An adaptive SAR interferogram filtering method based on non-local mean

Abstract

The application discloses a non-local mean-based adaptive SAR interferogram filtering method, and relates to the technical field of synthetic aperture radar. Firstly, two input SAR images are used to generate an interferogram and calculate the correlation coefficient of each pixel to form a correlation coefficient matrix. Phase information of the interferogram is extracted. Pixels that do not need filtering are screened. For pixels that need filtering, different search window radii are determined according to the correlation coefficient values, and the search window radius corresponding to the pixels that do not need filtering is set to 0. For each pixel that needs filtering, the weight of each pixel in the corresponding search window is determined, and the filtering result of each pixel is obtained by using a weighted average method to form a filtered SAR interferogram.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of synthetic aperture radar, and particularly relates to an adaptive SAR interferogram filtering method based on non-local mean. BACKGROUND

[0002] Synthetic aperture radar (SAR) is a kind of high-resolution imaging radar, which has the advantages of all-weather, all-day, and large-scale observation. The synthetic aperture radar interferometry (InSAR) imaging technology uses two SAR images with different observation angles to generate a SAR interferogram, and the elevation information of the observed target is obtained according to the interferometric phase. The imaging mechanism of SAR leads to the existence of many speckle noises in the SAR interferogram, and the existence of the noises will affect the subsequent target observation accuracy. In order to realize high-precision target observation, the SAR interferogram image needs to be filtered.

[0003] The non-local mean filtering method is a kind of filtering method suitable for SAR interferogram. The main idea of this algorithm is to use the high redundancy of natural images and the decorrelation of noises, to find similar image blocks in a certain window range, and to obtain the filtered image block by weighted averaging of these similar blocks. The advantage of the algorithm is that it can better preserve the edge and texture features of the image.

[0004] The non-local mean filtering method currently widely used generally uses a fixed-size window to filter the entire image, but for a SAR interferogram with a large scene range, the observed target features in the scene are quite different, and the use of a fixed filtering window cannot take into account all the targets in the image, and some details may be lost.

[0005] Some improved non-local mean filtering methods currently proposed are almost not specifically for SAR interferogram, and the existing schemes cannot adjust the filtering parameters according to the prior information such as the correlation coefficient and amplitude value of the SAR interferogram itself when processing the SAR interferogram, resulting in poor filtering effect. SUMMARY

[0006] Therefore, the present application provides an adaptive SAR interferogram filtering method based on non-local mean, which can adaptively determine the filtering window according to the correlation coefficient information and amplitude information of the SAR interferogram, so as to realize high-quality SAR interferogram filtering.

[0007] To achieve the above purpose, the technical scheme of the present application is as follows: an adaptive SAR interferogram filtering method based on non-local mean, taking two SAR image complex matrices S1 and S2 as input, and the neighborhood window radius is f; the adaptive SAR interferogram filtering method comprises the following steps:

[0008] Step (1) generates an interferogram S and calculates the correlation coefficient of each pixel using the input two SAR images S1 and S2, and forms a correlation coefficient matrix C.

[0009] Step (2) extracts the phase information of the interferogram S.

[0010] Step (3) screens the pixels that do not need to be filtered according to the correlation coefficient.

[0011] Step (4) extracts the amplitude information S of the interferogram S A , and performs edge extraction according to the amplitude information S A , and the extracted edge contour part of the pixels is the pixel that does not need to be filtered.

[0012] The pixels other than the pixels that do not need to be filtered are the pixels that need to be filtered.

[0013] Step (5) determines different search window radii according to the correlation coefficient values of the pixels that need to be filtered, and sets the search window radius corresponding to the pixels that do not need to be filtered as 0.

[0014] Step (6) selects a pixel that needs to be filtered, determines the weight of each pixel in the search window corresponding to the pixel, and obtains the filtering result of the current pixel by using the weighted average method.

[0015] Step (7) continues to select the next pixel that needs to be filtered, and repeats step (6) until the filtering results of all the pixels that need to be filtered are obtained, and the filtered SAR interferogram is formed, and the output is the complex matrix of the filtered SAR interferogram.

[0016] Further, step (1) generates an interferogram S and calculates the correlation coefficient of each pixel using the input two SAR images S1 and S2, and forms a correlation coefficient matrix C; specifically:

[0017] A d1xd2 rectangular window is constructed in the two images S1 and S2, and the values of d1 and d2 are set by oneself, and the correlation coefficient of each pixel is calculated, and the formula is as follows

[0018]

[0019] Wherein, represents the pixel point information contained in the rectangular window centered on the kth pixel in the mth SAR image, and m takes the value of 1 or 2; represents the pixel point information contained in the rectangular window centered on the kth pixel in the nth SAR image, and n takes the value of 1 or 2, and m and n are not equal; is the conjugate transpose of

[0020] ​The correlation system of each pixel obtained is combined to form a correlation coefficient matrix C.

[0021] Further, step (3) screens the pixels that do not need to be filtered according to the correlation coefficient; specifically, the high-quality PS points are screened out according to the correlation coefficient matrix C calculated in step (1), and the high-quality PS points are set as the pixels that do not need to be filtered, and the high-quality PS points are the pixel points whose correlation coefficients exceed a set threshold.

[0022] Further, the high-quality PS points are the pixel points whose correlation coefficients exceed a set threshold, and the set threshold is 0.85.

[0023] Further, in step (4), the amplitude information S A Edge extraction is performed, and specifically, the edge detection operator used is a log operator.

[0024] Further, in step (5), for the pixels that need to be filtered, different search window radii are determined according to the correlation coefficient values, and specifically, the search window radius is adjusted in a stepwise manner according to the size of the correlation coefficient, and a smaller search window is used in a place where the correlation coefficient is larger.

[0025] Further, the search window radius is adjusted in a stepwise manner according to the size of the correlation coefficient, and a smaller search window is used in a place where the correlation coefficient is larger, and the search window radius corresponding to the pixels whose correlation coefficient values are in the range of [0.75, 0.85] is 5, and the search window radius corresponding to the pixels whose correlation coefficient values are below 0.75 is 9.

[0026] Further, in step (6), a pixel that needs to be filtered is selected, the weight is determined pixel by pixel in the search window corresponding to the pixel, and the filtering result of the current pixel is obtained by using the weighted average method, and specifically,

[0027] For the selected pixel x that needs to be filtered, the search window radius corresponding to the pixel x is f x ; the weight w(x, y) is determined pixel by pixel in the search window Ω x corresponding to the pixel x.

[0028]

[0029] wherein v(x) represents an image block with the pixel x as the center element and with the half side length f as f, and the image block is recorded as a first image block, and f xy ; v(y) represents another image block selected by sliding in the current search window, and the image block is recorded as a second image block, y is the center pixel of the second image block, the first image block and the second image block have the same size; w(x, y) is the weight corresponding to the pixel y; is a normalization coefficient, and y takes the search window Ωx All pixels within; h is the filter parameter;

[0030] The filtering is performed using a weighted average method, and the filtering formula is as follows:

[0031]

[0032] in This represents the filtering result for the first image patch, taking... The filtering result at the center point is used as the filtering result for the pixel x that needs to be filtered;

[0033] Beneficial effects:

[0034] This invention enables adaptive adjustment of filtering parameters based on the characteristics of the original SAR interferogram, so that the filtered SAR interferogram retains detailed information of the target structure, improves the filtering quality, and thus enhances the accuracy of target observation. Attached Figure Description

[0035] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0037] like Figure 1 The process shown in this invention involves user input of two SAR image complex matrices S1 and S2, filtering parameter h, and neighborhood window radius f; the output is the filtered SAR interferogram complex matrix. The specific steps are as follows:

[0038] Step (1) Use the two input SAR images S1 and S2 to generate an interferogram S and calculate the correlation coefficient.

[0039] Specifically, this step is implemented as follows:

[0040] Based on the principle of interferometric correlation, an interferogram S is generated using two input SAR images S1 and S2. A rectangular window of d1×d2 is constructed in each of the two images, and the correlation coefficient for each pixel is calculated using the following formula:

[0041]

[0042] Step (2) Extract the phase information of the SAR interferogram. Specifically, this step is implemented as follows:

[0043] Due to the imaging mechanism of SAR, speckle noise in SAR interferograms follows a multiplicative noise model, which is not conducive to subsequent filtering. However, phase noise has additive characteristics. Therefore, the phase information of the interferogram is directly extracted for subsequent filtering.

[0044] Step (3) Filter out pixels that do not need filtering based on the correlation coefficient. Specifically, this step is implemented as follows:

[0045] SAR interferograms are generated by interfering two SAR images from different observation perspectives, and the correlation coefficient is the standard for judging the target scattering intensity. Areas with very high correlation coefficients can be considered to have very strong observed targets, which are high-quality PS points. Correspondingly, the noise level at these locations is very low, and the target information can be obtained without filtering. Therefore, high-quality PS points are defined as pixels that do not require filtering, and high-quality PS points are pixels with a correlation coefficient exceeding a set threshold. In this embodiment of the invention, the threshold is set to 0.85.

[0046] Step (4) Filter out pixels that do not need filtering based on the amplitude information of the SAR interferogram. Specifically, this step is implemented as follows:

[0047] The amplitude information in SAR interferograms is primarily used to assist in identifying point and line targets. Point and line targets are characterized by their small size and dense structure. These targets can be easily filtered out directly using nonlocal mean filtering. However, due to the nature of SAR interferograms, the amplitude information often contains clear target outlines. Therefore, edge information can be extracted from the amplitude information to assist in identifying smaller point and line targets. During the filtering process, these targets can be filtered out, thus preserving detailed information.

[0048] The specific operation method is to extract the amplitude information S of the complex matrix S of the interferogram. A Edge extraction is performed using the log edge detection operator, and the extracted edge contours do not require filtering.

[0049] Step (5) Search for the radius of the window based on the correlation coefficient. Specifically, this step is implemented as follows:

[0050] Based on the principle of nonlocal mean filtering, the search window radius determines the distance between similar image patches and the image patch to be filtered. A larger search window allows for more image patches from distant regions, resulting in stronger filtering. As analyzed earlier regarding correlation coefficients, point targets with high correlation coefficients exhibit strong scattering and low noise levels. Introducing too many distant image patches can corrupt target information, thus reducing filtering quality. Therefore, the search window is adjusted in a stepwise manner based on the correlation coefficient, using a smaller search window for areas with higher correlation coefficients.

[0051] In this invention, the search window radius for pixels with correlation coefficients in the range of [0.75, 0.85] is set to 5, and the search window radius for pixels with correlation coefficients below 0.75 is set to 9.

[0052] The specific correspondence is as follows:

[0053]

[0054] Step (6) For a selected pixel x that needs filtering, the corresponding search window radius is f. x ; in the search window Ω x Within the range, the weights w(x,y) are determined pixel by pixel.

[0055]

[0056] Where v(x) represents an image patch centered at pixel x with half-side length f, denoted as the first image patch, and f < f xy v(y) represents another image patch selected by sliding within the current search window, denoted as the second image patch, where y is the center pixel of the second image patch, and the first and second image patches have the same size; w(x,y) is the weight corresponding to pixel y; Ω is the normalization coefficient, where y takes the search window Ω. x All pixels within; h is the filter parameter;

[0057] The filtering is performed using a weighted average method, and the filtering formula is as follows:

[0058]

[0059] in This represents the filtering result for the first image patch, taking... The filtering result at the center point is used as the filtering result for the pixel x that needs to be filtered;

[0060] Step (7) Continue to select the next pixel that needs to be filtered, and repeat step (6) until the filtering results of all pixels that need to be filtered are obtained, forming a filtered SAR interferogram, and the output is the complex matrix of the filtered SAR interferogram.

[0061] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive SAR interferogram filtering method based on nonlocal means, characterized in that, Using two complex matrices S1 and S2 from SAR images as input, and a neighborhood window radius of f, the adaptive SAR interferogram filtering method includes the following steps: Step (1) Use the two input SAR images S1 and S2 to generate an interferogram S and calculate the correlation coefficient of each pixel to form a correlation coefficient matrix C; Step (2) Extract the phase information of the interferogram S; Step (3) Filter out pixels that do not need to be filtered based on the correlation coefficient; Step (4) Extract the amplitude information S of the interferogram S. A According to the amplitude information S A Edge extraction is performed, and the extracted edge contour pixels are those that do not require filtering. Pixels other than those that do not need filtering are the pixels that need filtering; Step (5) For pixels that need to be filtered, determine different search window radii based on their correlation coefficient values, and set the search window radius for pixels that do not need to be filtered to 0. Step (6) Select a pixel that needs to be filtered, determine the weight of each pixel in its corresponding search window, and obtain the filtering result of the current pixel by weighted average. Step (7) Continue to select the next pixel that needs to be filtered, and repeat step (6) until the filtering results of all pixels that need to be filtered are obtained, forming a filtered SAR interferogram, and the output is the complex matrix of the filtered SAR interferogram.

2. The adaptive SAR interferogram filtering method based on nonlocal means as described in claim 1, characterized in that, Step (1) uses the two input SAR images S1 and S2 to generate an interferogram S and the correlation coefficient of each pixel to form a correlation coefficient matrix C; specifically: Construct rectangular windows of size d1×d2 in both images S1 and S2, where the values ​​of d1 and d2 are set manually. Calculate the correlation coefficient for each pixel using the following formula. in, This represents the pixel information contained in the rectangular window centered on the k-th pixel in the m-th SAR image, where m takes the value 1 or 2; This represents the pixel information contained in the rectangular window centered at the k-th pixel in the n-th SAR image, where n takes the value 1 or 2, and m and n are not equal; W k n* is The conjugate transpose of; The correlation system of each obtained pixel is combined to form a correlation coefficient matrix C.

3. The adaptive SAR interferogram filtering method based on nonlocal means as described in claim 1, characterized in that, Step (3) involves filtering pixels that do not require filtering based on the correlation coefficient; specifically: Based on the correlation coefficient matrix C obtained in step (1), high-quality PS points are selected. High-quality PS points are defined as pixels that do not need to be filtered, and high-quality PS points are pixels whose correlation coefficient exceeds a set threshold.

4. The adaptive SAR interferogram filtering method based on nonlocal means as described in claim 3, characterized in that, The high-quality PS points are pixels with a correlation coefficient exceeding a set threshold, where the set threshold is 0.

85.

5. The adaptive SAR interferogram filtering method based on nonlocal means as described in any one of claims 1 to 4, characterized in that, In step (4), based on the amplitude information S A Edge extraction is performed; specifically, the edge detection operator used is the log operator.

6. The adaptive SAR interferogram filtering method based on nonlocal means as described in any one of claims 1 to 4, characterized in that, In step (5), for the pixels that need to be filtered, different search window radii are determined according to their correlation coefficient values. Specifically, the search window radius is adjusted in a stepwise manner according to the magnitude of the correlation coefficient, and a smaller search window is used where the correlation coefficient is larger.

7. The adaptive SAR interferogram filtering method based on nonlocal means as described in claim 6, characterized in that, The search window radius is adjusted in a stepwise manner based on the correlation coefficient. A smaller search window is used for pixels with a higher correlation coefficient. Specifically, the search window radius is set to 5 for pixels with a correlation coefficient between [0.75, 0.85], and to 9 for pixels with a correlation coefficient below 0.

75.

8. The adaptive SAR interferogram filtering method based on nonlocal means as described in claim 6, characterized in that, Step (6) involves selecting a pixel to be filtered, determining the weight of each pixel within its corresponding search window, and obtaining the filtering result of the current pixel using a weighted average method. Specifically: For a selected pixel x that needs filtering, the corresponding search window radius is f. x ; in the search window Ω x Within the range, the weights w(x,y) are determined pixel by pixel. Where v(x) represents an image patch centered at pixel x with half-side length f, denoted as the first image patch, and f < f xy v(y) represents another image patch selected by sliding within the current search window, denoted as the second image patch, where y is the center pixel of the second image patch, and the first and second image patches have the same size; w(x,y) is the weight corresponding to pixel y; Ω is the normalization coefficient, where y takes the search window Ω. x All pixels within; h is the filter parameter; The filtering is performed using a weighted average method, and the filtering formula is as follows: in This represents the filtering result for the first image patch, taking... The filtering result at the center point is used as the filtering result for the pixel x that needs to be filtered.

Images