Method for obtaining deformation graph covering whole mountainous area through rail ascending / descending InSAR deformation image fusion

Through the deformation intensity value generation feature matrix and deformable window feedback mechanism in InSAR technology, the differences and noise problems of observation results of rising and falling rails are solved, and more accurate surface deformation information fusion and noise removal are achieved, and the surface deformation map is improved. Quality and details are retained.

CN119022776APending Publication Date: 2024-11-26CHINA THREE GORGES CORPORATION +2
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411123105.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The observations of existing InSAR technology under the conditions of rising rails and falling rails reflect differences and complementarity in surface deformation, and the deformation zones and secondary salt and pepper noise are prone to occur during image fusion, resulting in incomplete surface deformation information.

Method used

The characteristic matrix is ​​generated by deformation intensity values ​​to guide pixel-level feature fusion, and a feedback mechanism is implanted in the deformable window. The window structure is adaptively adjusted through noise detection and deformable filters to remove the fused noise and realize the rising/downward track image. High-quality integration.

Benefits of technology

It improves the spatial perception ability of surface deformation information and the accuracy of identifying targets, suppresses salt and pepper noise, maintains image details, reduces single track detection deviation and uncertainty, and provides more comprehensive surface deformation information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119022776A_ABST
    Figure CN119022776A_ABST
Patent Text Reader

Abstract

The invention discloses a method for obtaining a deformation graph covering the whole mountainous area through rail ascending / descending InSAR deformation image fusion. The method comprises the following steps: S1, generating an InSAR image; s2, preprocessing and preliminary fusion; s3, performing an optimization process; s4, evaluating the structure; and S5, checking and outputting. The invention provides a method for obtaining a deformation image covering the whole mountainous area through rail ascending / descending InSAR deformation image fusion, and the method has the following beneficial effects: (1) a pixel-level InSAR rail ascending / descending fusion algorithm based on the guidance of a fusion decision matrix can utilize a deformation value image as a fusion guidance basis to realize the fusion of homologous different-rail InSAR data; the InSAR result space perception capability and the target identification accuracy are improved; and (2) a feedback mechanism is implanted in a deformable window median filtering technology to perform directional denoising, so that the phenomena of image definition reduction and surface feature edge blurring caused by a conventional filtering and denoising method are solved to a certain extent, and details of the fused image are kept while salt-and-pepper noise is suppressed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of surface deformation detection, and in particular to a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images. Background Art

[0002] Differential Interferometry Synthetic Aperture Radar (DInSAR, also known as InSAR) is a high-precision microwave observation technology for the Earth. It uses repeated observations of two coherent SAR (Synthetic Aperture Radar) images of the same area to form interference deformation, and obtains surface deformation information based on the phase information of image pixels. Due to the side-view imaging mode, the sensitivity of SAR satellites in observing surface deformation in the flight direction is reduced, which is closely related to geometric distortions such as shadows and overlaps. As a result, the InSAR observation results under ascending and descending orbit conditions are different and complementary from the actual situation in reflecting the intensity and range of InSAR deformation. Therefore, how to effectively fuse the ascending and descending orbit observation results is the key to reconstructing the complete deformation characteristics of the surface. With the help of ascending and descending orbit image fusion, the unique information between different detection objects is organically combined, and by removing secondary salt and pepper noise, enhancing ground feature and filling missing information, it is expected to reduce the deviation and uncertainty of single orbit detection of target objects, and provide more comprehensive surface deformation information. Based on the InSAR observation results, in order to achieve high-quality fusion of InSAR ascending and descending orbit observation results, it is necessary to solve two main problems: unclear deformation area and secondary salt and pepper noise in the image fusion process. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images, using deformation intensity values ​​to generate a feature matrix to guide pixel-level feature fusion, and implanting a feedback mechanism in the deformable window median filter technology to directionally remove fusion noise, thereby achieving high-quality fusion of ascending / descending orbit images.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images, comprising the following steps: S1, generate InSAR images; S2, preprocessing and preliminary fusion; S3, optimization process; S4, evaluation structure; S5. Check and output.

[0005] Preferably, the step S1 is specifically as follows: In the InSAR image generation stage, the Stacking-InSAR technology is used to generate the InSAR deformation value map and deformation visibility map of the study area.

[0006] Preferably, the step S2 is specifically as follows: The target area is cropped, data is calibrated, cleaned and checked on the ascending and descending orbit images. The best of human vision, orbit information and image content is taken as the main image from the InSAR ascending and descending orbit visible image, and the deformation decision matrix is ​​obtained using the corresponding deformation value map. The pixel-level fusion algorithm is then used to guide the transformation of pixel values ​​in the three-band InSAR visible image to complete the preliminary fusion.

[0007] Preferably, the step S3 is specifically as follows: First, the noise of the preliminary fusion result of the visible image is detected pixel by pixel and divided into two categories: noise and valid information. Then, the deformable filter is used to adaptively adjust the window structure to obtain the median value of the pixels around the noise, and the noise points in the image are replaced to restore the fused image under low-density impulse noise.

[0008] Preferably, the step S4 is specifically as follows: The image is segmented according to the superpixel principle, and then multiple indicators are evaluated in RGB and YUV space respectively. That is, the expression advantage of the RGB color system in three-dimensional space and the advantage of the HSV color model that can extract the image chromaticity, saturation and brightness separately are combined to determine the quality of image fusion.

[0009] Preferably, in step S5, after checking the parameter information, the fused image is finally geocoded to generate a fused deformation map that is easy to analyze by GIS.

[0010] Preferably, the step S2 is specifically as follows: First, according to the geographical location of the study area, the primary and secondary image pairs are selected from the InSAR deformation value map under the conditions of small geometric distortion and high imaging quality. In this process, it is necessary to clean the pixels without information in the InSAR observation results. The actual rule is to use the ascending and descending orbit observation images and the corresponding NoData matrix to generate intensity images S and M, that is: ; ; In the above formula, is the deformation value map of the main image, It is the deformation value diagram from the image. is a data matrix with no values, is the main InSAR image, It is from InSAR images; On this basis, it is necessary to select a more reliable main image from the basic image pair as a reference. This requires first calculating the decision matrix of the ascending and descending orbit surface deformation numerical map, and obtaining the eigenvalues ​​and eigenvectors to obtain the deformation decision matrix, that is: ; Where the decision matrix is ​​A, S and M represent the main and auxiliary intensity maps respectively, and the quantities that are irrelevant to i and j are regarded as constants, and we get: ; in The calculation method is to first and Take the absolute value to obtain the unsigned value of the surface deformation; then select the largest absolute value and assign it to , indicating the absolute strongest deformation value of the pixel; Y coefficient matrix, converting the specific data value into a label , corresponding to three types: unsigned, main image and auxiliary image; Since the selection of pixel-level fusion rules is very flexible, a fusion rule based on the decision matrix is ​​proposed using the value of the feature matrix: ; Where F reflects the change of pixel values ​​before and after fusion. If the quantities in the same band after fusion that are independent of i and j are regarded as constants, it can be further expressed as: ; in The specific calculation method is to select the pixels at the corresponding positions of the main and auxiliary images according to the data representation type in the decision matrix A, and then multiply them using the coefficient matrix. This fusion rule guides the pixel-level transformation and fusion of the three-band InSAR images through the decision matrix generated by the intensity value.

[0011] Preferably, the step S3 is as follows: The data processing process requires comparing the grayscale values ​​of all pixels in the image one by one through noise detection, and comparing the grayscale values ​​with or The same points are marked as suspected salt and pepper noise points, and other pixels are marked as signal points, that is: ; In the formula, To represent the gray value of the pixel located in the i-th row and column in the surface deformation image, Smax and Smin represent the maximum and minimum gray values ​​of all pixels in the image. represents the suspected salt and pepper noise point matrix; The deformable filter can adaptively adjust the window structure. The idea is that the kernel window (Kernel) is replaced by an irregular shape instead of a traditional regular rectangular window. The kernel function is expressed as , where d is a vector representing the deformation parameters; it is represented as a matrix K, where the elements , , ; Therefore, its column vector contains the required kernels, each with different deformation parameters; through singular value decomposition, the total least squares problem of finding the reference kernel and the interpolation function is solved, so that for a given number of J reference kernels, the mean square error (MSE) of the linear combination is minimized: ; when Keep every Assume that ,exist hour: ; In the formula , is the LS optimal interpolation function, is the convolution kernel size recommended by LS optimal interpolation. In many practical cases, there is a lot of overlap between kernels of different deformations.

[0012] The present invention provides a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images, which has the following beneficial effects: (1) Based on the pixel-level InSAR ascending and descending orbit fusion algorithm guided by the fusion decision matrix, the deformation value map can be used as the fusion guidance basis to realize the fusion of InSAR data from the same source but different orbits, thereby improving the spatial perception ability of InSAR results and the accuracy of target recognition; (2) A feedback mechanism is implanted in the deformable window median filter technology to perform directional denoising. This solves the problem of decreased image clarity and blurred edges of objects caused by previous filtering denoising methods to a certain extent, suppresses salt and pepper noise, and maintains the details of the fused image. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a preliminary fusion strategy diagram of the ascending / descending orbit of the present invention; Figure 3 It is a deformable median filter optimization fusion image strategy diagram under the decision feedback of the present invention; Figure 4This is a comparison diagram of the fusion results of the present invention; Figure 5 These are the results of different denoising methods under different amounts of salt and pepper noise in the present invention. DETAILED DESCRIPTION

[0014] like Figure 1 As shown, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images includes the following steps: S1, generate InSAR images; S2, preprocessing and preliminary fusion; S3, optimization process; S4, evaluation structure; S5. Check and output.

[0015] Preferably, the step S1 is specifically as follows: In the InSAR image generation stage, the Stacking-InSAR technology is used to generate the InSAR deformation value map and deformation visibility map of the study area.

[0016] Preferably, the step S2 is specifically as follows: The target area is cropped, data is calibrated, cleaned and checked on the ascending and descending orbit images. The best of human vision, orbit information and image content is taken as the main image from the InSAR ascending and descending orbit visible image, and the deformation decision matrix is ​​obtained using the corresponding deformation value map. The pixel-level fusion algorithm is then used to guide the transformation of pixel values ​​in the three-band InSAR visible image to complete the preliminary fusion.

[0017] Preferably, the step S3 is specifically as follows: First, the noise of the preliminary fusion result of the visible image is detected pixel by pixel and divided into two categories: noise and valid information. Then, the deformable filter is used to adaptively adjust the window structure to obtain the median value of the pixels around the noise, and the noise points in the image are replaced to restore the fused image under low-density impulse noise.

[0018] Preferably, the step S4 is specifically as follows: The image is segmented according to the superpixel principle, and then multiple indicators are evaluated in RGB and YUV space respectively. That is, the expression advantage of the RGB color system in three-dimensional space and the advantage of the HSV color model that can extract the image chromaticity, saturation and brightness separately are combined to determine the quality of image fusion.

[0019] Preferably, in step S5, after checking the parameter information, the fused image is finally geocoded to generate a fused deformation map that is easy to analyze by GIS.

[0020] Preferably, Figure 2 As shown, the step S2 is specifically as follows: First, according to the geographical location of the study area, the primary and secondary image pairs are selected from the InSAR deformation value map under the conditions of small geometric distortion and high imaging quality. In this process, it is necessary to clean the pixels without information in the InSAR observation results. The actual rule is to use the ascending and descending orbit observation images and the corresponding NoData matrix to generate intensity images S and M, that is: ; ; In the above formula, is the deformation value map of the main image, It is the deformation value diagram from the image. is a data matrix with no values, is the main InSAR image, It is from InSAR images; On this basis, it is necessary to select a more reliable main image from the basic image pair as a reference. This requires first calculating the decision matrix of the ascending and descending orbit surface deformation numerical map, and obtaining the eigenvalues ​​and eigenvectors to obtain the deformation decision matrix, that is: ; Where the decision matrix is ​​A, S and M represent the main and auxiliary intensity maps respectively, and the quantities that are irrelevant to i and j are regarded as constants, and we get: ; in The calculation method is to first and Take the absolute value to obtain the unsigned value of the surface deformation; then select the largest absolute value and assign it to , indicating the absolute strongest deformation value of the pixel; Y coefficient matrix, converting the specific data value into a label , corresponding to three types: unsigned, main image and auxiliary image; Since the selection of pixel-level fusion rules is very flexible, a fusion rule based on the decision matrix is ​​proposed using the value of the feature matrix: ; Where F reflects the change of pixel values ​​before and after fusion. If the quantities in the same band after fusion that are independent of i and j are regarded as constants, it can be further expressed as: ; in The specific calculation method is to select the pixels at the corresponding positions of the main and auxiliary images according to the data representation type in the decision matrix A, and then multiply them using the coefficient matrix. This fusion rule guides the pixel-level transformation and fusion of the three-band InSAR images through the decision matrix generated by the intensity value.

[0021] Preferably, Figure 3 As shown, the step S3 is specifically as follows: The data processing process requires comparing the grayscale values ​​of all pixels in the image one by one through noise detection, and comparing the grayscale values ​​with or The same points are marked as suspected salt and pepper noise points, and other pixels are marked as signal points, that is: ; In the formula, To represent the gray value of the pixel located in the i-th row and column in the surface deformation image, Smax and Smin represent the maximum and minimum gray values ​​of all pixels in the image. represents the suspected salt and pepper noise point matrix; The deformable filter can adaptively adjust the window structure. The idea is that the kernel window (Kernel) is replaced by an irregular shape instead of a traditional regular rectangular window. The kernel function is expressed as , where d is a vector representing the deformation parameters; it is represented as a matrix K, where the elements , , ; Therefore, its column vector contains the required kernels, each with different deformation parameters; through singular value decomposition, the total least squares problem of finding the reference kernel and the interpolation function is solved, so that for a given number of J reference kernels, the mean square error (MSE) of the linear combination is minimized: ; when Keep every Assume that ,exist hour: ; In the formula , is the LS optimal interpolation function, is the convolution kernel size recommended by LS optimal interpolation. In many practical cases, there is a lot of overlap between kernels of different deformations.

[0022] Taking a landslide-prone deformation area in the upper reaches of the Jinsha River as an example, S1: Pixel-level InSAR image fusion (DMPI) guided by decision matrix. This study randomly selected an area from the experimental area for effect comparison (see Appendix Figure 4), and the deformed body features in the ascending orbit observation results, descending orbit observation results and fusion results were statistically analyzed. After data fusion, the number of deformed patches identified increased by 34% compared with the descending orbit observation results, the total area increased by 61%, and the InSAR information density increased by 2.95%. The number of identified patches increased by 8 compared with the ascending orbit images, the total area increased by 30.16×105 km², the InSAR information density increased by 5.38%, and the remote sensing image texture increased to 4.6×10-2. .

[0023] S2: Deformable median filter optimized fusion image (DMFS) under decision feedback. This study constructs a simulation comparison experiment to verify the versatility and robustness of this denoising algorithm. The comparison methods include mean denoising method (Mean), median denoising method (Median), Gaussian denoising method (Gaussian), bilateral denoising method (Bilateral), non-local mean denoising method (FastNiMeans) and the deformable median filter (DMFS) under the feedback strategy proposed in this patent. Experimental results (see Appendix Figure 5) shows that the noise is reduced in the images after using the four denoising methods. The visual effect of the image after Gaussian filtering denoising is blurred, and artifacts appear at the edge of the deformed body; bilateral filtering will make the local area too smooth when processing large noise, and lose the image ground object details; the denoising effect of the image after non-local filtering is obvious, but fuzzy features will appear in some areas; after being processed by the method proposed in this patent, the edge contour of the image is clearer and the recovery of feature information is improved. This study selects edge preservation index (EPI), structural similarity (SSIM) and peak signal to noise ratio (PSNR) to evaluate the denoising effect. The closer the index value is to the value of the original Ori image, the better the quality of the denoised image. After mean filtering denoising, the overall blurriness of the image is greatly reduced. EPI is the maximum value of 0.743 when σ=0.001, and the PSNR index is always the minimum value. The resulting image after median filtering denoising can remove noise and improve clarity, but too smooth will lead to a reduction in the amount of ground object information. When σ=0.1, the maximum SSIM is 12.505, and the visual effect of c3 in the figure is better than other processing results. The results of Gaussian filtering show that the image clarity decreases, the boundary of the deformed body is blurred, and it is difficult to directly distinguish mountains and rivers. The PSNR value is better than that of median filtering and mean filtering. The image clarity after bilateral denoising is good, but the denoising effect is insufficient within the geometric deformation area, and the gradient edge information is discontinuous. EPI is the most obvious, and the maximum change of SSIM is 0.561. The image details after FastNiMeans denoising are well preserved. When σ=0.1, the maximum PSNR is 12.505, but it is too smooth, resulting in the loss of content information and terrain fluctuation change information. The overall clarity of the image after DMFS denoising is higher than that of other methods, and the texture of the ground objects is obvious, which is conducive to the intuitive interpretation of active landslides, and most of the evaluation indicators are better than other methods. Among them, the PSNR and SSIM values ​​change the most, which are 0.75 and 20.005 respectively, which is consistent with the subjective evaluation. Under different noise levels, compared with mean, median, Gaussian, bilateral and FastNiMeans, the denoising algorithm proposed in this patent can effectively remove SPN in the image and retain the original feature information to a large extent, solving the problems of decreased image clarity and blurred feature edges in previous denoising methods to a certain extent.

[0024] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending InSAR deformation images. It is characterized in that The following steps are involved: S1, generate InSAR images; S2, preprocessing and preliminary fusion; S3, optimization process; S4, evaluation structure; S5. Check and output.

2. According to claim 1, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending InSAR deformation images, It is characterized in that The step S1 is specifically as follows: In the InSAR image generation stage, the Stacking-InSAR technology is used to generate the InSAR deformation value map and deformation visibility map of the study area.

3. According to claim 1, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images, It is characterized in that The step S2 is specifically as follows: The target area is cropped, data is calibrated, cleaned and checked on the ascending and descending orbit images. The best of human vision, orbit information and image content is taken as the main image from the InSAR ascending and descending orbit visible image, and the deformation decision matrix is ​​obtained using the corresponding deformation value map. The pixel-level fusion algorithm is then used to guide the transformation of pixel values ​​in the three-band InSAR visible image to complete the preliminary fusion.

4. According to claim 1, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending orbit InSAR deformation images, It is characterized in that The step S3 is specifically as follows: First, the noise of the preliminary fusion result of the visible image is detected pixel by pixel and divided into two categories: noise and valid information. Then, the deformable filter is used to adaptively adjust the window structure to obtain the median value of the pixels around the noise, and the noise points in the image are replaced to restore the fused image under low-density impulse noise.

5. According to claim 1, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending InSAR deformation images, It is characterized in that The step S4 is specifically as follows: The image is segmented according to the superpixel principle, and then multiple indicators are evaluated in RGB and YUV space respectively. That is, the expression advantage of the RGB color system in three-dimensional space and the advantage of the HSV color model that can extract the image chromaticity, saturation and brightness separately are combined to determine the quality of image fusion.

6. According to claim 1, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending InSAR deformation images, It is characterized in that In step S5, after checking the parameter information, the fused image is finally geocoded to generate a fused deformation map that is easy to analyze by GIS.

7. According to claim 3, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending InSAR deformation images, It is characterized in that The step S2 is specifically as follows: First, according to the geographical location of the study area, the primary and secondary image pairs are selected from the InSAR deformation value map under the conditions of small geometric distortion and high imaging quality. In this process, it is necessary to clean the pixels without information in the InSAR observation results. The actual rule is to use the ascending and descending orbit observation images and the corresponding NoData matrix to generate intensity images S and M, that is: ; ; In the above formula, is the deformation value map of the main image, It is the deformation value diagram from the image. is a data matrix with no values, is the main InSAR image, It is from InSAR images; On this basis, it is necessary to select a more reliable main image from the basic image pair as a reference. This requires first calculating the decision matrix of the ascending and descending orbit surface deformation value map, and obtaining the eigenvalues ​​and eigenvectors to obtain the deformation decision matrix, that is: ; Where the decision matrix is ​​A, S and M represent the main and auxiliary intensity maps respectively, and the quantities that are irrelevant to i and j are regarded as constants, and we get: ; in The calculation method is to first and Take the absolute value to obtain the unsigned value of the surface deformation; then select the largest absolute value and assign it to , indicating the absolute strongest deformation value of the pixel; Y coefficient matrix, converting the specific data value into a label , corresponding to three types: unsigned, main image and auxiliary image; Since the selection of pixel-level fusion rules is very flexible, a fusion rule based on the decision matrix is ​​proposed using the value of the feature matrix: ; Where F reflects the change of pixel values ​​before and after fusion. If the quantities in the same band after fusion that are independent of i and j are regarded as constants, it can be further expressed as: ; in The specific calculation method is to select the pixels at corresponding positions of the main and auxiliary images according to the data representation type in the decision matrix A, and then multiply them using the coefficient matrix.

8. According to claim 4, a method for obtaining a deformation map covering the entire mountain area by fusing ascending / descending InSAR deformation images, It is characterized in that The specific steps of step S3 are as follows: The data processing process requires comparing the grayscale values ​​of all pixels in the image one by one through noise detection, and comparing the grayscale values ​​with or The same points are marked as suspected salt and pepper noise points, and other pixels are marked as signal points, that is: ; In the formula, To represent the gray value of the pixel located in the i-th row and column in the surface deformation image, Smax and Smin represent the maximum and minimum gray values ​​of all pixels in the image. represents the suspected salt and pepper noise point matrix; The deformable filter can adaptively adjust the window structure. The idea is that the kernel window is replaced by an irregular shape instead of a traditional regular rectangular window. The kernel function is expressed as , where d is a vector representing the deformation parameters; it is represented as a matrix K, where the elements , , ; Therefore, its column vector contains the required kernels, each with different deformation parameters; the total least squares problem of finding the reference kernel and the interpolation function is solved by singular value decomposition, so that the mean square error of the linear combination is minimized for a given number of J reference kernels: ; when Keep every Assume that ,exist hour: ; In the formula , is the LS optimal interpolation function, It is the convolution kernel size recommended by LS optimal interpolation.

Citation Information

Cited By

  • InSAR point target selection method and device based on double-branch fusion

    CN122024011A

  • Mountain area SAR image coherence change detection method and equipment fused with terrain

    CN122265871A