; Let μ be the variance of the initial noise vector in the block, τ be the prior variance distribution of the initial noise vector in the block, and μ be the variance of the initial noise vector in the block. k The prior mean of the initial scaling factor in the block.

3. The cross-polarization SAR noise suppression method based on Bayesian estimation according to claim 2, characterized in that, In S303, the expression for calculating the noise scaling factor is as follows: p(k|σ 0 )∝L(σ 0 |k)·p(k) Where K is s,i Calculate the noise scaling factor for each block.

4. The cross-polarization SAR noise suppression method based on Bayesian estimation according to claim 3, characterized in that, In S402, the formula for calculating the power balance factor is as follows: Among them, start right and start left It refers to the data boundary at the extracted sub-band boundary. It is the result of subtracting the scaled noise vector, where j is the block within the selected area; K is the balance coefficient of the j-th block in the selected region; N is the number of pixels in the selected region; b It is the power balance factor of the sub-band.

5. A cross-polarization SAR noise suppression system based on Bayesian estimation, characterized in that, For performing the method according to any one of claims 1 to 4, comprising: The first acquisition module is used to acquire raw cross-polarization data and perform preprocessing. The second acquisition module is used to acquire noise vector information corresponding to the cross-polarization data and strip boundary information of each sub-band; The first calculation module is used to calculate the noise scaling factor of the sub-band based on the Bayesian method; The second calculation module is used to calculate the power balance coefficient of the sub-band; The denoising module is used to reconstruct a two-dimensional noise field based on the noise scaling factor and power balance factor. Denoising is achieved by subtracting the reconstructed two-dimensional noise field from the backscattering coefficient of the original cross-polarization data.

6. A computer storage medium, characterized in that, The computer storage medium stores a computer program; when the computer program is run on the computer, it causes the computer to perform the method described in any one of claims 1 to 4.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 4.

Images