Adaptive uniform region extraction phase noise reduction method based on navigational star InBSAR (Interferometric Binary Synthetic Aperture Radar) system

Abstract

The invention discloses a method for extracting a self-adaptive uniform region and reducing phase noise based on a GNSS-based InSAR (Global Navigation Satellite System-Based InSAR) system. The method mainly aims at the problem that a traditional noise reduction algorithm based on a uniform region under a GNSS-based InSAR system is not applicable, and firstly, an improved Canny edge detection algorithm is used for carrying out self-adaptive uniform region extraction on an SAR image according to a resolution unit center and a gradient size. Then, a coherence matrix is introduced, a main phase component is obtained through eigenvalue decomposition, and the phase change of a deformation area is obtained; the method overcomes the limitation of a traditional InSAR imaging algorithm, is suitable for a GNSS-based InSAR system, can reduce phase noise interference, and lays a foundation for deformation monitoring application.

Description

Technical Field

[0001] This invention belongs to the field of bistatic synthetic aperture radar technology, specifically relating to an adaptive uniform region extraction method for reducing phase noise based on a navigation satellite bistatic InSAR system. Background Technology

[0002] Bistatic synthetic aperture radar interferometry (BSAR) based on a global navigation satellite system is a typical application of bistatic SAR. This system uses orbiting navigation satellites as external radiation sources and deploys stationary receivers on the ground to form a bistatic system. Compared to traditional SAR systems, navigation satellite-based interferometry systems offer advantages such as shorter re-orbit periods and lower costs. Furthermore, with over 30 BeiDou navigation satellites in orbit, at any given time, a target area can receive illumination from at least four satellites. This makes it possible to perform three-dimensional deformation measurements by combining the measurement results from different satellites.

[0003] However, achieving accurate deformation measurement using GNSS-based InSAR systems still faces many challenges. First, traditional InSAR imaging algorithms image in the range-Doppler domain, while GNSS-based InSAR systems image on the ground. This makes the resulting image directly affected by the ground, and the focus position shift caused by elevation angle errors cannot be compensated for. This also affects the selection of PS (parameter scatter) points and uniform regions, making traditional InSAR imaging algorithms unsuitable in this context. Second, the inherent low resolution and low signal-to-noise ratio of GNSS-based InSAR systems mean that the resulting image can be considered as composed of resolution cells. When there are several closely spaced strong scattering points, the interferometric phase of different strong scattering points will affect each other, resulting in lower measurement accuracy compared to traditional InSAR systems. Therefore, a method for interferometric phase denoising suitable for GNSS-based InSAR systems is urgently needed. Summary of the Invention

[0004] In view of this, and addressing the shortcomings of existing methods, this invention provides an adaptive uniform region extraction method for reducing phase noise based on GNSS-based InSAR systems. This method primarily addresses the inapplicability of traditional uniform region-based noise reduction algorithms in GNSS-based InSAR systems. It proposes first using an improved Canny edge detection algorithm to adaptively extract uniform regions from the SAR image based on the resolution cell center and gradient magnitude. Then, a coherence matrix is ​​introduced, and the principal phase component is obtained through eigenvalue decomposition to obtain the phase change in the deformed region. This method overcomes the limitations of traditional InSAR imaging algorithms, is applicable to GNSS-based InSAR systems, and can reduce phase noise interference, laying the foundation for deformation monitoring applications.

[0005] The technical solution of this invention is: an adaptive uniform region extraction and phase noise reduction method based on a navigation satellite InBSAR system, comprising:

[0006] Step 1: Extraction of Adaptive Uniform Regions: By comparing actual SAR image data with theoretical resolution cells from satellite trajectory simulation, the location of persistent scatterers (PS) is accurately determined; overlapping resolution cells are adaptively segmented based on gradient magnitude calculation and dynamic threshold; independent support regions are generated by combining connected component analysis to isolate phase interference from neighboring scatterers and improve phase extraction accuracy.

[0007] Step 2, Main Phase Extraction: By constructing a temporal coherence matrix that supports multiple pixels within the region, the matrix is ​​decomposed into eigenvalues ​​to extract the main phase; combined with the phase stability coefficient to screen persistent scatterers (PS), noise and non-main scattering information are suppressed, ultimately improving the deformation monitoring accuracy to the millimeter level.

[0008] Beneficial effects:

[0009] 1. This method proposes adaptive support region segmentation. Addressing the issue of phase distortion in overlapping regions of neighboring scatterers caused by directly using the phase of a single pixel in traditional methods, this method proposes an adaptive edge detection algorithm based on gradient magnitude and dynamic threshold. By segmenting overlapping resolution units, independent support regions are generated, effectively reducing mutual interference between neighboring scatterers and significantly improving the accuracy and reliability of phase extraction.

[0010] 2. This method proposes a temporal support region filtering approach to address the problem of random focus position shifts caused by baseline errors and terrain undulations in time series, which traditional methods struggle to handle effectively. This method innovatively proposes a temporal support region filtering algorithm based on maximum intersection. By selecting stable intersection pixels across multiple temporal support regions, it compensates for spatial shifts and suppresses noise interference, thereby enhancing the consistency of phase in the temporal dimension.

[0011] 3. This method proposes a multi-pixel coherence matrix main phase extraction, addressing the shortcomings of traditional methods that rely on single-pixel phase and are susceptible to noise and scattering interference. By constructing a coherence matrix supporting multiple pixels within the region and utilizing eigenvalue decomposition to extract the main phase, noise and non-main scattering information are effectively suppressed. Experimental results show that this method improves the deformation monitoring accuracy of natural scatterers from 24.2 mm to 13.0 mm using traditional methods, while achieving a monitoring accuracy of 4.0 mm for artificial transponders, realizing millimeter-level high-precision monitoring. Attached Figure Description

[0012] Figure 1 A schematic diagram of the algorithm described in the invention;

[0013] Figure 2 These are SAR images based on a GNSS-based InSAR system.

[0014] Figure 3 ,for Figure 2 The enlarged images of the area within the red box show the results for the transponder and three strong scattering points.

[0015] Figure 4 ,for Figure 3 Results of separating resolution units;

[0016] Figure 5 ,for Figure 4 Phase changes in different resolution units within a region;

[0017] Figure 6 , for eigenvalue influence analysis;

[0018] Figure 7 , , are the extraction results of the principal phase components of the transponder and the natural scattering point, respectively;

[0019] Figure 8 , , represent the accuracy of the transponder and natural scattering point deformation measurements, respectively. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] This invention discloses a method for reducing phase noise in InSAR navigation satellites based on adaptive uniform region extraction. First, a uniform region is extracted by selecting the center of a resolution cell. Then, the principal phase component is extracted and atmospheric phase is compensated. Finally, by extracting the phase information from the resolution cell, the deformation of the PS (parameter scatter) point region can be obtained, enabling deformation monitoring of the selected target region. The algorithm flow of this invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0022] Step 1: Extraction of Adaptive Uniform Regions

[0023] By comparing actual SAR image data with theoretical resolution cells from satellite trajectory simulation, the location of persistent scatterers (PS) is accurately determined; overlapping resolution cells are adaptively segmented based on gradient magnitude calculation and dynamic threshold; and independent support regions are generated by combining connected component analysis to isolate phase interference from neighboring scatterers and improve phase extraction accuracy.

[0024] A SAR image is obtained by imaging the acquired raw data using a back projection algorithm.

[0025]

[0026] Where, σ A (where PSF is the scattering coefficient of target A) A (This represents the point spread function of the resolution unit, P) S It is the location of the navigation satellite transmitter, P A It is the target location, P D It is the location of the receiver.

[0027] Imaging is a super-resolution processing procedure, where each PS point exists as a resolution unit on the image, and each resolution unit contains multiple pixels. The center of the PS point is extracted based on local maxima. PS() consists of a continuous 3-dB range from the center of each PS point, represented as:

[0028] PS(A)={∪B|Im(B)>Im(A)-3} (2)

[0029] The selection of resolution units is only related to the amplitude of the filtered image. When PS points are close to each other, their resolution units will overlap, therefore the resolution units need to be segmented into uniform regions. Here we use an improved Canny edge detection algorithm. The extraction of uniform regions is related to the gradient magnitude, and its calculation formula is:

[0030]

[0031] Where (x, y) represents the coordinates within the image, and the threshold weights are obtained by the frequency of gradient magnitude occurrences. Pixels with smaller gradient magnitudes are more likely to be edges of uniform regions, and the threshold weights are expressed as:

[0032]

[0033] The threshold T is calculated as follows:

[0034] T=ημ(x,y)Im(x,y) (5)

[0035] Where η is the gradient weight, used to adjust the scattering coefficient for different scenes. When M(x,y)<T, the corresponding pixel is used as the boundary, and the connected region containing the PS point is selected as the uniform region.

[0036] Step 2: Principal Phase Extraction

[0037] By constructing a temporal coherence matrix that supports multiple pixels within a region, the matrix is ​​decomposed into eigenvalues ​​to extract the principal phase. By combining the phase stability coefficient to screen persistent scatterers (PS), noise and non-principal scattering information are suppressed, ultimately improving the deformation monitoring accuracy to the millimeter level.

[0038] Before performing atmospheric phase compensation, it is necessary to extract the phase affected by the scattering characteristics of the PS point. The coherence matrix is ​​composed of pixels throughout the uniform region. Assuming all pixels are located within the uniform region, the complex coherence matrix is ​​expressed as follows:

[0039]

[0040] d(PS) = [d1(PS), d2(PS), ..., d K (PS) T (7)

[0041] Where H represents the conjugate transpose operator, and d(PS) represents the regularized complex vector. The pixel phase model extracted from the uniform region is represented as:

[0042]

[0043] Thus, the complex coherence matrix is ​​also expressed as...

[0044]

[0045] in The correlation matrix is ​​a positive semi-definite Hermitian matrix, written here as

[0046] Γ=UΛU -1 =UΛU T (10)

[0047] where Λ=diag(λ1,λ2,…,λ K ( is a K×K (dimensional) non-negative real eigenvalue diagonal matrix. U=[u1,u2,…,u K [(These are the corresponding orthogonal eigenvectors. The eigenvalues ​​are arranged in descending order, and the correlation matrix is ​​then expressed as...]

[0048]

[0049] in This represents the main scattering characteristics of the PS point. The influence of noise and other small scattering points mainly exists in Γ2, Γ3, ..., Γ K The phase of the eigenvector μ1 corresponding to the largest eigenvalue λ1 is the principal phase component, denoted as...

[0050]

[0051] in It is the processed principal phase component.

[0052] After extracting the principal phase component, atmospheric phase compensation is required. This compensation only needs to cover the atmospheric phase error generated during radar signal propagation between the PS point and the receiver. Atmospheric phase exists within the principal phase component. Atmospheric phase compensation is achieved by simultaneously collecting signals from different satellites. This atmospheric phase component is represented as...

[0053]

[0054] Where M is the number of satellite signals processed. This is the m-th principal phase component. After compensating for atmospheric phase and neglecting the influence of noise, the deformation is expressed as...

[0055]

[0056] After completing the above steps, the deformation results of the monitored area can be obtained. To verify the effectiveness of this method, root mean square error (RMSE) is used for analysis. The calculation formula is as follows:

[0057] σ=RMSE(ΔR-ΔR theory (15)

[0058] Where ΔR theory It is the theoretical deformation value.

[0059] The following explains the processing results of the measured data. In the actual measurement, a location in Chongqing was used as the experimental scenario, and the data collection period was 30 days. Deformation monitoring was conducted using seven BeiDou satellites. Taking the results from BeiDou-2-IGSO1 as an example, the imaging results are as follows... Figure 2 and Figure 3As shown. The transponder has a high scattering intensity, while the natural scatterer has a very low scattering intensity, and the scattering intensities of the natural scattering are very close. After extracting the center point of the resolution cell, a range of 3dB is selected as the resolution cell. The resolution cell corresponding to the transponder is complete, but the resolution cells corresponding to the natural scattering are superimposed. If the PS point is selected in the traditional way, multiple resolution cells will overlap, making phase extraction difficult. Therefore, the resolution cells should be separated to obtain a uniform region. In this study, η is 0.4. Gradient magnitudes below T are used as edges, and the uniform region selected after separating the resolution cells is as follows. Figure 4 As shown. A T value of 0.3 allows for the differentiation of mutually influencing resolution cells. Phase changes are as follows... Figure 5 As shown, the maximum phase difference is 0.09 rad and 0.45 rad. The transponder has a high scattering intensity and is less affected by noise and other scattering, thus exhibiting small phase fluctuations. The deformation monitoring results will vary significantly depending on the selected pixel location. Traditional methods use the center phase for processing, which makes accuracy difficult to guarantee. The proposed algorithm is used to extract the principal phase component. Eigenvalue influence analysis is as follows... Figure 6 As shown, the transponder exhibits strong scattering characteristics, while other scattering characteristics are very weak. The largest eigenvalue is much larger than the second largest eigenvalue. For natural scattering points, the non-maximum eigenvalues ​​are also relatively large. This indicates that other scattering information exists besides the primary scattering properties.

[0060] When the atmospheric phase is compensated, the obtained deformation phase is as follows: Figure 7 As shown, the phase difference between the principal phase component and the transponder center is small, which is consistent with the eigenvalue decomposition results. Extracting the principal phase component reduces the phase noise from natural scattering, making the extraction of the deformed phase more accurate.

[0061] The scene remained unchanged during the experiment, therefore the theoretical deformation was zero. The deformation monitoring accuracy is shown in the figure. For the transponder, the effects of other small scattering and noise resulted in the worst accuracy, 15.9 mm. Clearly, the proportion of extracted principal phase components is relatively small. After extracting the principal phase components, the deformation monitoring accuracy is within 4 mm. Compared to directly extracting the phase of pixels, this method obtains the true scattering phase of the scattering more accurately. For natural scattering, the worst deformation monitoring accuracy is 24.2 mm. By using principal phase component extraction, the accuracy is within 13 mm. Simultaneously, the experiment verifies the high-precision millimeter-level deformation monitoring capability based on GNSS-based InBSAR.

[0062] Traditional methods can only consider the phase of a single pixel. Using the phase of a single pixel to represent the phase of the entire resolution cell leads to low phase extraction accuracy and poor deformation monitoring. In this algorithm, the homogeneous region is adaptively selected. A coherence matrix is ​​introduced to represent the scattering characteristics of the entire resolution cell. The principal phase component of the resolution cell is obtained through eigenvalue decomposition. Analysis of this experimental case demonstrates the effectiveness of the proposed method, which effectively solves the problems of focal position offset and phase inconsistency, resulting in more accurate deformation monitoring.

[0063] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.

Claims

1. An adaptive uniform region extraction method for reducing phase noise based on a navigation satellite bistatic InSAR system, characterized in that, include: Step 1: Extraction of adaptive uniform regions: By comparing the actual data of SAR images with the theoretical resolution cells of satellite trajectory simulation, the location of persistent scatterer (PS) is accurately located; The overlapping resolution cells are adaptively segmented based on gradient magnitude calculation and dynamic threshold; independent support regions are generated by combining connected component analysis to isolate phase interference from neighboring scatterers and improve phase extraction accuracy. Step 2, Main Phase Extraction: By constructing a temporal coherence matrix that supports multiple pixels within the region, the matrix is ​​decomposed into eigenvalues ​​to extract the main phase; combined with the phase stability coefficient to screen persistent scatterers (PS), noise and non-main scattering information are suppressed, ultimately improving the deformation monitoring accuracy to the millimeter level.

2. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step one, the back projection algorithm is used to image the acquired raw data to obtain a SAR image; Where, σ A Is the scattering coefficient of target A, PSF A It represents the point spread function of the resolution unit, P. S This is the location of the navigation satellite transmitter, P. A Yes, that is the target location, P D Yes, that's the location of the receiver.

3. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step one, the center of the PS point is extracted based on the local maximum value; PS(·) consists of a continuous 3-dB range from the center of each PS point, represented as: PS(A)={∪B|Im(B)>Im(A)-3} Im(A) represents the pixel intensity of the center point of PS candidate point A, and Im(B) represents the pixel value of the neighboring region B.

4. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step one, the formula for extracting the uniform region is: Where (x, y) represents the coordinates within the image, and the threshold weights are obtained by the frequency of gradient magnitude occurrences; pixels with smaller gradient magnitudes are more likely to be edges of uniform regions, and the threshold weights are expressed as: The threshold T is calculated as follows: T=ημ(x,y)Im(x,y) Where η is the gradient weight, used to adjust the scattering coefficient for different scenes; when M(x,y)<T, the corresponding pixel is used as the boundary, and the connected region containing the PS point is selected as the uniform region.

5. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step two, assuming all pixels are located in a uniform region, the complex coherence matrix is ​​represented as: d(PS)=[d1(PS),d2(PS),…,d K (PS)] T Where d(PS) represents the set of time series vectors of each pixel within the PS point unit, d K (PS) is the time series vector of the Kth pixel in the PS point resolution unit, and Γ(PS) is the coherence matrix of the PS point, which is obtained by averaging the outer product over L days.

6. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step two, the pixel phase model extracted from the uniform region is represented as: in The correlation matrix is ​​a positive semi-definite Hermitian matrix: C = UΛU -1 =UΛU T where Λ=diag(λ1,λ2,…,λ K U is a K×K dimensional non-negative real eigenvalue diagonal matrix; U = [u1, u2, ..., u...] K ] represents the corresponding orthogonal eigenvectors; eigenvalues Sort in descending order, then the correlation matrix is ​​represented as follows: in This represents the main scattering characteristics of the PS point; the influence of noise and other small scattering points mainly exists in Γ2, Γ3, ..., Γ K Yes; the phase of the eigenvector μ1 corresponding to the largest eigenvalue λ1 is the principal phase component, expressed as... in It is the processed principal phase component.

7. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step two, the atmospheric phase is represented as follows: Where M is the number of satellite signals processed; It is the m-th principal phase component; after compensating for atmospheric phase, ignoring the influence of noise, the deformation is expressed as... in, To compensate for the deformation phase after atmospheric phase, λ is the signal wavelength.

8. The adaptive uniform region extraction and phase noise reduction method based on a navigation satellite bistatic InSAR system according to claim 1, characterized in that, In step two, the root mean square error is used for analysis, and the calculation formula is as follows: σ=RMSE(ΔR-ΔR theory ) Where ΔR theory It is the theoretical deformation value.

Images