A seasonal coherent curve weighted time series interferometric radar phase connection method

Abstract

The application provides a seasonal coherent curve weighted time series interferometric radar phase connection method, and belongs to the technical field of data processing, and specifically comprises the following steps: acquiring time series SAR single-view complex images of a research area, estimating a sample covariance matrix and a sample coherence matrix; then adaptively identifying seasonal coherent stacks based on the sample coherence matrix, giving seasonal labels to each image, and dividing the coherence matrix into seasonal areas and cross-seasonal areas according to the seasonal labels; fitting the coherence attenuation curves of each area respectively, and constructing a seasonal block prior weight matrix; then fusing the prior weight matrix with a data quality weight matrix based on the coherence amplitude to obtain a final weight matrix; using the final weight matrix to regularize and symmetrize the sample covariance matrix or the coherence matrix; and finally extracting a principal eigenvector through eigenvalue decomposition to restore time series single-scene phases. Through the method, the stability and spatial continuity of phase estimation in a significant seasonal decorrelation area are improved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a method for phase connection of seasonal coherence curve weighted temporal interferometric radar. Background Technology

[0002] Currently, in vegetated areas or mid-to-high latitude regions with significant freeze-thaw cycles, time-series InSAR data are often simultaneously affected by both overall low coherence and seasonal low coherence. Summer vegetation growth, precipitation, and soil moisture changes alter surface scattering characteristics; winter snow cover, freeze-thaw cycles, and changes in surface freezing conditions introduce other changes in scattering mechanisms. These factors result in the long-term coherence matrix not decaying uniformly, but exhibiting a more pronounced seasonal patchy structure: image pairs within the same season (summer-summer or winter-winter) typically have high coherence, while cross-seasonal image pairs (summer-winter, etc.) typically have the lowest coherence. Traditional PS, SBAS, or artificial network construction methods mainly rely on spatiotemporal baseline thresholds and empirically selected interferometer pairs, which can avoid low-coherence connections to some extent, but lack a unified quantitative expression of seasonal coherence structure. When the study area, time span, image gaps, or seasonal transition times change, the thresholds and networks often need to be readjusted, resulting in insufficient method portability and stability.

[0003] Phase connection methods, especially the EMI (Eigenvalue-based Maximum Likelihood Interferometric phase estimator) method based on eigenvalue decomposition, can estimate the phase of a single scene from the complex coherence matrix using the information from the full combination of interferences. This method is highly automated and makes full use of information. Classical EMI has already constructed a basis matrix using coherence amplitudes. However, in scenarios with overall low coherence superimposed with seasonal decoherence, a large number of low-quality connections spanning seasons or long time intervals still participate in the matrix solution. Although these connections may still appear as certain values ​​in the statistical block coherence, their actual phase constraint ability is very weak, easily introducing noise into the EMI eigenvalue decomposition process, leading to problems such as spatial islands, temporal discontinuities, and instability in subsequent unwrapping.

[0004] It is evident that there is an urgent need for a seasonal coherence curve-weighted temporal interferometric radar phase connection method with high connectivity adaptability and stability. Summary of the Invention

[0005] In view of this, the present disclosure provides a seasonal coherence curve weighted temporal interferometric radar phase connection method, which at least partially solves the problems existing in the prior art.

[0006] This disclosure provides a seasonal coherence curve weighted temporal interferometric radar phase connection method, including: Step 1: Acquire time-series SAR single-look complex images of the study area and preprocess them to obtain registered time-series SLC data, thereby estimating the sample covariance matrix and sample coherence matrix. Step 2: Based on the sample coherence matrix, the local coherence time series curves of each time series SAR single-view complex image are statistically analyzed and smoothed. The local minimum value of the local coherence time series curve is identified as the dividing point. Based on this, all time series SAR single-view complex images are adaptively divided into multiple seasonal coherence stacks. Step 3: Based on the seasonal coherence stack, assign a seasonal label to each time series SAR single-look complex image, and divide the off-diagonal elements in the sample coherence matrix into coherent regions in the same season and low coherence regions across seasons according to the seasonal label. Step 4: For each image pair in the same seasonal coherent region and the cross-seasonal low coherence region, the relationship between the coherence amplitude and the time interval of the image pair is statistically analyzed, and the coherence attenuation curve corresponding to each region is fitted. The image pair consists of two time-series SAR single-look complex images. Step 5: Based on the region and time interval of each image pair, obtain typical coherence values ​​from the corresponding coherence attenuation curve, and map the typical coherence values ​​to the preset weight interval to construct the seasonal block prior weight matrix. Step 6: Construct a data quality weight matrix based on the coherence amplitude of the sample coherence matrix, and multiply the seasonal block prior weight matrix with the data quality weight matrix element by element to obtain the final weight matrix; Step 7: Use the final weight matrix to regularize the sample covariance matrix or sample coherence matrix, and then perform symmetry processing on the regularized matrix. Step 8: Perform eigenvalue decomposition on the symmetric matrix, extract its principal eigenvectors, and recover the phase of a single scene in the time series based on the principal eigenvectors to obtain the phase connection result.

[0007] According to a specific implementation of an embodiment of this disclosure, step 2 specifically includes: Step 2.1: An asymmetric sliding window strategy is used to statistically analyze the local coherence of each time-series SAR single-view complex image. Specifically, for the first-stage time-series SAR single-view complex image, subsequent multiple time-series SAR single-view complex images are selected to form adjacent interferometric pairs. For the second-stage time-series SAR single-view complex image, preceding multiple time-series SAR single-view complex images are selected to form adjacent interferometric pairs. For the middle time-series SAR single-view complex image, adjacent multiple time-series SAR single-view complex images are selected to form adjacent interferometric pairs. The mean coherence amplitude of the adjacent interferometric pairs is calculated as the local coherence index of the time-series SAR single-view complex image to obtain the local coherence time-series curve. Step 2.2: Smooth the local coherence time series curve to obtain a smoothed coherence curve; Step 2.3: Identify the local minimum in the smooth coherence curve as the boundary point, and introduce constraint parameters during the identification process. Then, adaptively divide all time-series SAR single-look complex images into several seasonal coherence stacks. The constraint parameters include the minimum time interval constraint between adjacent seasonal coherence stacks and the minimum coherence difference constraint between high coherence and low coherence stages. The minimum time interval constraint is used to avoid misjudging short-term noise fluctuations as boundary points. The minimum coherence difference constraint is adaptively determined according to the coherence level of the study area.

[0008] According to a specific implementation of an embodiment of this disclosure, step 3 specifically includes: Step 3.1: Based on the obtained seasonal coherence stack, assign a seasonal label to each time series SAR single-look complex image. The seasonal label includes a summer label and a winter label. Each time series SAR single-look complex image belonging to the same seasonal coherence stack is assigned the same seasonal label. Step 3.2: Based on the seasonal labels, divide the off-diagonal elements in the sample coherence matrix into summer-summer region, winter-winter region, and summer-winter region. The summer-summer region and the winter-winter region are coherent regions within the same season, while the summer-winter region is a low-coherence region across seasons.

[0009] According to a specific implementation of this disclosure, the coherence attenuation curve is fitted using an exponential attenuation model, the expression of which is:

[0010] in, , Indicates the region type. Indicates summer-summer region. Indicates winter-winter region. Indicates the summer-winter region. , , These are the fitting parameters for the corresponding region. This indicates the time interval between image pairs.

[0011] According to a specific implementation of this disclosure, the preset weight range is [ [1] is used to preserve the connectivity of a time network while suppressing low-quality connections.

[0012] According to a specific implementation of an embodiment of this disclosure, the data quality weight matrix The element is defined as:

[0013] Where i and j are the SAR image time-series indices. Let be the complex coherence value between the i-th and j-th SAR images. These are parameters used to adjust the intensity of high and low coherence differentiation.

[0014] According to a specific implementation of an embodiment of this disclosure, the sample covariance matrix is... When performing regularization, the regularized matrix The element is defined as:

[0015] For the sample coherence matrix When performing regularization, the regularized matrix The element is defined as:

[0016] in, The element in the i-th row and j-th column of the final weight matrix W. Let be the element in the i-th row and j-th column of the sample covariance matrix C.

[0017] According to a specific implementation of this disclosure, the time series single-scene phase is obtained by extracting complex phases from each element of the principal feature vector, and the overall phase constant term is eliminated by fixing the phase of the reference image to zero.

[0018] According to a specific implementation of this disclosure, after step eight, the method further includes: constructing the phase connection results into an interferometric phase sequence and inputting it into a conventional time series InSAR inversion process to obtain the surface deformation time series results.

[0019] The seasonal coherence curve weighted temporal interferometric radar phase connection scheme in this embodiment includes: Step 1, acquiring time-series SAR single-view complex images of the study area and preprocessing them to obtain registered time-series SLC data, thereby estimating the sample covariance matrix and sample coherence matrix; Step 2, based on the sample coherence matrix, statistically smoothing the local coherence temporal curves of each time-series SAR single-view complex image, identifying the local minimum of the local coherence temporal curve as the boundary point, thereby adaptively dividing all time-series SAR single-view complex images into multiple seasonal coherence stacks; Step 3, assigning a seasonal label to each time-series SAR single-view complex image according to the seasonal coherence stack, and dividing the off-diagonal elements in the sample coherence matrix into coherent regions within the same season and low-coherence regions across seasons based on the seasonal label; Step 4, statistically processing the image pairs within each coherent region within the same season and low-coherence regions across seasons. Step 5: Calculate the relationship between the coherence amplitude and the time interval of the image pair, and fit the coherence attenuation curve corresponding to each region. The image pair consists of two time-series SAR single-look complex images. Step 6: Based on the region and time interval of each image pair, obtain typical coherence values ​​from the corresponding coherence attenuation curves, and map these typical coherence values ​​to a preset weight interval to construct a seasonal block prior weight matrix. Step 7: Construct a data quality weight matrix based on the coherence amplitude of the sample coherence matrix, and multiply the seasonal block prior weight matrix element-wise with the data quality weight matrix to obtain the final weight matrix. Step 8: Regularize the sample covariance matrix or sample coherence matrix using the final weight matrix, and perform symmetry processing on the regularized matrix. Step 9: Perform eigenvalue decomposition on the symmetric matrix, extract its principal eigenvectors, and recover the single-view phase of the time series based on the principal eigenvectors to obtain the phase connection result.

[0020] The beneficial effects of the embodiments disclosed herein are as follows: (1) Solve the problem that traditional artificial mesh construction or fixed threshold methods are insufficient in expressing seasonal coherence differences. By identifying seasonal stacks and distinguishing between three types of coherent regions, namely summer-summer, winter-winter and summer-winter, the weight construction can reflect the differences in the true seasonal scattering state.

[0021] (2) To address the problem that classical EMI still absorbs a large amount of low-quality cross-seasonal information under overall low coherence conditions. Based on the basic eigenvalue decomposition framework of EMI, this invention constructs the final weight matrix W by using the seasonal block prior weight matrix P and the data quality weight matrix D, thereby reducing the contribution of long-term or cross-seasonal connections that are close to the decoherence state in the solution.

[0022] (3) This invention addresses the problem that weighting based solely on coherence amplitude is insufficient to reflect seasonal structure. Coherence amplitude can only reflect the observation quality of a single image pair. In contrast, this invention further utilizes three types of seasonal coherence curves to describe the coherence differences between different seasonal combinations within the same time interval, thus enabling the weights to simultaneously convey both data source and seasonal structure meanings.

[0023] (4) To address the problem that excessive weakening of low-coherence connections may lead to insufficient temporal network constraints. This invention limits the seasonal weights to a preset range. In the embodiment, the minimum retention weight is 0.85 and the maximum weight is 1. In other applications, the minimum retention weight can be adjusted within a similar range according to the overall coherence level and network connectivity of the study area, thereby taking into account both noise suppression and the continuity of long-term sequence constraints.

[0024] (5) Improve the temporal stability and spatial continuity of phase connection results in the significant region of seasonal decoherence, and provide more stable phase input for subsequent StaMPS unwrapping and deformation rate estimation. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic flowchart illustrating a seasonal coherence curve weighted temporal interferometric radar phase connection method provided in this embodiment of the disclosure; Figure 2 A schematic diagram illustrating the specific implementation process of a seasonal coherence curve weighted temporal interferometric radar phase connection method provided in this embodiment of the present disclosure; Figure 3 A schematic diagram of the temporal coherence matrix of a Sentinel-1A / B 32 orbital image provided in an embodiment of this disclosure; Figure 4 This disclosure provides an embodiment of the intention to divide the summer and winter stacks of a year according to local valleys of temporal coherence. Figure 5 A schematic diagram of the seasonal block division results of the coherence matrix provided in this embodiment of the disclosure; Figure 6 A schematic diagram illustrating the statistical and fitting results of coherence between the same season and across seasons provided in the embodiments of this disclosure; Figure 7 A schematic diagram of the weight matrix P constructed based on seasonal segmentation and three types of coherence curves provided in this embodiment of the disclosure; Figure 8This is a schematic diagram of the deformation rate comparison results provided in the embodiments of this disclosure, wherein (a) is the traditional EMI on the left, and (b) is the seasonally weighted EMI of this application; Figure 9 This is a schematic diagram showing the spatial distribution comparison results of the deformation rate standard deviation provided in the embodiments of this disclosure, where (a) is the traditional EMI on the left and (b) is the seasonally weighted EMI of this application; Figure 10 This is a statistical diagram illustrating the distribution of the standard deviation of time series deformation provided in an embodiment of this disclosure. Detailed Implementation

[0027] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0028] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0029] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0030] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The illustrations only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0031] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0032] This disclosure provides a seasonal coherence curve weighted temporal interferometric radar phase connection method, which can be applied to the phase connection process of interferometric radar in environmental monitoring scenarios.

[0033] See Figure 1 This is a flowchart illustrating a seasonal coherence curve weighted temporal interferometric radar phase connection method provided in an embodiment of this disclosure. Figure 1 and Figure 2 As shown, the method mainly includes the following steps: Step 1: Acquire time-series SAR single-look complex images of the study area and preprocess them to obtain registered time-series SLC data, thereby estimating the sample covariance matrix and sample coherence matrix. The method proposed in this application belongs to the time series InSAR phase connectivity processing method. Its basic idea is as follows: First, the complex coherence matrix is ​​estimated from the long-term SAR data. Then, seasonal coherence structures are identified from the matrix, and three types of coherence attenuation curves, namely summer-summer, winter-winter, and summer-winter, are fitted. The curve values ​​are then mapped to a seasonal weight matrix P. The seasonal weight matrix P is fused with the data quality weight matrix D to obtain the final weight matrix W. W is then used to perform weighted correction on the phase connectivity matrix, so that the contribution of low-quality seasonal decoherence connectivity in EMI solution is reduced, while connectivity with good coherence in the same season, short time interval, or good coherence is relatively preserved.

[0034] In practice, firstly, long-term SAR single-look complex images covering the study area are acquired, and the number of images is denoted as [missing information]. The acquisition time of each scene's image is recorded as follows:

[0035] The aforementioned SAR images undergo conventional InSAR preprocessing, including orbit correction, master-slave image registration, radiometric correction, removal of flat phase, and removal of topographic phase, to obtain registered time-series SLC data. For the center pixel to be estimated, a set of homogeneous pixels in its neighborhood that meet the homogeneity condition is selected, denoted as:

[0036] in, The number of homogeneous pixels. Indicates the first The corresponding homogeneous pixels Dimensional observation vector.

[0037] Estimating the sample covariance matrix based on homogeneous pixel set :

[0038] in, This represents the complex conjugate transpose. Further normalization of the sample covariance matrix yields the sample coherence matrix. :

[0039] in, Indicates the first Jinghedi Complex coherence values ​​between SAR images This represents the coherence amplitude of the corresponding image pair. This coherence matrix is ​​used to characterize the statistical correlation between time-series SAR images and is the basis for subsequent seasonal block identification and phase connectivity solutions.

[0040] Step 2: Based on the sample coherence matrix, the local coherence time series curves of each time series SAR single-view complex image are statistically analyzed and smoothed. The local minimum value of the local coherence time series curve is identified as the dividing point. Based on this, all time series SAR single-view complex images are adaptively divided into multiple seasonal coherence stacks. In practice, considering that factors such as vegetation growth, precipitation, soil moisture changes, snow cover and freeze-thaw cycles can cause significant seasonal changes in the coherence of SAR images, this application does not adopt a fixed-length sub-stack partitioning method, but instead starts from the real coherence matrix to adaptively identify the seasonal coherence stack.

[0041] First, representative regions of interest are selected within the study area, and the local coherence of each time phase is statistically analyzed based on the coherence matrix obtained in step 1. To avoid boundary attenuation effects at the beginning and end of the sequence due to insufficient available adjacent interferometer pairs, this application employs an asymmetric sliding window statistical strategy. Specifically, for the... For each scene image, select several adjacent scene images to form a fixed number of adjacent interference pairs, and calculate the mean value of the corresponding coherence amplitude to obtain the first scene image. Local coherence indices of landscape images:

[0042] in, The number of adjacent interference pairs included in the statistics. In order to be with the first Image indices are used to link adjacent images in a sequence. For images at the beginning of a sequence, subsequent images are selected to form adjacent interference pairs; for images at the end of a sequence, preceding images are selected to form adjacent interference pairs, thus ensuring that the coherence statistics of different time phases are composed of the same number of valid samples.

[0043] Then, the obtained local coherence sequence Small-scale smoothing is performed to obtain smoothed temporal coherence curves. Since the surface scattering mechanism changes significantly during seasonal transitions, coherence often exhibits local troughs. Therefore, this application identifies the boundary points of seasonal coherence stacks by recognizing local minima in the smoothed coherence curves.

[0044] Two constraint parameters are introduced during the recognition process: 1. Minimum time interval between adjacent seasonal sub-stacks The preferred setting is 90 days to avoid misjudging short-term noise fluctuations as true seasonal boundaries. 2. Minimum coherence difference between the high coherence stage and the low coherence stage This parameter is adaptively determined based on the coherence level of the study area.

[0045] In a preferred embodiment, considering that natural scenes can typically be divided into two main seasonal coherence phases within a year, this application can pre-determine that each year contains two seasonal coherence sub-stacks, and gradually adjust them using an iterative search method. This process continues until the number of identified seasonal sub-stacks meets a preset requirement. This yields several seasonal coherent stacks, each with relatively consistent coherence within its stack, and different stacks corresponding to different seasons or different scattering states.

[0046] Step 3: Based on the seasonal coherence stack, assign a seasonal label to each time series SAR single-look complex image, and divide the off-diagonal elements in the sample coherence matrix into coherent regions in the same season and low coherence regions across seasons according to the seasonal label. In practice, based on the seasonal coherence stack identified in step 2, each SAR image is assigned a seasonal label. Let the first... The seasonal tags for the landscape images are:

[0047] in, This indicates an image that is closer to the summer scattering conditions during summer or spring / autumn. This indicates an image that is closer to the winter scattering state during winter or spring / autumn.

[0048] Based on seasonal labels, the off-diagonal elements of the coherence matrix are divided into three regions: (a) Summer – Summer region, denoted as SS region:

[0049] (b) Winter – Winter Region, denoted as WW region:

[0050] (c) Summer-Winter region, denoted as SW region:

[0051] The three types of regions mentioned above correspond to different seasonal incoherence behaviors. , and These represent the three types of image pair index sets. Typically, the SS region reflects the coherence relationship between images in summer or near-summer conditions; the WW region reflects the coherence relationship between images in winter or near-winter conditions; and the SW region reflects the coherence relationship between cross-seasonal images. Due to the significant differences in scattering mechanisms between cross-seasonal images, the SW region usually has lower coherence and has a greater impact on the spectral structure stability of the phase connectivity matrix.

[0052] It should be noted that this application does not presuppose that winter images necessarily have high coherence. Instead, it models the three types of regions (SS, WW, and SW) separately based on real data. Therefore, for mid-to-high latitude regions with significant snow cover and freeze-thaw cycles, WW regions can be modeled and weakened independently, avoiding the simplistic view of winter images of the same season as stable, strongly connected regions.

[0053] Step 4: For each image pair in the same seasonal coherent region and the cross-seasonal low coherence region, the relationship between the coherence amplitude and the time interval of the image pair is statistically analyzed, and the coherence attenuation curve corresponding to each region is fitted. The image pair consists of two time-series SAR single-look complex images. In practice, for any image pair The time interval is defined as:

[0054] in, The unit is year.

[0055] The relationship between coherence amplitude and time interval in the three regions SS, WW, and SW is statistically analyzed, and the average coherence attenuation functions for the three regions are fitted:

[0056] in, This indicates that the SS region is within a time interval of Typical coherence level at that time; This indicates that the WW region is in a time interval of Typical coherence level at that time; This indicates that the SW region is within a time interval of Typical coherence level at that time.

[0057] The above function can be fitted using an exponential decay model, a logarithmic descent model, or a linear model. When the amount of data is sufficient, the exponential decay model is preferred.

[0058] in, , , , These are the fitting parameters for the corresponding region. When the amount of data is small or the coherent change trend is weak, linear fitting or piecewise averaging can also be used to characterize the coherent change patterns of different seasonal regions.

[0059] Step 5: Based on the region and time interval of each image pair, obtain typical coherence values ​​from the corresponding coherence attenuation curve, and map the typical coherence values ​​to the preset weight interval to construct the seasonal block prior weight matrix. In practice, a weight matrix P is constructed from the seasonal curves. For an image pair (i,j), the corresponding curve value is selected from the three types of curves according to its category. And normalize and map it to a preset weight range. [1], to obtain the seasonal weights This mapping ensures that the closer the curve's coherence level is to a high-quality, same-seasonal connection, the closer the weight is to 1; the closer the curve's coherence level is to a long-term low-coherence or cross-seasonal connection, the closer the weight is to... The mapping method can also employ quantile normalization or other monotonic mapping forms; in one embodiment, Setting the weight to 0.85 and the highest weight to 1 allows the weight matrix to gently weaken low-coherence connections without compromising temporal network connectivity. For the diagonal elements, let... =1; for off-diagonal elements, force... = This is to ensure that the weight matrix is ​​symmetric.

[0060] Step 6: Construct a data quality weight matrix based on the coherence amplitude of the sample coherence matrix, and multiply the seasonal block prior weight matrix with the data quality weight matrix element by element to obtain the final weight matrix; In practice, due to the weight matrix constructed in step 5 This primarily reflects the prior seasonal structure, while the actual observation quality of the coherence matrix elements should still be determined by the actual coherence amplitude. Therefore, this application further constructs a data quality weight matrix. :

[0061] in, The intensity adjustment parameter distinguishes between high and low coherence. When When the value is large, the weight difference between high-coherence image pairs and low-coherence image pairs is amplified; when When the weights are smaller, the weight differences are relatively weakened.

[0062] Seasonal block prior weight matrix With data quality weight matrix Perform the Hadamard product to obtain the final weight matrix. :

[0063] in, This represents the Hadamard product, which is the product of corresponding elements.

[0064] This fusion method enables the final weights to reflect the seasonal block structure and adapt to the actual coherence level, avoiding over-constraint caused by relying solely on seasonal priors and underutilization of seasonal structural information caused by relying solely on the original coherence matrix.

[0065] Step 7: Use the final weight matrix to regularize the sample covariance matrix or sample coherence matrix, and then perform symmetry processing on the regularized matrix. In practice, based on the final weight matrix obtained in step 6 The sample covariance matrix or coherence matrix constructed in step 1 is then regularized.

[0066] When using the coherence matrix When the input is a regularized coherence matrix It can be represented as:

[0067] When the covariance matrix When the input is a regularized covariance matrix It can be represented as:

[0068] To ensure that the matrix satisfies the Hermitian structure required for phase connection solutions, further symmetry processing can be performed:

[0069] or:

[0070] Through the above processing, the spectral structure instability caused by cross-seasonal weak connections, local low coherence noise, and anomalous coherence elements in the original sample matrix is ​​suppressed, while the effective observation constraints required for time series phase connections are still preserved.

[0071] Step 8: Perform eigenvalue decomposition on the symmetric matrix, extract its principal eigenvectors, and recover the phase of a single scene in the time series based on the principal eigenvectors to obtain the phase connection result.

[0072] In practice, after obtaining the regularized covariance matrix or coherence matrix, this application uses the maximum likelihood phase estimation method based on eigenvalue decomposition to recover the phase of a single scene in the time series.

[0073] Regularized covariance matrix For example, we can perform feature decomposition on it:

[0074] in, The eigenvector matrix, Let be the diagonal matrix of eigenvalues. Let the principal eigenvector corresponding to the largest eigenvalue be . ,but Characterize the dominant phase consistency pattern in the regularization matrix.

[0075] Extract single-scene phase estimation results based on the principal feature vector:

[0076] in, This indicates a complex phase extraction operation. The first eigenvector of the main feature vector Each element. To eliminate the overall phase constant term, the phase of the first image or a designated reference image is fixed to zero:

[0077] in, The reference image index is used. The final time-series single-scene phase vector is obtained as follows:

[0078] This phase vector can be used to construct an interferometric phase sequence after phase connection, and then input into a conventional time series InSAR inversion process to obtain the time series results of surface deformation.

[0079] Compared with directly solving the EMI of the original sample matrix, this application introduces seasonal block prior regularization before eigenvalue decomposition, which makes the main eigenvector less affected by cross-seasonal low coherence noise and local anomalous coherence elements, thereby improving the stability of the phase connectivity estimation results.

[0080] The working principle of this application's technical solution lies in the fact that seasonal decoherence is not a random, uniform decay, but rather a low-quality connectivity problem with a clear temporal structure and seasonal differences. While classical EMI utilizes coherence amplitude, it cannot explicitly distinguish weakly constrained connections caused by seasonal scattering mechanism variations. This application constructs a weight matrix P using three types of seasonal coherence curves, ensuring that connections with good coherence within the same season, short time intervals, or long time intervals contribute significantly to the EMI solution, while low-coherence connections across seasons or long time intervals are gently weakened. Since the value of P remains within a finite range close to 1, this method avoids temporal network fragmentation caused by hard thresholding and edge deletion, while simultaneously reducing the cumulative perturbation of a large number of low-quality connections on eigenvalue decomposition, thereby improving the phase connectivity stability in scenarios involving both overall low coherence and seasonal low coherence.

[0081] Compared with existing PS, SBAS, artificial meshing methods and classical EMI phase connection methods, the technical advantages of this application are mainly reflected in the following aspects: (1) It can utilize the seasonal coherence patterns in real data. By fitting coherence decay curves to the three types of regions SS, WW and SW respectively, this method no longer relies solely on fixed time windows or artificial empirical thresholds, but constructs a weight matrix based on the coherence temporal characteristics of the study area itself, adapting to the coherence changes caused by seasonal factors such as vegetation, precipitation, snow cover and freeze-thaw.

[0082] (2) It can reduce the impact of low-quality cross-seasonal connections on the basis of classical EMI. The classical EMI basis matrix has already utilized the coherence amplitude, but it may still absorb a large number of near-incoherent connections under the condition of overall low coherence. This method uses the final weight matrix W to weight and correct the phase connection matrix, thereby reducing the contribution of low-quality connections that cross seasons, long time intervals or rapidly degrade, thus improving the stability of phase estimation.

[0083] (3) The method requires minimal modification and has strong compatibility. This application is based on the basic solution framework of EMI, and only adds the construction steps of seasonal curve weight matrix P, data quality weight matrix D and final weight matrix W before eigenvalue decomposition. It has good compatibility with existing time series processing workflows such as Distributed Scattering SAR and StaMPS, and is easy to embed into existing software workflows.

[0084] (4) Balancing noise suppression and network connectivity. The weight matrix P is restricted to a preset range, which is [0.85,1] in the embodiment. Therefore, low-coherence connections are weakened rather than completely eliminated, which can avoid long-term network disconnection caused by excessive edge deletion, and at the same time reduce the interference of invalid low-coherence information on EMI solution.

[0085] (5) It is beneficial to improve the quality of subsequent unwrapping and deformation inversion. In areas with significant seasonal decoherence, this method can reduce local instability and spatial islanding in the phase connection results, improve the continuity of the deformation rate field and time series deformation results, and provide a more stable data foundation for long-term deformation monitoring in volcanic areas.

[0086] The seasonal coherence curve weighted temporal interferometric radar phase connection method provided in this embodiment estimates the sample coherence matrix of long-term SAR single-look complex images and adaptively identifies seasonal coherence stacks based on this matrix. This ensures that the seasonal division results are derived from the actual scattering state changes in the study area rather than from human empirical thresholds, avoiding the problem of repeated trial and error adjustments required when moving between different study areas using fixed time windows or fixed thresholds in traditional methods. By dividing the off-diagonal elements of the coherence matrix into three regions—summer-summer, winter-winter, and summer-winter—coherence attenuation curves are fitted to each region, and a seasonal block prior weight matrix is ​​constructed. P effectively suppresses the contribution of cross-seasonal low-coherence connections in subsequent eigenvalue decomposition, solving the problem that classical EMI easily absorbs a large amount of cross-seasonal low-quality information under overall low coherence conditions, leading to unstable phase estimation. By fusing the seasonal block prior weight matrix P with the data quality weight matrix D based on coherence amplitude through Hadamard product, the final weight W simultaneously considers both the seasonal structure prior and the observation quality of individual image pairs, avoiding the problem of seasonal patterns being masked when relying solely on coherence amplitude, or ignoring actual data quality differences when relying solely on seasonal prior. The value range of the seasonal block prior weight matrix P is [ [1], in the embodiments Setting the threshold to 0.85 allows for the gentle weakening rather than complete removal of low-quality cross-seasonal connections. This suppresses the perturbation of low-quality connections on eigenvalue decomposition while avoiding the temporal network disconnection problem caused by hard threshold edge deletion. By performing symmetry processing on the regularized matrix before eigenvalue decomposition, the principal eigenvectors can more accurately reflect the dominant phase consistency patterns between time-series images. The recovered single-scene phase is more spatially continuous and temporally stable, providing higher-quality phase input for subsequent unwrapping and deformation inversion. Without changing the basic eigenvalue decomposition framework of classical EMI, only adding seasonal curve fitting and weight matrix construction steps before eigenvalue decomposition, it has good compatibility with existing mainstream time-series InSAR processing workflows and is easy to embed into existing software systems.

[0087] The method disclosed herein will be further described below with reference to a specific embodiment. This invention has been verified using real Sentinel-1A / B satellite time-series SAR data of a volcanic area. The data time range is from June 4, 2015 to December 11, 2021, totaling 153 images. The experiment uses the classical EMI method as the closest comparison method to the prior art, and embeds the seasonal coherence curve weighted EMI method proposed in this application into the same time-series InSAR processing flow to examine its improvement effect on phase connectivity stability and deformation inversion results under seasonal low coherence scenarios.

[0088] To ensure interpretability of the comparison, both sets of experiments used the same SAR data, the same preprocessing procedures, the same homogeneous pixel selection strategy, and the same subsequent unwrapping and deformation inversion parameters; the only difference was whether seasonal weights were used to improve the final weight matrix W during the phase connection stage. Therefore, the differences in the results mainly reflect the impact of seasonal curve weighting on the stability of EMI solution.

[0089] The study area is a volcanic region, which simultaneously exhibits factors such as vegetation cover, seasonal humidity changes, winter snow cover, and freeze-thaw cycles, showing obvious seasonal incoherence characteristics, making it suitable for verifying the method of this invention.

[0090] First, the Sentinel-1A / B data were preprocessed using conventional InSAR, and then the coherence matrix was estimated based on the time-series SLC data. Figure 3 The temporal coherence matrix of T32 orbital imagery of a volcanic region is presented. It can be seen that the overall coherence of this region is low, and there are blocky coherence structures that alternate along the time direction, indicating that coherence is affected not only by time intervals but also by seasonal differences in scattering states.

[0091] like Figure 4 As shown, the entire data stack is then seasonally divided based on the troughs in the local coherence time-series curves, distinguishing between summer / rainy season and winter / dry season conditions throughout the year. This division is used to assign a seasonal label to each image and serves as the basis for subsequent identification of the three coherence regions: SS, WW, and SW. The seasonal block division results of the coherence matrix are shown below. Figure 5 As shown, interference subsets within the same colored grid are considered as coherent cases within the same season. After assigning seasonal labels, the relationship between the coherence of three types of image pairs—summer-summer, winter-winter, and summer-winter—and the time interval was statistically analyzed, and curve fitting was performed. Figure 6The results show that the summer-summer coherence region exhibits the highest coherence; the winter-winter coherence region initially shows the second highest coherence, but gradually degrades and approaches the level of cross-seasonal coherence as the time interval increases; the summer-winter cross-seasonal region shows the lowest coherence. This phenomenon indicates that the low-coherence connectivity in the study area has a clear seasonal structure, making it suitable for constructing a weight matrix using the three types of seasonal coherence curves.

[0092] Based on the three types of seasonal coherence curves, a seasonal weight matrix P is further constructed, such as... Figure 7 As shown. The range of values ​​for this weight matrix is ​​[ [1] In this embodiment The value is set to 0.85, with the highest weight set to 1. This setting is not a hard, fixed parameter, but rather used to gently reduce the contribution of low-coherence connections in experimental data; in other regions or under different data conditions, Adjustments can be made based on overall coherence levels, seasonal differences, and network connectivity. Figure 7 It can be seen that the P matrix still retains the complete temporal network structure, but it assigns a lower contribution to low-reliability connections across seasons and long time intervals, and a higher contribution to connections within the same season and short time intervals, which is consistent with... Figure 6 The revealed seasonal coherence degradation pattern.

[0093] according to Figure 7 The seasonal weight matrix P is shown, and combined with the data quality weight matrix D, the final weight matrix W is obtained; then, the phase connection matrix is ​​weighted and corrected using W to obtain the deformation rate fields corresponding to the traditional EMI and the method of this application, respectively. Figure 8 (a) shows the traditional EMI results, and (b) shows the seasonally weighted EMI results. The comparison demonstrates that the method described in this application can reduce the interference of seasonally low-coherence connections on the phase connection results, making the deformation rate field more spatially continuous and reducing local anomalous islands and unstable regions. Since the two sets of experiments are consistent except for the phase connection matrix, this phenomenon indicates that the seasonal curve weights can improve the stability of EMI in low-coherence regions.

[0094] The standard deviation of the time series deformation results is further statistically analyzed to evaluate the temporal stability of phase connection and subsequent unwrapping. Figure 9 This indicates that traditional EMI is prone to spatial discontinuities in standard deviation and local islanding phenomena in some low-coherence regions, which is usually related to the instability of PL / EMI solution and subsequent unwrapping instability. After adopting the method of this application, the standard deviation distribution of the time series is smoother, which is beneficial for obtaining more stable unwrapping and deformation inversion results through subsequent processing such as StaMPS.

[0095] Figure 10The statistical distribution of the standard deviation of time-series deformation is presented. It can be seen that the method proposed in this application can obtain more usable points and improve overall temporal stability in low-coherence or seasonally decoherent regions. For a small number of highly coherent points or PS points with relatively simple scattering mechanisms, traditional EMI may still have a local advantage; however, the goal of the method proposed in this application is to improve the phase connectivity stability in distributed scatterers and seasonally low-coherence regions. Therefore, it is more in line with the long-term deformation monitoring needs of this volcanic region in terms of point retention, spatial continuity, and temporal stability in low-coherence regions. This result also shows that the method proposed in this application does not simply aim to be superior to EMI in all regions, but rather provides stability enhancement specifically for the technical problem of seasonally low coherence.

[0096] In summary, real-world data experiments demonstrate that the seasonal coherence curve weighted EMI method proposed in this application can reduce the negative impact of cross-seasonal and long-term low-coherence connections by constructing a weight matrix using seasonal coherence curves without altering the classical EMI solution framework. This method starts from the characteristics of the seasonal coherence matrix ( Figure 3 Through seasonal stacking, fitting of three types of decoherent curves, construction of the P matrix, and fusion of the final weight matrix W, the deformation results are ultimately reflected in the improvement of spatial continuity and temporal stability, forming a complete verification chain from physical phenomena and weight construction to inversion effect, which is practically feasible.

[0097] It should be understood that the various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof.

[0098] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for phase connection of seasonal coherence curve weighted temporal interferometric radar, characterized in that, include: Step 1: Acquire time-series SAR single-look complex images of the study area and preprocess them to obtain registered time-series SLC data, thereby estimating the sample covariance matrix and sample coherence matrix. Step 2: Based on the sample coherence matrix, the local coherence time series curves of each time series SAR single-view complex image are statistically analyzed and smoothed. The local minimum value of the local coherence time series curve is identified as the dividing point. Based on this, all time series SAR single-view complex images are adaptively divided into multiple seasonal coherence stacks. Step 3: Based on the seasonal coherence stack, assign a seasonal label to each time series SAR single-look complex image, and divide the off-diagonal elements in the sample coherence matrix into coherent regions in the same season and low coherence regions across seasons according to the seasonal label. Step 4: For each image pair in the same seasonal coherent region and the cross-seasonal low coherence region, the relationship between the coherence amplitude and the time interval of the image pair is statistically analyzed, and the coherence attenuation curve corresponding to each region is fitted. The image pair consists of two time-series SAR single-look complex images. Step 5: Based on the region and time interval of each image pair, obtain typical coherence values ​​from the corresponding coherence attenuation curve, and map the typical coherence values ​​to the preset weight interval to construct the seasonal block prior weight matrix. Step 6: Construct a data quality weight matrix based on the coherence amplitude of the sample coherence matrix, and multiply the seasonal block prior weight matrix with the data quality weight matrix element by element to obtain the final weight matrix; Step 7: Use the final weight matrix to regularize the sample covariance matrix or sample coherence matrix, and then perform symmetry processing on the regularized matrix. Step 8: Perform eigenvalue decomposition on the symmetric matrix, extract its principal eigenvectors, and recover the phase of a single scene in the time series based on the principal eigenvectors to obtain the phase connection result.

2. The method according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: An asymmetric sliding window strategy is used to statistically analyze the local coherence of each time-series SAR single-view complex image. Specifically, for the first-stage time-series SAR single-view complex image, subsequent multiple time-series SAR single-view complex images are selected to form adjacent interferometric pairs. For the second-stage time-series SAR single-view complex image, preceding multiple time-series SAR single-view complex images are selected to form adjacent interferometric pairs. For the middle time-series SAR single-view complex image, adjacent multiple time-series SAR single-view complex images are selected to form adjacent interferometric pairs. The mean coherence amplitude of the adjacent interferometric pairs is calculated as the local coherence index of the time-series SAR single-view complex image to obtain the local coherence time-series curve. Step 2.2: Smooth the local coherence time series curve to obtain a smoothed coherence curve; Step 2.3: Identify the local minimum in the smooth coherence curve as the boundary point, and introduce constraint parameters during the identification process. Then, adaptively divide all time-series SAR single-look complex images into several seasonal coherence stacks. The constraint parameters include the minimum time interval constraint between adjacent seasonal coherence stacks and the minimum coherence difference constraint between high coherence and low coherence stages. The minimum time interval constraint is used to avoid misjudging short-term noise fluctuations as boundary points. The minimum coherence difference constraint is adaptively determined according to the coherence level of the study area.

3. The method according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Based on the obtained seasonal coherence stack, assign a seasonal label to each time series SAR single-look complex image. The seasonal label includes a summer label and a winter label. Each time series SAR single-look complex image belonging to the same seasonal coherence stack is assigned the same seasonal label. Step 3.2: Based on the seasonal labels, divide the off-diagonal elements in the sample coherence matrix into summer-summer region, winter-winter region, and summer-winter region. The summer-summer region and the winter-winter region are coherent regions within the same season, while the summer-winter region is a low-coherence region across seasons.

4. The method according to claim 1, characterized in that, The coherent decay curve is fitted using an exponential decay model, the expression of which is: in, , Indicates the region type. Indicates summer-summer region. Indicates winter-winter region. Indicates the summer-winter region. , , These are the fitting parameters for the corresponding region. This indicates the time interval between image pairs.

5. The method according to claim 1, characterized in that, The preset weight range is [ [1], used to preserve the connectivity of temporal networks while suppressing low-quality connections. As the lower bound of the weights, and .

6. The method according to claim 1, characterized in that, The data quality weight matrix The element is defined as: Where i and j are the SAR image time series indices. Let be the complex coherence value between the i-th and j-th SAR images. These are parameters used to adjust the intensity of high and low coherence differentiation.

7. The method according to claim 1, characterized in that, For the sample covariance matrix When performing regularization, the regularized matrix The element is defined as: For the sample coherence matrix When performing regularization, the regularized matrix The element is defined as: in, The element in the i-th row and j-th column of the final weight matrix W. Let be the element in the i-th row and j-th column of the sample covariance matrix C.

8. The method according to claim 1, characterized in that, The time series single-scene phase is obtained by extracting complex phases from each element of the main feature vector, and the overall phase constant term is eliminated by fixing the phase of the reference image to zero.

9. The method according to any one of claims 1 to 8, characterized in that, Following step eight, the method further includes: constructing the phase connection results into an interferometric phase sequence and inputting it into a conventional time series InSAR inversion process to obtain the time series results of surface deformation.

Images