GB-SAR real-time deformation monitoring method and device based on DPS-InSAR and computer equipment

Abstract

The invention discloses a GB-SAR real-time deformation monitoring method and device based on a DPS-InSAR and computer equipment. Performing time-sharing updating and incremental calculation on continuous GB-SAR observation data through sliding window interference processing; selecting a PS target point set based on the amplitude deviation index and the time sequence average coherence coefficient; an effective scattering area is determined in the initial SLC image set, a homogeneous pixel set is identified in the effective scattering area, and then a DS target point set is determined in combination with the number of homogeneous pixels and posterior coherence; merging the PS target point set and the DS target point set, and determining an interferogram needing phase unwrapping; a PS target point is used as a space reference field, the high-coherence reference characteristic of the PS target point is used for guiding DS target point phase path selection, and unwrapping is carried out based on a weighted phase unwrapping model; and performing atmospheric phase correction on the unwrapped phase diagram to obtain a phase diagram after the atmospheric phase is eliminated. According to the invention, minute-level high-frequency observation data of a low-coherence area can be subjected to efficient processing and real-time deformation inversion.

Description

Technical Field

[0001] This application relates to the field of surface deformation monitoring technology, and in particular to a GB-SAR real-time deformation monitoring method, device and computer equipment based on DPS-InSAR. Background Technology

[0002] Currently, surface deformation monitoring has become a key support for geological disaster early warning and risk assessment. Ground-based synthetic aperture radar (GB-SAR), as an active remote sensing technology with all-weather observation capabilities, leverages its high spatiotemporal resolution (meter-level spatial resolution and minute-level temporal resolution) and short-range observation advantages to monitor deformation in key areas such as landslides, mine slopes, and dams. Currently, GB-SAR deformation monitoring mainly relies on permanent scatterer interferometry (PS-InSAR) technology. This technology, by selecting targets with stable backscattering characteristics in a time series (such as buildings and exposed rocks) for interferometry processing, can achieve millimeter-level deformation monitoring, and its technical system is relatively mature. However, in typical geological disaster areas such as mine slopes and landslide areas, the surface is often covered by exposed soil or shallow vegetation, resulting in significant fluctuations in scatterer signals over time and low coherence. This leads to a limited number of PS points, making it difficult to meet the needs of large-scale, high-density deformation inversion. To address the challenges of monitoring low-coherence regions, distributed scatterer interferometry (DS-InSAR) technology has gradually been developed. This technology mainly extracts phase information based on statistically homogeneous sets of pixels.

[0003] However, existing DS-InSAR research mainly focuses on spaceborne platforms with long revisit cycles (days to weeks). Research on real-time processing for the minute-level high-frequency observations and complex dynamic scenes unique to GB-SAR systems remains scarce. Although some studies have explored methods for selecting DS points in GB-SAR, these efforts are mostly based on offline datasets with limited image quantities, failing to fully consider the real-time processing requirements of continuous GB-SAR observation modes. Furthermore, most DS-InSAR processing frameworks still employ offline batch processing, which prevents deformation information from being updated in real-time as data is continuously acquired, resulting in significant lag in analysis results. Additionally, offline batch processing statically selects DS points based on the entire time series, making it difficult to identify and remove scatterers that gradually become ineffective during monitoring, thus affecting the accuracy and reliability of deformation inversion. Summary of the Invention

[0004] Therefore, embodiments of this application provide a GB-SAR real-time deformation monitoring method based on DPS-InSAR, which can significantly improve processing efficiency while ensuring monitoring accuracy, realize high-precision and real-time monitoring of surface deformation under ground-based SAR platform, and provide technical support for rapid identification and early warning of geological disasters.

[0005] Firstly, this application provides a GB-SAR real-time deformation monitoring method based on DPS-InSAR.

[0006] This application is achieved through the following technical solution: A real-time deformation monitoring method based on DPS-InSAR and GB-SAR includes: S1: Obtain a preset number of SLC image sets, perform joint processing on the SLC image sets, and generate the cumulative deformation of the first time segment; S2: Based on the time baseline minimization criterion, a master image is selected from the SLC image set, and an interferogram stack is generated by performing complex conjugate multiplication with the other slave images using the master image as a reference. S3: Based on the preset amplitude deviation index threshold and time-series average coherence coefficient threshold, select a PS target point set from the SLC image set; S4: Calculate the average amplitude matrix of the SLC image set in the time dimension, and determine the effective scattering area based on the preset amplitude mask threshold. Within the effective scattering area, use the confidence interval hypothesis testing algorithm to identify the homogeneous pixel set, and use the maximum likelihood estimation eigenvalue decomposition algorithm to optimize the phase of the homogeneous pixel set to obtain the optimized phase result. Combine the homogeneous pixel number threshold and the posterior coherence threshold to determine the DS target point set. S5: Combine the PS target point set and the DS target point set into a high coherence target point set, calculate the pixel difference between adjacent SLC images based on the interferogram stack, compare the pixel difference with the preset judgment conditions, and determine the interferogram that needs to be unwrapped in phase. S6: Using the PS target point as the spatial reference field, search for K nearest PS target points in the effective neighborhood of each DS target point to be unwrapped, and unwrap it based on the weighted phase unwrapping model to obtain the unwrapped phase map; S7: Perform atmospheric phase correction on the unwrapped phase map to obtain a phase map after removing the atmospheric phase; S8: When M new images are acquired, the previous SLC image set is updated according to the sliding window with a fixed step size to form a new SLC image set. Then, return to execute S1 and repeat steps S2 to S7 using the new SLC image set to achieve real-time updating of deformation detection results.

[0007] In a preferred embodiment of this application, the preset number of SLC image sets is 30 images. When M new images are acquired, the previous SLC image set is updated using a sliding window with a fixed step size to form a new SLC image set, including: When three new images are acquired, the three earliest images in the current SLC image set are removed, and the three new images are added to form a new SLC image set.

[0008] In a preferred embodiment of this application, the method can be further configured to select a PS target point set from the SLC image set based on a preset amplitude deviation index threshold and a time-series average coherence coefficient threshold, including: Calculate the amplitude deviation index and temporal average coherence coefficient of each pixel in the SLC image; A pixel is selected as a PS target point when its amplitude deviation index is less than the preset ADI threshold and its temporal average coherence coefficient is greater than the preset coherence coefficient threshold.

[0009] In a preferred embodiment of this application, the step of calculating the average amplitude matrix of the SLC image set in the time dimension and determining the effective scattering region based on a preset amplitude mask threshold, wherein the effective scattering region includes: The average amplitude matrix of each SLC image is calculated in the time dimension. The average amplitude value is calculated based on the average amplitude matrix, and a binary mask matrix is ​​constructed. Pixel regions with amplitude greater than d / 3 are identified as effective scattering regions.

[0010] In a preferred embodiment of this application, step S4, which uses a confidence interval hypothesis testing algorithm to identify a set of homogeneous pixels, may further include: The confidence interval boundary parameters of each homogeneous pixel set are cached, and the confidence interval boundary parameters are used to adaptively update the homogeneous pixel set during the real-time update phase.

[0011] In a preferred embodiment of this application, the step of using the PS target point as a spatial reference field, searching for K nearest PS target points within the effective neighborhood of each DS target point to be unwrapped, and performing unwrapping based on a weighted phase unwrapping model includes: Establish a KD-Tree spatial index structure between DS target points and PS target points; For each DS target point to be unwrapped, search for the K nearest unwrapped PS target points in its effective neighborhood; Based on the normalized distance weights between the DS target point and the K PS target points, a weighted phase unwrapping model is constructed to calculate the unwrapping phase of the DS target point. The effective neighborhood is the region centered on the DS target point and at a distance less than the preset spatial resolution.

[0012] In a preferred example of this application, it may further be configured to include: When there are no unwrapped PS target points in the effective neighborhood of the DS target point, the unwrapping phase is calculated by inverse distance weighted interpolation based on the unwrapped DS target points around it.

[0013] In a preferred example of this application, it may further be configured to include: During the real-time update phase, based on the confidence interval boundary parameters of the homogeneous pixel set determined in the previous round, pixels with unstable scattering characteristics caused by environmental changes are detected and located. The HTCI algorithm is used to update the local homogeneous pixel set in the region where the unstable pixel is located; Based on the preset expected number of PS target points, the amplitude deviation index threshold and coherence coefficient threshold are dynamically adjusted to maintain the stability of the number of PS target points. Based on the preset expected number of DS target points, the posterior coherence threshold is dynamically adjusted to maintain the stability of the number of DS target points.

[0014] Secondly, this application provides a GB-SAR real-time deformation monitoring device based on DPS-InSAR.

[0015] This application is achieved through the following technical solution: A real-time deformation monitoring device based on DPS-InSAR and GB-SAR, used to execute the real-time deformation monitoring method based on DPS-InSAR and GB-SAR described in the first aspect above, comprising: The data preprocessing module is used to acquire a preset number of SLC image sets, perform joint processing on the SLC image sets, and generate the cumulative deformation of the first time segment; it is also used to update the previous round of SLC image sets according to a sliding window with a fixed step size when M new images are acquired, form a new SLC image set, return to execute S1, and repeat steps S2 to S7 using the new SLC image set to realize the real-time update of deformation detection results. The interferogram generation module is used to select a master image from the SLC image set based on the time baseline minimization criterion, and generate an interferogram stack by performing complex conjugate multiplication with the master image and other slave images based on the master image. The PS target point selection module is used to select a set of PS target points from the SLC image set based on a preset amplitude deviation index threshold and a time-series average coherence coefficient threshold. The DS target point selection module is used to calculate the average amplitude matrix of the SLC image set in the time dimension, and determine the effective scattering area based on a preset amplitude mask threshold. Within the effective scattering area, a confidence interval hypothesis testing algorithm is used to identify a set of homogeneous pixels, and a maximum likelihood estimation eigenvalue decomposition algorithm is used to optimize the phase of the set of homogeneous pixels to obtain the optimized phase result. The DS target point set is determined by combining the homogeneous pixel number threshold and the posterior coherence threshold. The interferogram filtering module is used to combine the PS target point set and the DS target point set into a high coherence target point set, calculate the pixel difference between adjacent SLC images based on the interferogram stack, compare the pixel difference with preset judgment conditions, and determine the interferograms that need phase unwrapping. The unwrapping module is used to take the PS target point as the spatial reference field, search for K nearest PS target points in the effective neighborhood of each DS target point to be unwrapped, and perform unwrapping based on the weighted phase unwrapping model to obtain the unwrapped phase map. The phase correction module is used to perform atmospheric phase correction on the unwrapped phase map to obtain a phase map after removing the atmospheric phase.

[0016] Thirdly, this application is achieved through the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described GB-SAR real-time deformation monitoring methods based on DPS-InSAR.

[0017] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following: (1) This application proposes a sliding window interferometric processing architecture with adjustable step size, which can perform time-division updating and incremental calculation of continuous GB-SAR observation data. While ensuring the accuracy of deformation monitoring, this method achieves minute-level deformation field updates, which significantly improves the real-time performance and overall computational efficiency of the system.

[0018] (2) This application constructs a phase unwrapping model based on the spatial constraints of permanent scatterers (PS). By leveraging the high coherence reference characteristics of PS points, the phase path selection of distributed scatterers (DS) points is guided, effectively solving the problem of unstable phase entanglement in low signal-to-noise ratio regions, thereby achieving high-precision phase reconstruction under complex terrain conditions.

[0019] (3) In view of the characteristics of GB-SAR high-frequency observation, an adaptive update algorithm for the scatterer set was designed. Combined with temporal consistency and correlation detection, the automatic maintenance and elimination of scatterers is realized, ensuring the continuity of monitoring results in time and the stability in space. Attached Figure Description

[0020] Figure 1 A flowchart illustrating a GB-SAR real-time deformation monitoring method based on DPS-InSAR provided in an embodiment of this application; Figure 2 A schematic diagram of a sliding window provided in an embodiment of this application; Figure 3 A schematic diagram of the monitoring blind zone and phase wrapping of adjacent frames provided in an embodiment of this application; Figure 4 A schematic diagram of the deformation monitoring results of PS-InSAR and DPS-InSAR for 10 consecutive hours; Figure 5 A schematic diagram showing the deformation monitoring results of PS-InSAR and DPS-InSAR for 20 consecutive hours; Figure 6 A schematic diagram of the deformation monitoring results of PS-InSAR and DPS-InSAR for 30 consecutive hours; Figure 7 A schematic diagram showing the deformation monitoring results of PS-InSAR and DPS-InSAR for 40 consecutive hours; Figure 8 A schematic diagram showing the deformation monitoring results of PS-InSAR and DPS-InSAR for 52 consecutive hours; Figure 9 This is a schematic diagram of the temporal amplitude variation curve of pixels in the construction area; Figure 10 A schematic diagram of the scatter plot of the target points of PS and DS; Figure 11 A schematic diagram illustrating the phase change between adjacent SLC images before and after optimization; Figure 12 This is a schematic diagram illustrating the time-series characteristics of meteorological parameters. Figure 13 This is a schematic diagram illustrating the change in the number of PS / DS target points. Detailed Implementation

[0021] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] Furthermore, the term "and / or" in this application is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this application, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0024] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0025] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0026] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0027] See Figure 1 As shown, the first exemplary embodiment of this application provides a real-time deformation monitoring method for GB-SAR based on DPS-InSAR, the specific steps of which include: S1: Obtain a preset number of SLC image sets, perform joint processing on the SLC image sets, and generate the cumulative deformation of the first time segment; S2: Based on the time baseline minimization criterion, the main image is selected from the SLC image set, and an interferogram stack is generated by performing complex conjugate multiplication with the other secondary images using the main image as the reference. S3: Select a set of PS target points from the SLC image set based on the preset amplitude deviation index threshold and time-series average coherence coefficient threshold; S4: Calculate the average amplitude matrix of the SLC image set in the time dimension, and determine the effective scattering area based on the preset amplitude mask threshold. Within the effective scattering area, use the confidence interval hypothesis testing algorithm to identify the homogeneous pixel set, and use the maximum likelihood estimation eigenvalue decomposition algorithm to optimize the phase of the homogeneous pixel set to obtain the optimized phase result. Combine the homogeneous pixel number threshold and the posterior coherence threshold to determine the DS target point set. S5: Combine the PS target point set and the DS target point set into a high coherence target point set, and calculate the pixel difference between adjacent SLC images based on the interferogram stack. Compare the pixel difference with the preset judgment conditions to determine the interferogram that needs to be unwrapped in phase. S6: Using the PS target point as the spatial reference field, search for K nearest PS target points in the effective neighborhood of each DS target point to be unwrapped, and unwrap it based on the weighted phase unwrapping model to obtain the unwrapped phase map; S7: Perform atmospheric phase correction on the unwrapped phase map to obtain a phase map after removing the atmospheric phase. S8: When M new images are acquired, the previous SLC image set is updated according to the sliding window with a fixed step size to form a new SLC image set. Then, return to execute S1 and repeat steps S2 to S7 using the new SLC image set to achieve real-time updating of deformation detection results.

[0028] In a preferred embodiment, the preset number of SLC image sets is 30 images. When M new images are acquired, the previous SLC image set is updated according to a sliding window with a fixed step size to form a new SLC image set, including: When three new images are acquired, the three earliest images in the current SLC image set are removed, and the three new images are added to form a new SLC image set.

[0029] Specifically, the system first initializes. When 30 SLC images have been acquired, the first 30 SLC images are jointly processed to generate the cumulative deformation field for the first time series. Then, in the real-time update phase, for every 3 new SLC images, the system executes a "discard three, add three" window iteration strategy. This means discarding the earliest 3 SLC images in the window (e.g., images 1-3 for the first real-time update) and adding the latest 3 acquired SLC images (e.g., images 31-33 for the first real-time update), forming a new processing window containing images 4-33. Based on this, the system inherits the phase calculation results from the previous window and superimposes the incremental deformation field from the newly acquired images 31-33 to generate the updated cumulative deformation field for the entire sequence of images 1-33. As monitoring continues, the processing window dynamically slides in steps of 3 images, for example, images 7-36, images 10-39, etc., ultimately forming a generalized window model from (n-29) to n. See [link to relevant documentation]. Figure 2 As shown.

[0030] In the GB-SAR real-time dynamic monitoring system, DPS-InSAR technology achieves continuous updating of deformation time-series information by introducing a sliding window mechanism. Considering the strong dependence of distributed scatterer (DS) technology on the statistical stability of multi-temporal images, it is usually necessary to process 25 images jointly to ensure the robustness and reliability of the phase estimation results. However, real-time monitoring scenarios require the system to output deformation results within a short time interval to meet the timeliness requirements of engineering applications. Therefore, this application proposes a sliding window strategy based on step size control to balance deformation estimation accuracy and monitoring real-time performance, setting the step size to 3 images. It should be noted that this step size is determined based on the acquisition and processing time-series characteristics of the GB-SAR system. The acquisition time of a single SLC image is approximately 3.5 minutes, so the time span of 3 images is approximately 10.5 minutes; at the same time, a single deformation update processing cycle takes approximately 6 minutes. Using a 3-image step size not only ensures that the system can complete a complete deformation update cycle within 10.5 minutes, but also effectively balances the accuracy and real-time requirements of deformation estimation.

[0031] After acquiring 30 SLC images, the data from the initialization phase is first preprocessed. Based on the time baseline minimization criterion, a master image is selected from the SLC image set. Using the master image as a reference, an interferogram stack is generated by performing complex conjugate multiplication with the remaining slave images. The specific steps include: Based on the time baseline minimization criterion, the 15th SLC image was selected as the master image. Using the master image as the reference, 30 pairs of interferometric phase maps were generated through complex conjugate multiplication. The master image was multiplied by the other 29 slave images respectively to generate 29 pairs of master-slave interferograms, and a pair of conjugate multiplication results between the master image and itself was also included. , in, Indicates the phase of a single master image. This represents the phase of the i-th image. Indicates the interference phase.

[0032] Simultaneously, a temporal intensity sequence set was constructed: ,in Indicates amplitude information.

[0033] In some embodiments, selecting a PS target point set from the SLC image set based on a preset amplitude deviation index threshold and a time-series average coherence coefficient threshold includes: Calculate the amplitude deviation index and temporal average coherence coefficient of each pixel in the SLC image; A pixel is selected as a PS target point when its amplitude deviation index is less than the preset ADI threshold and its temporal average coherence coefficient is greater than the preset coherence coefficient threshold.

[0034] Specifically, a two-parameter threshold criterion based on physical scattering characteristics is used to identify and select permanent scatterers. The Amplitude Dispersion Index (ADI) is utilized. The temporal scattering stability of the target point is characterized, and then the temporal mean coherence is used. The stability and quality of the pixel interference phase are evaluated to assess the phase quality. A pixel is identified as a PS target point only when both its amplitude deviation index and temporal average coherence coefficient simultaneously meet the corresponding threshold conditions.

[0035] In some embodiments, the average amplitude matrix of the SLC image set is calculated in the time dimension, and the effective scattering region is determined based on a preset amplitude mask threshold, including: The average amplitude matrix of each SLC image is calculated in the time dimension. The average amplitude value is calculated based on the average amplitude matrix, and a binary mask matrix is ​​constructed. Pixel regions with amplitude greater than d / 3 are identified as effective scattering regions.

[0036] For distributed scatterers, this application employs a DS target point selection strategy combining HTCI verification and EMI algorithm. Accurate selection of specific distributed scatterers is a crucial step in DPS-InSAR data processing, with the main challenge being the weakening effect of spatiotemporal decorrelation on the interferometric phase quality, thus affecting the accuracy of deformation information acquisition. To address this issue, this application adopts a collaborative framework of Statistically Homogeneous Pixel (SHP) identification and phase optimization to achieve efficient extraction and utilization of DS target points.

[0037] Considering the presence of numerous purely noisy regions in GB-SAR echo data, directly performing homogeneous pixel identification and phase optimization across the entire image would not only incur enormous computational costs but may also introduce invalid or interfering information. To improve processing efficiency, this application introduces a masking strategy based on amplitude features to eliminate noisy pixels. In practical implementation, an average amplitude matrix is ​​calculated on the original SLC image over time. Based on this average amplitude matrix, the average amplitude value d of the entire image is calculated, and a binary mask matrix is ​​constructed. Regions with amplitudes greater than d / 3 are considered effective scattering areas, and homogeneous pixel identification and phase optimization are performed within these effective scattering areas.

[0038] Specifically, the confidence interval hypothesis testing algorithm uses interval estimation to select homogeneous pixels in the effective scattering region. The specific form of the interval estimation is expressed as: , In the formula, x represents the estimated sample statistic, and P represents the interval where the sample statistic x falls. The probability, and The boundaries of the confidence interval. , where is the confidence level.

[0039] In the HTCI algorithm, the selection of homogeneous pixels is based on the confidence interval determination of the mean pixel scattering characteristics, which can be expressed mathematically as follows: , Where N represents the stack size, For standard gamma distribution in quantiles at the location, The estimated value of the reference pixel is derived from the initial homogeneous pixel set. The temporal intensity of all pixels is obtained by averaging. Let be the temporal intensity mean of the pixel to be detected, when Within the confidence interval, it represents that pixel. Reference pixel Homogeneous pixels.

[0040] When performing phase optimization, this application adopts the eigenvalue decomposition maximum likelihood estimation algorithm (EMI). This algorithm is theoretically close to the Cramér-Rao low bound and has high computational efficiency.

[0041] After phase optimization, combined with the number of homogeneous pixels Threshold of posterior coherence To determine the final DS target point. The determination method has 95% of the PS target points. Value less than .

[0042] After determining the PS and DS target points, they are merged and unified as a set of highly coherent target points. The PS target points satisfy the phase consistency theorem. By utilizing the redundant information of the entire interferogram stack for phase linking, the phase of the DS target points can also meet the consistency requirements. The interferometric phase of adjacent SLC images can be calculated using the following formula: , Where o is the primary image number, and m and n are the secondary image numbers.

[0043] The above calculation method can shorten the time baseline between the main image and the auxiliary image, thereby reducing the probability of phase entanglement and the impact of atmospheric phase delay.

[0044] Traditional phase unwrapping methods in spaceborne InSAR systems, such as the branching method or least-squares fitting method, typically rely on processing large, continuous spatial regions or smooth terrain areas. These methods are highly applicable to spaceborne platforms because their interferograms cover a wide area, have long time baselines, and exhibit relatively uniform spatial coherence. However, in GB-SAR systems, due to severe irregular occlusion and uneven spatial distribution of scatterers, traditional phase unwrapping methods perform poorly in terms of unwrapping accuracy and stability. Therefore, when performing DPS-InSAR phase unwrapping in GB-SAR, it is crucial to design an adaptive unwrapping strategy that integrates spatial constraints and local weight modeling to improve the robustness of unwrapping at low signal-to-noise ratio (DS) points.

[0045] To address the noise sensitivity issue in the phase unwrapping process of distributed scatterers, this application proposes a two-stage unwrapping framework that integrates the spatial constraints of permanent scatterers and the advantages of DS density. It achieves accurate reconstruction of the phase field by constructing a hierarchical spatial index and a dynamic weighting mechanism.

[0046] In some embodiments, the PS target point is used as a spatial reference field, and for each DS target point to be unwrapped, K nearest PS target points are searched within its effective neighborhood. Unwrapping is performed based on a weighted phase unwrapping model, including: Establish a KD-Tree spatial index structure between DS target points and PS target points; For each DS target point to be unwrapped, search for the K nearest unwrapped PS target points in its effective neighborhood; Based on the normalized distance weights between the DS target point and K PS target points, a weighted phase unwrapping model is constructed to calculate the unwrapping phase of the DS target point. The effective neighborhood is the region centered on the DS target point and within a distance of less than the preset spatial resolution.

[0047] In practical implementation, considering the short acquisition time interval of GB-SAR SLC images, most adjacent interferograms do not exhibit phase entanglement, leading to ineffective calculations by traditional global unwrapping strategies. Phase unwrapping is time-consuming; therefore, this application first determines whether interferograms are entangled before processing the interferograms that need unwrapping, thereby reducing unwrapping time and computational cost.

[0048] Specifically, the phase difference is calculated for each highly coherent target point and its neighboring pixels: , in, Represents pixels phase, Representing the neighborhood phase, This indicates the phase difference.

[0049] When the phase difference exceeds the threshold of 1.5π, it is considered a jump, and a jump counter is used. Record the number of jumps. When the number of jumps exceeds the preset threshold of 150, it is marked as needing phase unwrapping.

[0050] For DS point sets and PS point set Based on the spatial distribution characteristics, a KD-Tree multidimensional spatial index structure is established to realize DS points. Fast nearest neighbor search: , Calculate the normalized distance weight matrix: , This distance weight represents the spatial contribution of the PS point to the phase unwrapping of the DS point; the closer the PS point is, the higher its weight. Factors to prevent zero-distance anomalies.

[0051] The untangling is performed along the distance path, and its formula is expressed as: , In the formula, This represents the i-th row and j-th column. Represents the unwrapped phase, and represents the recovered true continuous phase value. This represents the entangled phase, which is the original phase obtained after interferometry. Its value range is typically (...). , This represents the integer number of wrapping operations. The recursion starting point is selected from the first valid PS point in each column to ensure the physical continuity of the unwrapping path.

[0052] During phase unwrapping, noise interference may cause local row unwrapping failures. To address this, this application proposes a dual quality control mechanism that achieves precise repair through the synergistic effect of statistical outlier detection and spatial autocorrelation constraints. The specific steps include: Systematic unwrapping errors, such as a column-wide phase shift of ±2π, can cause the column phase mean to deviate significantly from the global statistical characteristics. Statistical analysis can quickly locate anomalous columns with overall deviations. The column mean and global mean of the unwrapped phase matrix are calculated as follows: , in, This represents the unwrapped phase value of the pixel in the i-th row and j-th row of the unwrapped phase matrix. This represents the average phase value of the i-th column. This indicates the number of PS target points in the i-th column. This represents the global phase mean, which is calculated from the unwrapped phase mean of all PSs.

[0053] Define an outlier threshold, if If the i-th column is an outlier, then the i-th column is marked as an outlier column. This is an empirical coefficient.

[0054] Due to the geometric characteristics of SAR imaging, the phase fields between adjacent columns exhibit strong spatial correlation. Based on the spatial continuity assumption, reconstructing anomalous columns using the phase values ​​of normal columns can effectively preserve the spatial continuity characteristics of the deformation field. This application uses anomalous columns... Centered on the data, search for the 20 most recent non-abnormal columns in both left and right directions. To ensure spatial proximity, neighboring rows are given higher weights to reflect spatial decay effects, and each pixel in anomaly columns is assigned a different weight. Perform weighted interpolation.

[0055] Since distributed scatterers (DS) typically have a low signal-to-noise ratio, their phase information is easily affected by noise, leading to significant uncertainties and thus phase unwrapping errors. To improve the accuracy of unwrapping, this application proposes introducing permanent scatterer (PS) points as a spatial reference field for each DS point to be unwrapped. Search within its effective neighborhood Find the nearest neighbor PS points and construct the following weighted phase unwrapping model: , in, Represents the DS pixel to be untangled The untangling phase result, Represents the k-th PS point in the neighborhood. The untangled phase, Indicates the entanglement phase. Indicates DS pixels Effective neighborhood set, effective neighborhood set The constraint must be met, namely, the distance must be less than 130 times the spatial resolution. Indicates DS pixels PS points in the neighborhood The spatial weighting coefficients between them are used to suppress far-field noise through spatial weighting. The DS points processed in the above way are marked as unwrapped.

[0056] In some preferred embodiments, when the effective neighborhood set of a certain DS target point When the PS point is empty, meaning there are no available PS points to participate in the auxiliary untangling, a local phase field is constructed based on the already untangled DS point set, and inverse distance weighted interpolation is used: , in, for The M nearest neighbors of DS point have been untangled.

[0057] Atmospheric phase correction aims to eliminate phase delay errors in interferograms caused by spatiotemporal variations in atmospheric water vapor. Since spatially adjacent scatterers are affected by the same atmospheric disturbances, phase variations exhibit regional consistency. This application utilizes a spatial clustering-inverse distance weighted joint correction method. Specifically, it divides homogeneous regions through spatial clustering, uses the intra-cluster phase mean to characterize the regional atmospheric components, and combines this with inverse distance weighted interpolation to achieve spatially continuous correction.

[0058] To meet the timeliness and stability requirements of continuous surface deformation monitoring and to achieve closed-loop feedback and online adaptive updating of the deformation inversion system, this application also includes: During the real-time update phase, based on the confidence interval boundary parameters of the homogeneous pixel set determined in the previous round, pixels with unstable scattering characteristics caused by environmental changes are detected and located. The HTCI algorithm is used to update the local homogeneous pixel set in the region where the unstable pixel is located; Based on the preset expected number of PS target points, the amplitude deviation index threshold and coherence coefficient threshold are dynamically adjusted to maintain the stability of the number of PS target points. Based on the preset expected number of DS target points, the posterior coherence threshold is dynamically adjusted to maintain the stability of the number of DS target points.

[0059] Specifically, the improvements in the real-time update process are mainly reflected in the following three aspects: (1) The adaptive update strategy of homogeneous pixel set is adopted to replace the global re-identification process, which significantly reduces the computational complexity and improves the processing efficiency; (2) The dynamic optimization algorithm of PS / DS target points based on adaptive learning rate adjustment is introduced to replace the static scatterer set, so as to maintain the stability of the number and spatial distribution of monitoring points under rainfall or environmental disturbance conditions; (3) The low-quality PS and DS target points are integrated into a hybrid target set, and the high-quality PS phase is used as the spatial reference for joint unwrapping to enhance the spatial continuity of the phase field.

[0060] Among them, an adaptive update strategy using a homogeneous pixel set is adopted: To achieve rapid detection of rusty pixels, this application uses the confidence interval determined in the previous round of homogeneous pixel identification. For efficient positioning, it should be noted that rust pixels mainly originate from two types of land cover changes: the first type is where the reference pixel itself undergoes land cover changes, and its amplitude time-series average exceeds the existing confidence interval. The second scenario is that the reference pixel itself remains stable, while the homogeneous pixels in its neighborhood change, causing the reference pixel to lose its original statistical homogeneity.

[0061] After detection, for the newly acquired interferogram stack, this application utilizes the HTCI algorithm to perform localized homogeneous pixel set updates on the identified rust pixel regions, achieving dynamic adjustment and adaptive maintenance of the homogeneous pixel SHP. Simultaneously, the corresponding confidence intervals are also updated to ensure the robustness and recognition accuracy of the model under time-varying scattering conditions.

[0062] Dynamic update strategy for target points of permanent scatterers (PS) and distributed scatterers (DS): This dynamic update mechanism achieves adaptive adjustment of the threshold through an iterative optimization process, maintaining the statistical stability of the number of target points while ensuring recognition accuracy, thereby improving the robustness and adaptability of the system in long-term monitoring.

[0063] Among them, the dynamic selection of the target point of the permanent scatterer (PS): After each round of SLC stack loading, the initial magnitude deviation exponential threshold is first applied. and coherence threshold Extract the initial PS target point set, and denote the number of initial PS target points as . To keep the number of PS target points fluctuating stably around the target mean, the objective function is defined as follows: , in, This represents the number of PS target points after the k-th iteration. This represents the average number of PS target points in all processing results. The system has entered a dynamic adjustment phase.

[0064] Set the learning rate factor and Initialize to 0.1, step size is Then the amplitude difference threshold With coherence threshold The following adjustments will be made dynamically: , , In the formula, This represents the ADI threshold for the k-th iteration. This represents the coherence coefficient threshold for the k-th iteration. This represents the ADI threshold in the (k-1)th iteration. This represents the coherence coefficient threshold for the (k-1)th iteration. This represents the system adjustment factor for the (k-1)th iteration. The PS point selection process is then re-executed using the updated threshold. This iterative process continues until the iteration stopping condition is met. The iteration stopping condition is... Or the number of iterations has reached the maximum number of iterations. Finally, the updated PS point set and its quantity are obtained. The results are recorded in a statistical vector to participate in the next round of mean calculation and threshold update.

[0065] Among them, the dynamic selection of the target point of the distributed scatterer (DS) is as follows: A dynamic maintenance mechanism similar to that used for permanent scatterers is employed. Within identified homogeneous pixel regions, an initial coherence threshold is applied. Select target points of distributed scatterers that meet the conditions, and record the initial number as follows. The objective function is defined as follows: , in, The mean number of target points in the distributed scatterer over historical iterations, when At that time, perform iterative updates to the target point threshold of the distributed scatterer: , in, This is the learning rate factor, which dynamically increases with the number of iterations to accelerate convergence. After each iteration, the number of DS target points is recalculated and updated. The process continues until the convergence condition is met or the maximum number of iterations is reached. The final set of DS target points will then be used for subsequent deformation analysis.

[0066] In some embodiments, this application introduces a phase unwrapping method based on hierarchical quality control. Specifically, before performing dynamic threshold adjustment, a set of high-quality permanent scatterer (PS) points is extracted from the data based on initially set amplitude deviation index thresholds and coherence thresholds. This high-quality PS target point set is considered a phase-stable and reliable reference, used to guide the phase unwrapping process of low-quality PS target points and distributed scatterer (DS) target points selected after dynamic threshold adjustment. The reason for adopting this hierarchical strategy is that, in the subsequent dynamic PS target point selection stage, in order to improve the spatial density and coverage of target points, it is necessary to appropriately relax the amplitude deviation index and coherence thresholds. However, the reduction of the thresholds will inevitably lead to a decrease in the overall phase stability of the selected PS target points, thereby affecting their reliability as a direct unwrapping reference. By introducing a high-quality PS target point set as a hierarchical reference, the spatial integrity and phase continuity of the results can be maintained while ensuring unwrapping accuracy.

[0067] In some embodiments, considering that during deformation monitoring, as the sliding window advances and the PS and DS target points are dynamically updated, it is necessary to synchronously update the two-dimensional deformation field obtained in the previous round to accurately reflect the spatiotemporal evolution of surface deformation in the new time period. Therefore, this application proposes an atmospheric correction phase increment accumulation strategy based on a sliding window to achieve dynamic updating of the deformation matrix.

[0068] Specifically, assuming the phase matrix after atmospheric correction is If it is the initialization phase, then it is , This represents the total length of the time series, which is 30 in this application. For the number of monitoring points, As the sliding step size, this application sets it to 3, so any monitoring point The cumulative phase increment in the k-th round is expressed as: , This phase increment is used to update the two-dimensional deformation matrix from the previous round. Thus, the updated deformation field is obtained. .

[0069] For monitoring points that newly appear in the current round but were not included in the previous round If the corresponding position in the historical deformation matrix is ​​empty, then the deformation is estimated by interpolation using the deformation of existing monitoring points in its neighborhood: , in, Point The local neighborhood, The average deformation of the non-missing value points in the neighborhood; For monitoring points where historical values ​​already exist in the deformation matrix, the deformation update follows the direct accumulation principle: , Only the PS and DS target points selected in the current batch are retained as valid monitoring targets, while other non-target points are set to missing values. During each round of dynamic updates, only the permanent scatterer (PS) and distributed scatterer (DS) target points identified in the current batch are retained as valid monitoring targets, and all other non-target locations are set to null values. Through this update strategy, the deformation matrix achieves reasonable interpolation estimation for newly added points, continuous updating of existing points, and adaptive removal of non-target points in each round of the sliding window, thereby effectively ensuring the spatial smoothness and temporal consistency of the deformation field, providing a stable and reliable data foundation for subsequent deformation time series analysis and anomaly identification.

[0070] Another embodiment of this application provides a GB-SAR real-time deformation monitoring device based on DPS-InSAR for performing the above-described method. The device specifically includes: The data preprocessing module is used to acquire a preset number of SLC image sets, perform joint processing on the SLC image sets, and generate the cumulative deformation of the first time segment; it is also used to update the previous round of SLC image sets according to a sliding window with a fixed step size when M new images are acquired, forming a new SLC image set, returning to execute S1, and repeating steps S2 to S7 using the new SLC image set to realize the real-time update of deformation detection results. The interferogram generation module is used to select the master image from the SLC image set based on the time baseline minimization criterion, and generate an interferogram stack by performing complex conjugate multiplication with the other slave images based on the master image. The PS target point selection module is used to select a set of PS target points from the SLC image set based on preset amplitude deviation index threshold and time-series average coherence coefficient threshold. The DS target point selection module is used to calculate the average amplitude matrix of the SLC image set in the time dimension, and determine the effective scattering area based on the preset amplitude mask threshold. Within the effective scattering area, the confidence interval hypothesis testing algorithm is used to identify the homogeneous pixel set, and the maximum likelihood estimation eigenvalue decomposition algorithm is used to optimize the phase of the homogeneous pixel set to obtain the optimized phase result. Combined with the homogeneous pixel number threshold and the posterior coherence threshold, the DS target point set is determined. The interferogram filtering module is used to combine the PS target point set and the DS target point set into a high coherence target point set, and calculate the pixel difference between adjacent SLC images based on the interferogram stack. The pixel difference is compared with the preset judgment conditions to determine the interferograms that need to be unwrapped in phase. The unwrapping module is used to take the PS target point as the spatial reference field, search for K nearest PS target points in the effective neighborhood of each DS target point to be unwrapped, and perform unwrapping based on the weighted phase unwrapping model to obtain the unwrapped phase map. The phase correction module is used to perform atmospheric phase correction on the unwrapped phase map to obtain a phase map after removing the atmospheric phase.

[0071] This application uses an open-pit iron ore mining area as the experimental zone. This area exhibits a typical stepped open-pit mining landform, with the main slope consisting of hematite-quartzite layers formed by multi-level artificial mining. The surface is locally covered with sparse, low shrubs, representing a typical scenario with both complex terrain and weak coherence distribution characteristics. In this environment, vegetation interference makes it difficult for traditional PS-InSAR methods to select sufficient stable permanent scatterers, limiting their continuous monitoring capabilities for steep slope areas. To address these issues, this application focuses on the integrated application of ground-based SAR and DPS-InSAR technologies for real-time dynamic monitoring. By combining the high temporal resolution advantage of GB-SAR with the distributed scatterer phase optimization capability of DPS-InSAR, millimeter-level real-time displacement capture of potentially unstable areas on steep slopes in the mining area is achieved.

[0072] The dataset used in this application was acquired by a K-band Arc-orbit Ground-based Synthetic Aperture Radar (ArcSAR) system. The monitoring mission commenced at 18:09 on October 30, 2024, and continued until 22:08 on November 1, 2024, lasting a total of 52 hours, achieving continuous high-frequency observation of the target mine area. During this process, the radar system acquired a total of 912 complex scattering images, with an average time interval of approximately 3.5 minutes between adjacent images, demonstrating its ability to capture subtle and rapid surface deformations.

[0073] Throughout the monitoring period, the experimental area experienced complex meteorological and environmental changes, including significant diurnal temperature variations, multiple rainfalls, construction work interference, and high wind speeds. These factors significantly affected radar scattering characteristics, leading to a substantial reduction in the number of permanent scatterers (PS) during certain periods, resulting in large-area monitoring blind spots. (See [link to relevant documentation]). Figure 3 As shown in (a). Meanwhile, rapidly changing meteorological conditions can easily cause phase entanglement between adjacent images. (See [reference]). Figure 3 As shown in (b). In GB-SAR system applications, due to the prevalence of irregular occlusion, uneven spatial distribution of scatterers, and low coherence in vegetation-covered areas, traditional phase unwrapping methods are easily affected by noise and spatiotemporal decorrelation when processing such data, thus significantly increasing the complexity and uncertainty of the unwrapping process.

[0074] This application employed DPS-InSAR and traditional PS-InSAR technologies to conduct real-time surface deformation monitoring of the target mining area for 52 hours, and performed deformation field analysis every 10 hours. See [link to relevant documentation]. Figures 4 to 8 As shown, the deformation results of the two methods at different time periods are compared. Overall, the monitoring results indicate that no large-scale significant deformation occurred in the study area during continuous observation. The area marked with a red box exhibits localized anomalous deformation, mainly attributed to mining construction activities. To further verify this inference, the temporal amplitude variation curves of typical construction pixels within the red box area were extracted; see [link to relevant documentation]. Figure 9 As shown in the curve, the area exhibited significant amplitude fluctuations during the monitoring period, especially with a marked increase and fluctuation in amplitude within a specific time period, reflecting changes in the surface scattering characteristics due to construction disturbances.

[0075] Within the monitoring areas of [-1000, 1500m] in the azimuth direction and [0, 1400m] in the range direction, the deformation monitoring effects of DPS-InSAR and traditional PS-InSAR technologies exhibit significant spatial differentiation: PS-InSAR results show that surface deformation is distributed in discrete patches, especially in the sparse vegetation cover areas of [-500, 0m] and [200, 1000m] in the azimuth direction, where large monitoring blind spots exist, causing a break in the spatial continuity of the deformation field. This is mainly due to the fact that traditional methods rely on permanent scatterers (PS), which are easily affected by environmental interference and fail in areas with low scattering stability. In contrast, DPS-InSAR, by effectively mining distributed scatterer (DS) signals, successfully reconstructs a continuous strip-shaped deformation field within the same area, improving the completeness and spatial coverage of the monitoring. Experimental results show that combining distributed scatterer enhancement technology with high-frequency ground-based SAR observation data can effectively compensate for the monitoring blind spots of traditional PS-InSAR in mining environments with low vegetation cover, and improve the spatiotemporal resolution and reliability of surface deformation monitoring.

[0076] DPS-InSAR technology significantly improves the spatial coverage density of deformation monitoring by fusing the complementary characteristics of permanent scatterers (PS) and distributed scatterers (DS). After 52 hours of continuous monitoring, 346,080 PS target points and 102,098 DS target points were extracted, representing an overall increase of approximately 29.5% compared to the traditional PS-InSAR method. This advantage is mainly attributed to the effective supplementary role of DS target points in low-coherence regions. Figure 10 As shown, DS points (green) are concentrated in the shallow vegetation cover area in the azimuth direction (−500~0m) and the distance direction (200~1000m). Due to the dynamic changes in the vegetation canopy and the influence of time-varying humidity, PS points (blue) in this area are difficult to extract stably. By introducing homogeneous pixel selection and phase optimization strategies, DS points overcome the applicability limitations of traditional PS methods in natural surface monitoring and successfully capture deformation signals in vegetated areas.

[0077] See Figure 11 As shown, the temporal variation characteristics of the interferometric phase at five randomly selected target points within the aforementioned vegetation-covered area are illustrated. By comparing the trends before and after phase optimization, it can be observed that homogeneous pixel identification and phase optimization techniques significantly improve the signal quality in low-coherence regions. Before optimization ( Figure 11 a) The phase sequence exhibits strong non-stationarity, with its high-frequency random fluctuations mainly stemming from the instability of the vegetation canopy scattering mechanism and the combined influence of atmospheric disturbance noise; while after optimization ( Figure 11 (b) The phase sequence fluctuations were significantly reduced and tended to stabilize, indicating that spatiotemporal decorrelation noise was effectively suppressed. Statistical analysis further showed that the standard deviation of the phase sequence decreased from 0.787 rad to 0.653 rad after optimization, a reduction of approximately 17.0%; the proportion of fluctuations where the phase gradient instantaneously exceeded π / 2 decreased from 15.62% to 8.17%. These results validate the effectiveness of the optimization algorithm in reducing random noise and improving the stability of the phase temporal sequence, providing a more reliable foundation for subsequent deformation monitoring.

[0078] This application reveals the mechanism by which environmental disturbances affect ground-based SAR (GB-SAR) deformation monitoring results through joint analysis of temporal changes in meteorological parameters and the number of PS / DS target points. Figure 12 As shown, multiple rainfall events occurred during the monitoring period, some of which reached moderate intensity (marked by blue dots), accompanied by significant wind speed fluctuations. Combined with... Figure 13 The dynamic curves of the number of PS and DS target points reveal a significant spatiotemporal correlation between rainfall and strong wind events and the sharp decrease in the number of permanent scattering bodies (PS). Specifically, for example... Figure 13As shown in (a), the heavy rainfall that occurred between 03:00 and 04:00 on October 31st caused a sharp decrease in the number of target points, especially PS target points; subsequently, during two moderate rainfall events at 09:00 on October 31st and 06:00 on November 1st, the number of PS also showed a significant downward trend. However, after adopting a dynamic PS / DS target point adjustment strategy ( Figure 13 (b) The fluctuation of the PS count significantly decreased, and the sudden sharp decline that occurred during heavy rainfall events was no longer observed, indicating that the method has good robustness under rainfall interference conditions. Although higher wind speeds also have some impact on the number of target points, their interference effect is relatively weak compared to rainfall events.

[0079] Although rainfall significantly reduces the density of PS (Polarized scattering) points, the number of distributed scatterer (DS) target points remains relatively stable in most cases. This phenomenon is attributed to the pixel set identification and phase optimization mechanism based on statistical characteristics in DS-InSAR technology. Homogeneous pixel identification and joint phase optimization can reduce random noise interference introduced by rainfall to a certain extent. Although the scattering characteristics of individual pixels change with variations in soil moisture content, the overall scattering covariance matrix of the pixel cluster remains highly stable, thus maintaining the identification capability of DS points.

[0080] To systematically evaluate the real-time processing performance of the DPS-InSAR algorithm on a ground-based SAR platform, this application presents a quantitative analysis and optimization design of the algorithm's key processes and their running time.

[0081] The initialization phase is typically the most time-consuming part of the entire processing flow, mainly for two reasons: First, the system needs to accumulate 20–30 SLC images before deformation monitoring can begin. Since the SLC image acquisition interval in this experiment was 3.5 minutes, the data accumulation process itself introduced a time delay; Second, the selection of homogeneous pixels and phase optimization are the two most computationally intensive key steps.

[0082] To improve processing efficiency, this application introduces an adaptive masking technique for pure noise regions. This technique automatically identifies and removes pure noise pixels, reducing unnecessary computational overhead. Experimental results show that pure noise pixels account for approximately 46% of the total pixels. With an SLC image size of 1100×5000, without noise masking, homogeneous pixel identification takes approximately 130 seconds; after introducing the masking mechanism, this process is significantly reduced to approximately 70 seconds. In the phase optimization stage, despite employing a parallel computing strategy, the original processing still takes approximately 5 minutes. With the addition of noise masking, the processing time is reduced to approximately 3 minutes.

[0083] In the phase unwrapping and atmospheric correction section, this application employs an adjacent interferometric phase conversion and phase entanglement detection strategy to replace the traditional global unwrapping method. Because the time baseline of adjacent interferograms is short, most pixels do not become entangled, thus keeping the time for this step within one minute.

[0084] In summary, by introducing a noise mask and an optimized untangling mechanism, the overall processing time in the initialization phase is reduced to about 6 minutes, which significantly improves timeliness while ensuring accuracy and lays the foundation for subsequent real-time deformation monitoring.

[0085] During the real-time update phase, the homogeneous pixel set is adaptively updated based on the results of the previous round. The HTCI algorithm is re-executed only for detected rust pixel regions, significantly reducing the recognition time from approximately 70 seconds to about 20 seconds. Since the pixel scattering characteristics change relatively little within a short period, the number of rust regions is limited, further reducing the computational burden and improving overall efficiency. Furthermore, this phase only performs phase wrapping detection and atmospheric correction on the three newly added images, avoiding redundant calculations of the complete data stack. Overall, a complete update of a single sliding window takes approximately 4 minutes, improving the system's real-time monitoring capabilities and computational efficiency.

[0086] The specific limitations of the GB-SAR real-time deformation monitoring device based on DPS-InSAR provided in this embodiment can be found in the embodiment of the GB-SAR real-time deformation monitoring method based on DPS-InSAR described above, and will not be repeated here. Each module in the above-mentioned GB-SAR real-time deformation monitoring device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0087] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform the steps of the GB-SAR real-time deformation monitoring method based on DPS-InSAR as described in any of the above embodiments.

[0088] The working process, working details and technical effects of the computer equipment provided in this embodiment can be found in the embodiment of the GB-SAR real-time deformation monitoring method based on DPS-InSAR mentioned above, and will not be repeated here.

[0089] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system described in this application can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A real-time deformation monitoring method based on DPS-InSAR and GB-SAR, characterized in that, The method includes: S1: Obtain a preset number of SLC image sets, perform joint processing on the SLC image sets, and generate the cumulative deformation of the first time segment; S2: Based on the time baseline minimization criterion, a master image is selected from the SLC image set, and an interferogram stack is generated by performing complex conjugate multiplication with the other slave images using the master image as a reference. S3: Based on the preset amplitude deviation index threshold and time-series average coherence coefficient threshold, select a PS target point set from the SLC image set; S4: Calculate the average amplitude matrix of the SLC image set in the time dimension, and determine the effective scattering area based on the preset amplitude mask threshold. Within the effective scattering area, use the confidence interval hypothesis testing algorithm to identify the homogeneous pixel set, and use the maximum likelihood estimation eigenvalue decomposition algorithm to optimize the phase of the homogeneous pixel set to obtain the optimized phase result. Combine the homogeneous pixel number threshold and the posterior coherence threshold to determine the DS target point set. S5: Combine the PS target point set and the DS target point set into a high coherence target point set, calculate the pixel difference between adjacent SLC images based on the interferogram stack, compare the pixel difference with the preset judgment conditions, and determine the interferogram that needs to be unwrapped in phase. S6: Using the PS target point as the spatial reference field, search for K nearest PS target points in the effective neighborhood of each DS target point to be unwrapped, and unwrap it based on the weighted phase unwrapping model to obtain the unwrapped phase map; S7: Perform atmospheric phase correction on the unwrapped phase map to obtain a phase map after removing the atmospheric phase; S8: When M new images are acquired, the previous SLC image set is updated according to the sliding window with a fixed step size to form a new SLC image set. Then, return to execute S1 and repeat steps S2 to S7 using the new SLC image set to achieve real-time updating of deformation detection results.

2. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 1, characterized in that, The preset number of SLC image sets is 30 images. When M new images are acquired, the previous SLC image set is updated according to a sliding window with a fixed step size to form a new SLC image set, including: When three new images are acquired, the three earliest images in the current SLC image set are removed, and the three new images are added to form a new SLC image set.

3. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 1, characterized in that, Based on preset amplitude deviation index thresholds and time-series average coherence coefficient thresholds, a set of PS target points is selected from the SLC image set, including: Calculate the amplitude deviation index and temporal average coherence coefficient of each pixel in the SLC image; A pixel is selected as a PS target point when its amplitude deviation index is less than the preset ADI threshold and its temporal average coherence coefficient is greater than the preset coherence coefficient threshold.

4. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 1, characterized in that, The step involves calculating the average amplitude matrix of the SLC image set over time and determining the effective scattering region based on a preset amplitude mask threshold. Within the effective scattering region, the following is included: The average amplitude matrix of each SLC image is calculated in the time dimension. The average amplitude value is calculated based on the average amplitude matrix, and a binary mask matrix is ​​constructed. Pixel regions with amplitude greater than d / 3 are identified as effective scattering regions.

5. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 4, characterized in that, Step S4, which uses a confidence interval hypothesis testing algorithm to identify homogeneous pixel sets, also includes: The confidence interval boundary parameters of each homogeneous pixel set are cached, and the confidence interval boundary parameters are used to adaptively update the homogeneous pixel set during the real-time update phase.

6. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 1, characterized in that, The step of using the PS target point as a spatial reference field, searching for K nearest PS target points within the effective neighborhood of each DS target point to be unwrapped, and performing unwrapping based on a weighted phase unwrapping model includes: Establish a KD-Tree spatial index structure between DS target points and PS target points; For each DS target point to be unwrapped, search for the K nearest unwrapped PS target points in its effective neighborhood; Based on the normalized distance weights between the DS target point and the K PS target points, a weighted phase unwrapping model is constructed to calculate the unwrapping phase of the DS target point. The effective neighborhood is the region centered on the DS target point and at a distance less than the preset spatial resolution.

7. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 6, characterized in that, Also includes: When there are no unwrapped PS target points in the effective neighborhood of the DS target point, the unwrapping phase is calculated by inverse distance weighted interpolation based on the unwrapped DS target points around it.

8. The GB-SAR real-time deformation monitoring method based on DPS-InSAR according to claim 5, characterized in that, Also includes: During the real-time update phase, based on the confidence interval boundary parameters of the homogeneous pixel set determined in the previous round, pixels with unstable scattering characteristics caused by environmental changes are detected and located. The HTCI algorithm is used to update the local homogeneous pixel set in the region where the unstable pixel is located; Based on the preset expected number of PS target points, the amplitude deviation index threshold and coherence coefficient threshold are dynamically adjusted to maintain the stability of the number of PS target points. Based on the preset expected number of DS target points, the posterior coherence threshold is dynamically adjusted to maintain the stability of the number of DS target points.

9. A GB-SAR real-time deformation monitoring device based on DPS-InSAR, characterized in that, The apparatus for performing the method as described in any one of claims 1 to 8 comprises: The data preprocessing module is used to acquire a preset number of SLC image sets, perform joint processing on the SLC image sets, and generate the cumulative deformation of the first time segment; it is also used to update the previous round of SLC image sets according to a sliding window with a fixed step size when M new images are acquired, form a new SLC image set, return to execute S1, and repeat steps S2 to S7 using the new SLC image set to realize the real-time update of deformation detection results. The interferogram generation module is used to select a master image from the SLC image set based on the time baseline minimization criterion, and generate an interferogram stack by performing complex conjugate multiplication with the master image and other slave images based on the master image. The PS target point selection module is used to select a set of PS target points from the SLC image set based on a preset amplitude deviation index threshold and a time-series average coherence coefficient threshold. The DS target point selection module is used to calculate the average amplitude matrix of the SLC image set in the time dimension, and determine the effective scattering area based on a preset amplitude mask threshold. Within the effective scattering area, a confidence interval hypothesis testing algorithm is used to identify a set of homogeneous pixels, and a maximum likelihood estimation eigenvalue decomposition algorithm is used to optimize the phase of the set of homogeneous pixels to obtain the optimized phase result. The DS target point set is determined by combining the homogeneous pixel number threshold and the posterior coherence threshold. The interferogram filtering module is used to combine the PS target point set and the DS target point set into a high coherence target point set, calculate the pixel difference between adjacent SLC images based on the interferogram stack, compare the pixel difference with preset judgment conditions, and determine the interferogram that needs phase unwrapping. The unwrapping module is used to take the PS target point as the spatial reference field, search for K nearest PS target points in the effective neighborhood of each DS target point to be unwrapped, and perform unwrapping based on the weighted phase unwrapping model to obtain the unwrapped phase map. The phase correction module is used to perform atmospheric phase correction on the unwrapped phase map to obtain a phase map after removing the atmospheric phase.

10. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images