A method for InSAR multi-baseline phase unwrapping
Through mixed integer quadratic programming and Markov random field model, the InSAR multi-baseline phase detangling technology is improved, solving the problems of low efficiency and noise interference, and achieving more efficient and accurate phase detangling, especially in complex Phase gradient recognition of terrain areas.
Patent Information
- Application Number
- CN202111559202.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-20
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-12-20
Smart Images

Figure CN114371477B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of Interferometric Synthetic Aperture Radar (InSAR), and in particular to an InSAR multi-baseline phase unwrapping method. Background Art
[0002] InSAR processing can extract high-precision, large-scale digital surface models (DSM) of the ground, and plays an irreplaceable role in surveying and mapping applications. Phase unwrapping is a key core step in InSAR processing. Since the interferometric phase obtained through InSAR data is entangled, that is, between -π and π, phase unwrapping operations must be performed to restore the phase to the absolute phase before subsequent ground elevation inversion and other processing can be performed.
[0003] Phase unwrapping can be divided into single-baseline phase unwrapping technology and multi-baseline phase unwrapping technology. Single-baseline phase unwrapping technology performs phase unwrapping through an interference pattern, which mainly includes path tracking-based phase unwrapping methods, norm optimization-based phase unwrapping methods, statistics-based phase unwrapping methods, and network planning-based phase unwrapping methods. The phase unwrapping method based on path tracking uses branch tangents to connect phase residual points so that positive and negative residual points reach a balance, and then avoids branch tangents to integrate the phase to obtain the unwrapped phase. This method is prone to produce phase islands; the phase unwrapping method based on norm optimization establishes a norm (such as 0 norm, 1 norm, 2 norm) minimization objective function of the difference between the unwrapped phase gradient and the wrapped phase gradient, obtains the phase gradient by solving the optimization target, and integrates to obtain the unwrapped phase; the phase unwrapping method based on statistics restores the absolute phase based on the statistical information of the interference phase, and often performs phase unwrapping and filtering at the same time, which is easy to cause the global diffusion problem of errors; the phase unwrapping method based on network planning transforms the residual minimization problem into the minimization problem of network cost flow, and uses the minimum cost flow network planning algorithm for phase unwrapping. It is currently the most effective single baseline phase unwrapping method. However, single baseline phase unwrapping technology has a common shortcoming, that is, the use of the phase continuity assumption makes it difficult to unwrap the phase in areas with large terrain undulations such as mountains and hills. It is difficult to correctly identify the phase gradient in these places where there are phase jumps, resulting in unwrapping errors.
[0004] Multi-baseline phase unwrapping technology performs phase unwrapping through multiple interferograms. By combining the interferograms corresponding to different lengths of baselines, the gradient of the phase jump part can be correctly estimated, thereby improving the phase unwrapping accuracy. Multi-baseline phase unwrapping technology is divided into two steps. The first step is to establish the relationship between the baseline size, phase ambiguity and phase gradient, and use the Chinese remainder theorem to estimate the phase gradient. The second step is to establish the minimization criterion of the unwrapped phase and the entangled unwrapping under the gradient constraint, and solve the objective function to obtain the unwrapped phase. However, the current multi-baseline phase unwrapping technology still has problems such as low algorithm execution efficiency and susceptibility to noise interference. Summary of the invention
[0005] In view of the limitations of the current InSAR multi-baseline phase unwrapping technology, the present invention proposes an InSAR multi-baseline phase unwrapping method. The method uses Mixed Integer Quadratic Programming (MIQP) to improve the execution efficiency of gradient estimation and uses a Markov Random Field (MRF) model to improve the noise resistance of the phase unwrapping method, thereby achieving the improvement of the computational efficiency and accuracy of InSAR multi-baseline unwrapping.
[0006] The technical solution adopted by the present invention is:
[0007] An InSAR multi-baseline phase unwrapping method comprises the following steps:
[0008] (1) Input n interference patterns And the corresponding vertical baselines B1, B2, ..., B n , using the relationship between the interference phase, ambiguity, vertical baseline and terrain height between different interferograms, the 2-norm optimization model objective function f of the phase gradient ambiguity is established;
[0009] (2) The objective function f is converted into a standard quadratic form and the objective function is solved using mixed integer quadratic programming to obtain the phase gradient ambiguities k1, k2, ..., k n ;
[0010] (3) After obtaining the phase gradient ambiguity k1, k2, ..., k n In the case of , a Markov random field energy function E is established which can express both phase smoothness and phase closure.
[0011] (4) Using numerical methods to estimate the weight coefficients of the Markov random field energy function E, where the smoothness weight coefficient s is obtained by maximizing the neighborhood phase, and the closure weight coefficient c is represented by the phase coherence ρ of the interference pattern;
[0012] (5) Substitute the phase gradient ambiguities k1, k2, ..., k solved in the first step into n , transform the energy function E into a standard quadratic expression, use the quadratic programming method to minimize the energy function, and find the unwrapped phase;
[0013] Complete InSAR multi-baseline phase unwrapping.
[0014] Furthermore, the specific method of step (1) is:
[0015] (101) Input n interference patterns, where the winding phase corresponding to the i-th interference pattern is Fuzziness k i , vertical baseline B i The relationship with the terrain height h is as follows:
[0016]
[0017] Where λ represents the wavelength, R represents the distance from the antenna phase center to the ground target point, and θ represents the antenna incident angle;
[0018] Using the equal terrain height h, we obtain the equation:
[0019]
[0020] (102) Using the permutation and combination method, take the 2-norm of the difference between the two groups and then sum them:
[0021]
[0022] Furthermore, the Markov random field energy function E is:
[0023]
[0024] Among them, E(φ i,j ) represents the phase φ at the pixel point (i, j) i,j The energy function consists of two parts, which represent the smoothness and closure of the unwrapped phase, C i,j represents the phase value set adjacent to pixel (i, j), s i,j represents the weight of smoothness, c i,j Represents the weight of closure.
[0025] Furthermore, the specific method of step (5) is:
[0026] (501) Substitute the calculated phase gradient ambiguities k1, k2, ..., k n , removing the complex exponential part of the energy function E, the expanded and simplified expression is:
[0027]
[0028] Among them, s r and c Respectively represent the weight s in the row direction and the column direction, c r and c c Respectively represent the weight of weight c in the row direction and column direction, k r and k c denote the ambiguity of ambiguity k in the row and column directions respectively;
[0029] (502)E(φ i,j ) is a phase point φ in the interference pattern i,j Assuming that the interference pattern size is M×N pixels, all points in the interference pattern are modeled. After summing i and j, we convert and write them into matrix form as follows:
[0030]
[0031] E(Ψ) is the energy function for the entire interference pattern, where Ψ represents a one-dimensional vector obtained by straightening a two-dimensional interference pattern φ by columns;
[0032] φ i,j represents the i-th row and j-th column element of the matrix φ, Ψ t represents the tth element of the vector Ψ;
[0033] φ i,j The coefficient of t The mapping is: t = i + N·j, that is:
[0034] Ψ=[φ 0,0 ,φ 1,0 ,…,φ 1,N -1,…,φ M-1,N-1 ] T
[0035] F represents a first-order coefficient vector, F t represents the tth element of vector F, φ i,j The coefficient of F t The mapping is: t = i + N·j, that is:
[0036]
[0037] H represents the quadratic coefficient matrix, H i,j represents the i-th row and j-th column element of matrix H, φ l,m φ p,q The coefficient of H i,j The mapping is: i = l + M·m, j = p + M·q;
[0038] (503) Energy function The minimization iterative solution is performed to obtain the smoothed unwrapped phase Ψ, and finally Ψ is readjusted to a two-dimensional matrix form φ, which is the obtained unwrapped phase.
[0039] Compared with the background technology, the present invention has the following advantages:
[0040] 1. In view of the low efficiency of using 1-norm optimization to solve the phase gradient ambiguity in conventional InSAR multi-baseline phase unwrapping technology, this paper proposes to use 2-norm optimization to solve the phase gradient, which reduces the time complexity from O(K n(n-1) / 2 ) is reduced to O(n 3 ), which improves the computational efficiency.
[0041] 2. In order to address the inefficiency and error diffusion problems caused by the heuristic global optimization method used in conventional solutions to Markov random field problems, the present invention uses a Markov model to establish the relationship between the unwrapped phase, the wrapped phase and the phase ambiguity, and improves the expression of the Markov random field model so that it can be solved using a quadratic programming method. This improves efficiency on the one hand, and enhances the algorithm's ability to cope with noise on the other.
[0042] 3. The solution algorithms used in the present invention are all implemented using optimization algorithms, the solution results are deterministic and unique, and the execution time is controllable under a given accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 This is the InSAR multi-baseline phase unwrapping flow chart.
[0044] Figure 2 It is a schematic diagram of the Markov random field model representing the interaction between the real phase, entangled phase and unentangled phase of the interference pattern.
[0045] Figure 3 It is a schematic diagram of the neighborhood pixel set of the target pixel. DETAILED DESCRIPTION
[0046] The present invention will be further described below in conjunction with the accompanying drawings.
[0047] An InSAR multi-baseline phase unwrapping method is proposed. The method is based on Markov random fields and mixed integer quadratic programming and can recover the absolute phase from the interferogram entangled phase.
[0048] Assume there are n interference patterns: The size of the interference pattern is M×N, and the corresponding vertical baseline lengths are: B1, B2, ... B n , solve for the unwrapped phase: φ1,φ2,...,φ n.
[0049] like Figure 1 As shown, the steps of this method are:
[0050] (1) According to the InSAR observation geometry model, the winding phase of the i-th interferogram Fuzziness k i , vertical baseline B i The relationship with the terrain height h is as follows:
[0051]
[0052] Where λ represents the wavelength, R represents the distance from the antenna phase center to the ground target point, and θ represents the antenna incident angle. Using the terrain height h, the equation can be obtained:
[0053]
[0054] Here we propose a new 2-norm objective function, using permutations and combinations, taking the 2-norm difference between the two groups and then summing them:
[0055]
[0056] Since there are n interference patterns, we can get A combination of methods.
[0057] (2) The objective function f is converted into a standard quadratic form as follows:
[0058]
[0059] in:
[0060] X=[k1,k2,...,k n ] T ,X∈Z n
[0061] represents the ambiguity vector to be solved, the vector length is n,
[0062]
[0063] represents a diagonal matrix of observed winding phase values,
[0064]
[0065] definition:
[0066] P=4π·sum col (ΦQ)
[0067] sum col(·) represents the column-wise summation of the matrix, and the result is a row vector.
[0068] is a constant that is independent of the unknown number and can be ignored. Therefore, the objective function f(X) can be simplified to:
[0069]
[0070] Solving quadratic programming problems The ambiguity X is obtained, and then the phase gradient is obtained.
[0071] (3) When the phase gradient ambiguity is obtained, the energy function model E of the Markov random field is established, as follows: Figure 2 As shown in Figure 2, the energy function consists of two parts: phase smoothness and phase closure. The phase φ at the pixel point (i, j) i,j The energy function is:
[0072]
[0073] C i,j Represents the set of phase values adjacent to pixel (i, j), such as Figure 3 As shown, s i,j is the weight of smoothness, c i,j is the weight of closure.
[0074] (4) Estimate the weight coefficient of the Markov random field energy function and the smoothness weight coefficient s i,j The closedness weight coefficient c is obtained by maximizing the neighborhood phase. i,j Using the phase coherence representation of the interferogram:
[0075] Smoothness weights i,j Estimation method:
[0076]
[0077] Closure weight c i,j Calculated using the coherence coefficient:
[0078]
[0079] S1 and S2 generate two complex SAR images corresponding to the interferograms, respectively. E(·) represents expectation, and the superscript * represents the conjugate operation.
[0080] (5) Substitute the phase gradient ambiguities k1, k2, ..., k solved in the first step into n , the complex exponential part of the energy function E can be removed, and the expanded and simplified expression is:
[0081]
[0082] Use s respectively r and c Represents the weight of weight s in the row and column directions, with k r and k c represents the ambiguity of ambiguity k in the row and column directions.
[0083] Above E(φ i,j ) is for a phase point φ i,j If all points are considered, let Then after summing i and j, it can be converted and written into matrix form as follows:
[0084]
[0085] The above formula is the overall energy function.
[0086] Ψ represents a one-dimensional vector that straightens a two-dimensional interference pattern, and the vector length is MN, φ i,j To t Mapping: t = i + N·j, that is:
[0087] Ψ=[φ 0,0 ,φ 1,0 ,…,φ 1,N-1 ,…,φ M-1,N-1 ] T
[0088] F represents the first-order coefficient vector, φ i,j The coefficient of F t Mapping: t = i + N·j, that is:
[0089]
[0090] H represents the quadratic coefficient matrix, the matrix size is MN×MN, φ l,m φ p,q The coefficient of H i,j The mapping is: i=l+M·m, j=p+M·q.
[0091] Energy Function After minimizing the iterative solution, the smoothed unwrapped phase Ψ can be obtained.
[0092] (6) Finally, Ψ is resized into a two-dimensional matrix form φ to complete the InSAR multi-baseline phase unwrapping.
[0093] In summary, the present invention uses a mixed integer quadratic programming method to estimate the phase gradient ambiguity of the interferogram, and uses a Markov random field model to solve the unwrapped phase. The present invention proposes to use a 2-norm constraint to establish an optimization model for the phase gradient ambiguity, and converts the problem to be solved into a mixed integer quadratic programming problem; the present invention improves the energy function expression of the Markov random field model, so that the model is converted into a quadratic programming problem to simplify the solution process. Through the method proposed in the present invention, the execution efficiency and solution accuracy of InSAR multi-baseline phase unwrapping can be effectively improved, the solution result has the characteristics of determinism and uniqueness, and the execution time is controllable under a given accuracy.
Claims
1. An InSAR multi-baseline phase unwrapping method, characterized in that: The following steps are involved: (1) Input N interference patterns And the corresponding vertical baselines B1, B2, ..., B n , using the relationship between the interference phase, ambiguity, vertical baseline and terrain height between different interferograms, the 2-norm optimization model objective function f of the phase gradient ambiguity is established; the specific method is: (101) Input n interference patterns, where the winding phase corresponding to the i-th interference pattern is Fuzziness k i , vertical baseline B i The relationship with the terrain height h is as follows: Where λ represents the wavelength, R represents the distance from the antenna phase center to the ground target point, and θ represents the antenna incident angle; Using the equal terrain height h, we obtain the equation: (102) Using the permutation and combination method, take the 2-norm of the difference between the two groups and then sum them: (2) The objective function f is converted into a standard quadratic form and the objective function is solved using mixed integer quadratic programming to obtain the phase gradient ambiguities k1, k2, ..., k n ; (3) After obtaining the phase gradient ambiguity k1, k2, ..., k n In the case of , a Markov random field energy function E is established that can express both phase smoothness and phase closure. The Markov random field energy function E is: Among them, E(φ i,j ) represents the phase φ at the pixel point (i, j) i,j The energy function consists of two parts, which represent the smoothness and closure of the unwrapped phase, C i,j represents the phase value set adjacent to pixel (i, j), s i,j represents the weight of smoothness, c i,j The weight representing closure; (4) Using numerical methods to estimate the weight coefficients of the Markov random field energy function E, where the smoothness weight coefficient s is obtained by maximizing the neighborhood phase, and the closure weight coefficient c is represented by the phase coherence ρ of the interference pattern; (5) Substitute the phase gradient ambiguities k1, k2, ..., k solved in the first step into n , transform the energy function E into a standard quadratic expression, use the quadratic programming method to minimize the energy function, and find the unwrapped phase; the specific method is: (501) Substitute the calculated phase gradient ambiguities k1, k2, ..., h n , removing the complex exponential part of the energy function E, the expanded and simplified expression is: Among them, s r and c Respectively represent the weight s in the row direction and the column direction, c r and c c Respectively represent the weight of weight c in the row direction and column direction, k r and k c denote the ambiguity of ambiguity k in the row and column directions respectively; (502)E(φ i,j ) is a phase point φ in the interference pattern i,j Assuming that the interference pattern size is M×N pixels, all points in the interference pattern are modeled. After summing i and j, we convert and write them into matrix form as follows: E(Ψ) is the energy function for the entire interference pattern, where Ψ represents a one-dimensional vector obtained by straightening a two-dimensional interference pattern φ by columns; φ i,j represents the i-th row and j-th column element of the matrix φ, Ψ t represents the tth element of the vector Ψ; φ i,j The coefficient of t The mapping is: t = i + N ~ j, that is: Ψ=[φ 0,0 ,f 1,0 ,…,f 1,N-1 ,…,f M-1,N-1 ] T F represents a first-order coefficient vector, F t represents the tth element of vector F, φ i,j The coefficient of F t The mapping is: t = i + N·j, that is: H represents the quadratic coefficient matrix, H i,j represents the jth row and jth column element of matrix H, φ l,m φ p,q The coefficient of H i,j The mapping is: i = l + M·m, j = p + M·q; (503) Energy function Perform minimization iteration to obtain the smoothed unwrapped phase Ψ, and finally readjust Ψ to a two-dimensional matrix form φ, where φ is the obtained unwrapped phase; Complete InSAR multi-baseline phase unwrapping.
