A SAR image filtering method based on structural strength joint consistency

Abstract

The application discloses a SAR image filtering method based on structure strength joint consistency, and relates to the technical field of radar remote sensing image processing methods.The method comprises the following steps: acquiring double-time-phase SAR images of the same ground surface monitoring area, performing radiation correction and spatial registration to form a double-time-phase SAR image pair to be processed; respectively calculating structure consistency components and strength consistency components of the double-time-phase SAR image pair, and adopting a product strategy to construct a structure-strength joint consistency coefficient; calculating a spatial self-adaptive filtering weight based on the structure-strength joint consistency coefficient, and automatically adjusting filtering strength of flat areas and edge areas; and adopting a consistency-guided self-adaptive weighted fusion filtering to output a final denoising image.The method can capture real geometric features, effectively distinguish the features, and avoid edge blurring and texture loss.

Description

Technical Field

[0001] This invention relates to the field of radar remote sensing image processing methods, and specifically to a SAR image filtering method based on joint structural strength consistency. Background Technology

[0002] With the rapid development of remote sensing technology, Synthetic Aperture Radar (SAR) has become an important tool in the field of Earth observation due to its all-weather, all-day imaging capabilities and strong penetration into the physical properties of the Earth's surface. However, the SAR imaging mechanism inevitably introduces multiplicative speckle noise, which severely degrades image quality, obscures the edges and details of ground features, and leads to a decrease in the accuracy of subsequent data applications. Therefore, how to suppress speckle noise while maintaining the geometric structure of image edges is a challenge in the field of SAR image processing.

[0003] Despite the numerous existing SAR image filtering methods, significant limitations remain in complex scenarios: First, traditional spatial filtering often employs global smoothing strategies, which frequently sacrifice image clarity, resulting in blurred edges and lost textures. Second, methods based on local statistics rely solely on grayscale features, making it difficult to effectively distinguish between speckle noise and ground object edges, which share high-frequency characteristics, and can easily lead to missmoothing of structures. Third, constrained by the multiplicative noise mechanism, drastic fluctuations in strong scattering regions are often misjudged as real changes, making it difficult for existing technologies to achieve an ideal balance between "complete denoising" and "detail preservation." Summary of the Invention

[0004] The technical problem to be solved by this invention is how to provide a SAR image filtering method that can capture real geometric features, make effective distinctions, and avoid edge blurring and texture loss.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a SAR image filtering method based on joint structural strength consistency, the method comprising the following steps:

[0006] S1: Acquire dual-temporal SAR images of the same surface monitoring area, perform radiometric correction and spatial registration, and form a dual-temporal SAR image pair to be processed;

[0007] S2: Calculate the structural consistency component and intensity consistency component of the dual-temporal SAR image pair respectively, and construct the joint structural-intensity consistency coefficient using a product strategy;

[0008] S3: Calculate spatial adaptive filtering weights based on the joint structure-strength consistency coefficient, and automatically adjust the filtering intensity in flat and edge regions;

[0009] S4: Adaptive weighted fusion filtering guided by consistency is used to output the final denoised image.

[0010] The beneficial effects of adopting the above technical solution are as follows: 1) The structural consistency component constructed by the method described herein captures geometric topological features through gradient direction field. By utilizing the essential difference between the directional randomness of noise and the directional stability of the edge, it captures the real geometric features, thereby making effective distinctions and avoiding edge blurring and texture loss.

[0011] 2) The intensity consistency component of the method is constructed based on the normalized mean ratio. It normalizes the intensity difference between the two time phases and maps ground objects with different radiation levels to a unified consistency space, thereby eliminating the interference of absolute brightness on change detection. This effectively suppresses noise fluctuations in high-brightness areas and minimizes the risk of ground object misjudgment caused by noise fluctuations.

[0012] 3) The proposed method employs a consistency-guided adaptive weighting mechanism. Based on the structure-strength joint consistency coefficient, it calculates nonlinear filter weights, forming a spatially adaptive gating strategy: in regions of high consistency, the weights tend to smooth the reference image, suppressing random noise; in regions of low consistency, the weights tend to resemble the original image, preserving details. This mechanism achieves maximum denoising while retaining edge details, providing a high-quality data foundation for subsequent SAR image applications. Attached Figure Description

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0014] Figure 1 This is the main flowchart of the method described in an embodiment of the present invention;

[0015] Figure 2 This is the first temporal SAR image in this embodiment of the invention, SAR1;

[0016] Figure 3 This is the second temporal SAR image in this embodiment of the invention, SAR2;

[0017] Figure 4 This is a diagram showing the Gaussian filtering result applied to the entire SAR1 image in an embodiment of the present invention.

[0018] Figure 5 This is a diagram showing the filtering result of mean filtering on the entire SAR1 image in an embodiment of the present invention;

[0019] Figure 6 This is a diagram showing the filtering result of Lee filtering on the entire SAR1 image in an embodiment of the present invention;

[0020] Figure 7This is a diagram showing the filtering result of the filter proposed in this embodiment of the invention on the entire SAR1 image;

[0021] Figure 8 This is a diagram showing the Gaussian filtering result applied to the entire SAR2 image in an embodiment of the present invention.

[0022] Figure 9 This is a diagram showing the filtering result of mean filtering on the entire SAR2 image in an embodiment of the present invention;

[0023] Figure 10 This is a diagram showing the filtering result of Lee filtering on the entire SAR2 image in an embodiment of the present invention;

[0024] Figure 11 This is a diagram showing the filtering result of the proposed filter on the entire SAR2 image. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] Overall, such as Figure 1 As shown in the figure, this invention discloses a SAR image filtering method based on joint consistency of structure and strength. This method utilizes dual-temporal SAR images of the same monitoring area and adaptively guides the filtering process by constructing a joint consistency coefficient between structure and strength. Specifically, it includes the following steps:

[0028] S1: Acquire dual-temporal SAR images of the same surface monitoring area, perform radiometric correction and spatial registration, and form a dual-temporal SAR image pair to be processed;

[0029] S2: Calculate the structural consistency component and intensity consistency component of the dual-temporal SAR image pair respectively, and construct the joint structural-intensity consistency coefficient using a product strategy;

[0030] S3: Calculate spatial adaptive filtering weights based on the joint structure-strength consistency coefficient, and automatically adjust the filtering intensity in flat and edge regions;

[0031] S4: Adaptive weighted fusion filtering guided by consistency is used to output the final denoised image.

[0032] The above steps will be explained in detail below:

[0033] S1: Acquire dual-temporal SAR images of the same surface monitoring area, perform radiometric correction and spatial registration to form a dual-temporal SAR image pair to be processed. The original dual-temporal SAR images are as follows: Figure 2 and Figure 3 As shown, it is used for radiometric correction and spatial registration, providing a standardized data source for subsequent SAR image processing.

[0034] First, the dual-temporal SAR image data to be processed is acquired. Since the SAR images from different time phases were acquired at different times, spatial location deviations may exist; therefore, image standardization preprocessing is necessary. Radiometric correction is performed on the first and second temporal SAR images to eliminate systematic errors, and the image pixel values ​​are converted into backscattering coefficients. Subsequently, high-precision spatial geometric registration is performed to ensure that the two images are strictly aligned in the pixel coordinate system, forming the dual-temporal image pair to be processed.

[0035] S2: Calculate the structural consistency component and intensity consistency component of the dual-temporal SAR image pair separately, and construct the joint structure-intensity consistency coefficient using a product strategy. Based on the preprocessed dual-temporal SAR image pair, calculate the structural consistency component and intensity consistency component using gradient direction difference and normalized mean ratio respectively, and construct the joint consistency coefficient reflecting the spatiotemporal stability of ground features through nonlinear fusion.

[0036] S2-1: Constructing structural consistency components using gradient direction fields This study utilizes gradient direction field information to capture the geometric topological stability of ground features. Real ground feature edges maintain consistent orientation across different time phases, while the gradient direction of noise is random. First, the Sobel edge detection operator is used to analyze the first-phase SAR image. Second-phase SAR images Perform convolution operations to calculate the gradient components of the image in the horizontal direction. and the gradient component in the vertical direction .

[0037] The specific calculation formula is as follows: Let... The horizontal Sobel convolution kernel is used. The vertical Sobel convolution kernel is defined as follows:

[0038] ;

[0039] For any pixel Its horizontal gradient with vertical gradient Calculated separately as follows:

[0040] ;

[0041] ;

[0042] in, The input SAR image is the image to be processed. In the specific calculations, the above formulas are applied to the first temporal SAR image. Second-phase SAR images To obtain the gradient components of their respective time phases; symbol This represents a two-dimensional discrete convolution operation.

[0043] Subsequently, the gradient direction angle corresponding to each pixel in the two time phases is calculated using the four-quadrant arctangent function. and The calculation formula is as follows:

[0044] ;

[0045] ;

[0046] in, The gradient component in the horizontal direction of the first phase image; The gradient component in the vertical direction of the first phase image; This represents the gradient component in the horizontal direction of the second phase image; This represents the gradient component in the vertical direction of the second phase image;

[0047] The set of gradient direction angles calculated for all pixels in the image constitutes the gradient direction fields of the first and second time phases.

[0048] Based on the obtained dual-phase gradient direction angles, the minimum periodic angle difference is mapped to a structural consistency component using a nonlinear exponential decay strategy. The calculation formula is as follows:

[0049] ;

[0050] in, and These are the gradient direction angles for the first and second time phases, respectively. This represents a natural exponential function with base e. The function is used to handle the periodicity of angles. This is a structure sensitivity factor used to adjust the decay rate in response to differences in gradient direction.

[0051] S2-2: Constructing intensity consistency components using normalized mean ratio The normalized mean ratio statistic is used to capture the radiative stability of backscattering intensity of ground objects. A local sliding window is defined, and the local mean values ​​of the first and second temporal SAR images within the window are calculated respectively. and .

[0052] Based on the multiplicative noise model of SAR images, a normalized mean ratio statistic is constructed to obtain the intensity consistency component. The calculation formula is as follows:

[0053] ;

[0054] in, The constant is a non-zero small constant used for numerical stability control. This formula eliminates the influence of absolute brightness of ground features through ratio calculations, mapping ground features of different radiation levels to a unified interval while retaining only the normalized noise statistical characteristics. This gives the algorithm constant false alarm rate adaptability to noise in high-brightness areas, avoiding misjudgments caused by large noise fluctuations in strong reflection areas.

[0055] S2-3: Genetic structure-strength joint consistency coefficient A nonlinear product fusion strategy is employed to address the structural consistency components. The intensity consistency component Pixel-by-pixel multiplication is performed, meaning that for each pixel in the image, its corresponding structure consistency component value is multiplied by its intensity consistency component value. This ensures that low consistency values ​​in either component are dominated by multiplication and suppressed in the final joint coefficient, thus outputting high consistency values ​​only in regions where both geometric structure and radiative intensity remain stable. The structure-intensity joint consistency coefficient is then calculated. The calculation formula is:

[0056] ;

[0057] This fusion strategy implements a "one-vote veto" mechanism through product operations. When a pixel exhibits disordered gradient directions in the structural domain or significant radiative changes in the intensity domain, the joint consistency coefficient approaches 0, indicating that the pixel is a non-background stable region, i.e., an edge, texture, or change region.

[0058] S3: Calculate spatial adaptive filtering weights based on the structure-strength joint consistency coefficient, and automatically adjust the filtering intensity in flat and edge regions. Based on the aforementioned structure-strength joint consistency coefficient... Spatial adaptive filtering weights are calculated using the Sigmoid nonlinear function. The weights are set to automatically adjust the filtering strength based on the consistency coefficient. In regions where the consistency coefficient indicates high stability, weight values ​​tending towards smoothness are generated; in regions where the consistency coefficient indicates low stability, weight values ​​tending towards fidelity are generated. The calculation formula is as follows:

[0059] ;

[0060] in, This is the kurtosis factor, used to adjust the slope of the weighting curve and its sensitivity to changes in consistency. This is the center offset of the soft threshold, used to set the critical point for consistency judgment; The formula represents the natural exponential function; it is constructed using the Sigmoid function form and utilizes its saturation properties to achieve a non-linear mapping of the consistency coefficient; when the pixel consistency coefficient... Greater than At that time, weight A value close to 1 ensures that the pixel primarily adopts the smoothed and denoised image in the filtering result, thus pointing to a highly stable region; when the pixel's consistency coefficient approaches 1... Less than At that time, weight Approaching 0 causes the pixel to primarily adopt the original image containing edge details in the filtering result, thus pointing to a region of low stability.

[0061] S4: Employ consistency-guided adaptive weighted fusion filtering to output the final denoised image. Utilize the aforementioned spatial adaptive filtering weights... The original SAR image containing edge details and the smoothed and denoised SAR image are weighted and fused to output the final denoised SAR image. The formula for calculating the consistency-guided adaptive weighted filter is:

[0062] ;

[0063] in, The original SAR image containing edge details, The smoothed and denoised SAR image is generated by applying a large-window mean filter or Gaussian filter of a preset size to the original SAR image containing edge details, in order to provide a noise-free reference for flat areas.

[0064] This formula uses weights Pixel-level dynamic complementarity is achieved: when a pixel is located in a highly consistent background region, the fusion result mainly adopts... This achieves noise reduction; when pixels are located in low-consistency edge regions, the fusion result mainly adopts... This allows for the preservation of details, thereby effectively removing multiplicative noise while maximizing the retention of texture details and geometric structures.

[0065] The effectiveness of the present invention will be further illustrated by the following SAR image experiments.

[0066] 1) Experimental setup:

[0067] Experimental data are from the yellow_A image, such as Figure 2 and Figure 3 As shown. To verify that the SAR image filtering method based on joint structural strength consistency proposed in this invention has good noise suppression capability, a selection was made. Figure 2 and Figure 3 Experiments were conducted in flat regions I and II to verify the edge preservation capability of the method proposed in this invention. Figure 2 and Figure 3 Experiments were conducted on edge texture region III in the model. The performance of Gaussian filtering, mean filtering, and Lee filtering was compared with the method proposed in this invention.

[0068] 2) Results Analysis:

[0069] This experiment uses the equivalent number of looks (ENL) to evaluate the noise suppression capability of the filtering algorithm in the flat regions I and II of the SAR image. ENL is an important indicator of the smoothness of a SAR image, mainly used to evaluate the filter's ability to suppress speckle noise in a uniform background region. A larger ENL value indicates higher image smoothness and less residual speckle noise in that region. It is defined as:

[0070] ;

[0071] in, This represents the mean pixel intensity within a selected uniform region. This represents the standard deviation of pixel intensity within the region.

[0072] The Edge Preservation Index (ESI) was used to quantitatively analyze the edge preservation capability of SAR image edge region III. ESI measures the ability of a filter to preserve image edge details during denoising. It quantifies edge sharpness by comparing the gradient changes in the horizontal and vertical directions of the images before and after filtering. A higher ESI value (closer to 1) indicates that the filtered image edges are closer to the original image edges, and the edge preservation effect is better. Its definition is:

[0073] ;

[0074] in, These are the pixel values ​​of the SAR image after filtering and denoising. The numerator represents the pixel values ​​of the original noisy SAR image; the numerator represents the sum of the gradient intensities of the denoised image in both the horizontal and vertical directions, and the denominator represents the sum of the gradient intensities of the original image as a whole in both the horizontal and vertical directions.

[0075] The performance comparison results of various filtering algorithms for flat regions I and II and edge region III of SAR images are shown in Tables 1 and 2. The original image is shown below. Figure 2 and Figure 3 As shown, Gaussian filtering, mean filtering, Lee filtering, and the method proposed in this invention have different effects on... Figure 2 and Figure 3 The filtering result of the entire SAR image is as follows Figures 4 to 11 As shown.

[0076] Table 1 - Comparison of SAR1 denoising and edge preservation indices

[0077]

[0078] Table 2 - Comparison of SAR2 denoising and edge preservation indices

[0079]

[0080] As shown in Table 1, the method of the present invention increases the equivalent number of views by 119.62 in Region I and by 139.36 in Region II; Gaussian filtering increases the equivalent number of views by 62.60 in Region I and by 42.23 in Region II; mean filtering increases the equivalent number of views by 148.28 in Region I and by 93.79 in Region II; Lee filtering increases the equivalent number of views by 86.30 in Region I and by 61.00 in Region II; in the edge preservation index of Region III, Gaussian filtering is 0.48, mean filtering is 0.29, Lee filtering is 0.82, and the method of the present invention is 0.83. Observing Table 2, the method of the present invention increases the equivalent number of views by 74.50 after denoising in Region I compared to before denoising, and by 34.06 in Region II; Gaussian filtering increases the equivalent number of views by 24.44 after denoising in Region I compared to before denoising, and by 14.68 in Region II; mean filtering increases the equivalent number of views by 50.12 after denoising in Region I compared to before denoising, and by 27.97 in Region II; Lee filtering increases the equivalent number of views by 35.59 after denoising in Region I compared to before denoising, and by 19.13 in Region II; in the edge preservation index of Region III, Gaussian filtering is 0.40, mean filtering is 0.23, Lee filtering is 0.81, and the method of the present invention is 0.82. From the above results and... Figures 4 to 11 From this, we can see that:

[0081] (1) Regarding noise suppression in smooth regions, this invention exhibits excellent noise suppression performance in most scenarios. Although mean filtering results in a slightly higher ENL due to over-smoothing, it sacrifices image details. In contrast, the ENL index of this invention is significantly higher than that of Gaussian filtering and Lee filtering, proving that this method has good denoising capabilities while preserving details.

[0082] (2) Regarding the ability to preserve image edge details, the filtering method proposed in this invention has the highest edge preservation coefficient (ESI). Combined with... Figure 4 , Figure 5 , Figure 8 and Figure 9 It can be seen that while traditional Gaussian filtering and mean filtering remove some noise, they blur image edges and result in a low ESI index. Combined with... Figure 6 and Figure 10 It can be seen that while traditional Lee filtering preserves edge details, it also retains a significant amount of noise. Combined with... Figure 7 and Figure 11As can be seen, the filtering method proposed in this invention provides clear details at the image edges, has the highest ESI index, can accurately identify and protect edge pixels, and avoids edge blurring while effectively suppressing noise.

[0083] The method described in this application addresses the problem that traditional filtering methods rely solely on the gray-level variance within a local window for decision-making, failing to effectively distinguish between noise and edge textures that belong to the same high-frequency features. This leads to the filter mistakenly smoothing edges and destroying the geometric details of ground features. The structural consistency component constructed by the method described in this invention captures geometric topological features through a gradient direction field. By utilizing the essential difference between the directional randomness of noise and the directional stability of edges, it captures the true geometric features, thereby effectively distinguishing them and avoiding edge blurring and texture loss.

[0084] To address the technical challenge of SAR image multiplicative noise fluctuating drastically with increasing signal strength, leading to a high false alarm rate in strong scattering regions, the method described in this invention introduces an intensity consistency component constructed based on a normalized mean ratio. This component normalizes the intensity differences between the two time phases, mapping ground features of different radiation levels to a unified consistency space. This eliminates the interference of absolute brightness on change detection, effectively suppressing noise fluctuations in bright areas and minimizing the risk of misjudgment of ground features due to noise fluctuations.

[0085] Traditional filtering methods often employ fixed or linearly varying threshold or weighting mechanisms, making it difficult to balance the conflict between robust denoising in flat regions and detail preservation in edge regions. This invention addresses this issue by designing a consistency-guided adaptive weighting mechanism. Based on the structure-intensity joint consistency coefficient, it calculates nonlinear filtering weights, forming a spatially adaptive gating strategy: in regions of high consistency, the weights tend to smooth the reference image, suppressing random noise; in regions of low consistency, the weights tend to resemble the original image, preserving details. This mechanism achieves maximum denoising while retaining edge details, providing a high-quality data foundation for subsequent SAR image applications.

Claims

1. A SAR image filtering method based on joint structural strength consistency, characterized in that, The method includes the following steps: S1: Acquire dual-temporal SAR images of the same surface monitoring area, perform radiometric correction and spatial registration, and form a dual-temporal SAR image pair to be processed; S2: Calculate the structural consistency component and intensity consistency component of the dual-temporal SAR image pair respectively, and construct the joint structural-intensity consistency coefficient using a product strategy; S3: Calculate spatial adaptive filtering weights based on the joint structure-strength consistency coefficient, and automatically adjust the filtering intensity in flat and edge regions; S4: Adaptive weighted fusion filtering guided by consistency is used to output the final denoised image; S1 specifically includes the following steps: Acquire first-phase SAR images of the same surface monitoring area at two different time points. Second-phase SAR images Radiometric correction and spatial geometric registration are performed on the first and second temporal SAR images to align the two images in the pixel coordinate system, forming a pair of dual temporal images to be processed. S2 specifically includes the following steps: S2-1: Constructing structural consistency components using gradient direction fields ; S2-2: Constructing intensity consistency components using normalized mean ratio ; S2-3: A nonlinear product fusion strategy is adopted for the structural consistency components. The intensity consistency component Perform pixel-by-pixel multiplication to generate structure-strength joint consistency coefficients. ; S2-1 specifically includes the following steps: The Sobel edge detection operator was used to analyze the first temporal SAR image. Second-phase SAR images Perform convolution operations to calculate the gradient components of the image in the horizontal direction. and the gradient component in the vertical direction ; For any pixel Its horizontal gradient component With vertical gradient components Calculated separately as follows: ； ； in, The input SAR image to be processed, symbol This represents a two-dimensional discrete convolution operation. The horizontal Sobel convolution kernel is used. The vertical Sobel convolution kernel is defined as follows: ； The above formulas were applied to the first temporal SAR image respectively. Second-phase SAR images To obtain the gradient components of their respective time phases; Calculate the gradient direction angle for each pixel in two time phases using the four-quadrant arctangent function. and The calculation formula is as follows: ； ； in, The gradient component in the horizontal direction of the first phase image; The gradient component in the vertical direction of the first phase image; This represents the gradient component in the horizontal direction of the second phase image; This represents the gradient component in the vertical direction of the second phase image; The set of gradient direction angles calculated for all pixels in the image constitutes the gradient direction fields of the first and second time phases. Based on the obtained dual-phase gradient direction angles, the minimum periodic angle difference is mapped to a structural consistency component using a nonlinear exponential decay strategy. The calculation formula is as follows: ； in, and These are the gradient direction angles for the first and second time phases, respectively. This represents a natural exponential function with base e. The function is used to handle the periodicity of angles. This is a structural sensitivity factor used to adjust the decay rate in response to differences in gradient direction; S2-2 specifically includes the following steps: The local mean of the dual-temporal SAR image pairs within the local sliding window is calculated separately. The local mean is represented by a physical model of the product of the absolute backscattering intensity of ground features and the multiplicative noise term, and its formula is as follows: ； in, This represents the local mean of the dual-temporal SAR image within a local sliding window. Spatial coordinate points The true reflectance of ground features at that location; This represents the mean of the multiplicative speckle noise. The intensity consistency component is constructed by using the product of the local means of the two time phases as the numerator and the sum of squares of the two time phase mean statistics as the denominator, and performing normalized division. The calculation formula is as follows: ； in, and These are the local mean values ​​of the first and second time phases within the sliding window, respectively, both of which are the product of the absolute backscattering intensity of ground objects and the multiplicative noise term; It is a non-zero small constant used for numerical stability control; S2-3 specifically includes the following steps: For the structural consistency component The intensity consistency component Perform pixel-by-pixel multiplication to obtain the structure-strength joint consistency coefficient. The calculation formula is: ； in, For structural consistency components; For strength consistency components; S3 specifically includes the following steps: Based on the structure-strength joint consistency coefficient, a soft-threshold mapping model between smoothness and fidelity is constructed using the nonlinear saturation characteristics of the sigmoid function. The structure-strength joint consistency coefficient is calculated by presetting a soft-threshold center offset and a steepness adjustment factor. The numerical deviation relative to the center offset is amplified using the steepness adjustment factor, and the amplified deviation value is mapped to a spatial adaptive filtering weight taking values ​​in the (0,1) interval using the Sigmoid function. The calculation formula is as follows: ； in, This is the kurtosis factor, used to adjust the slope of the weighting curve and its sensitivity to changes in consistency. This is the center offset of the soft threshold, used to set the critical point for consistency judgment; Represents the natural exponential function; When the structure-strength joint consistency coefficient Greater than the center offset When, weight values ​​tend to be smoothed and denoised; when the joint consensus coefficient Less than center offset At that time, weight values ​​that tend to preserve detail are generated; S4 specifically includes the following steps: Raw SAR images containing edge details Apply a smoothing filter of a preset size to obtain a smoothed and denoised SAR image. , serving as a noise-free reference benchmark for flat areas; Using the spatial adaptive filtering weights In the original SAR image containing edge details Compared with the smoothed and denoised SAR image Pixel-level complementary linear weighted fusion is performed between them, and the final output is a filtered SAR image; The pixel-level complementary linear weighted fusion method includes the following steps: For each pixel location in the image spatial domain, the corresponding smoothed and denoised SAR image is generated. Pixel value multiplied by the spatial adaptive filtering weight and the original SAR image containing edge details corresponding to that location. Pixel value multiplied by 1 minus the spatial adaptive filtering weight The obtained complementarity coefficients are then summed to obtain the final output value for that pixel location. The calculation formula is as follows: ； in, The filtered image is the output. The original SAR image containing edge details, To smooth and denoise the SAR image, These are the spatial adaptive filter weights obtained in S3; When a pixel is located in a high-consistency background area, the weight control fusion result tends to be a smooth and denoised SAR image, thus achieving denoising; when a pixel is located in a low-consistency edge or texture area, the weight control fusion result tends to be the original SAR image containing edge details, thus achieving detail preservation, and finally outputting a filtered SAR image.

Images