An InSAR automatic unwrapping method based on coherence hierarchical constraint

Abstract

The application relates to a radar remote sensing data processing technology, in particular to an InSAR automatic unwrapping method based on a coherence hierarchical constraint. The method comprises the following steps: acquiring an interference wrapped phase diagram and a coherence coefficient diagram, extracting a coherence feature to generate a coherence level diagram containing high, medium and low three-level regions; performing initial unwrapping based on the coherence level diagram and recording a propagation source, and constructing a coherence level propagation diagram endowed with a propagation risk weight; detecting a propagation anomaly by using a closed loop error, a boundary cycle slip mutation and high coherence area abnormal intrusion verification; when the anomaly is found, the error path is traced back to a high coherence reliable anchor point, a rollback processing is performed, and the integer cycle slips on the error path are cleared; under the constraint of the reliable anchor point, candidate integer cycle slip combinations are re-screened for local re-unwrapping, and a final unwrapped phase diagram and a reliability marking diagram are obtained. The application can effectively solve the problem of misjudgment diffusion from a low coherence area to a high coherence area, and improve the stability and reliability of unwrapping results.

Description

Technical Field

[0001] This application relates to radar remote sensing data processing technology, and in particular to an InSAR automatic unwrapping method based on coherence hierarchical constraints. Background Technology

[0002] Synthetic Aperture Radar Interferometry (InSAR) technology utilizes the phase difference information of two or more SAR images to accurately invert surface deformation, generate digital elevation models, and monitor geological hazards such as landslides and mining subsidence. It has been widely applied in geological exploration, urban subsidence monitoring, and disaster prevention and mitigation. In the InSAR data processing workflow, phase unwrapping is a crucial step in restoring the restricted wrapped phase to a continuous absolute phase, and its result directly determines the accuracy of subsequent inversion parameters. Existing phase unwrapping techniques mainly include branch-cutting methods based on path tracking, minimum cost flow methods based on global optimization, weighted least squares methods, and unwrapping methods that have emerged in recent years, combining deep learning to predict phase gradients. These existing techniques typically guide the priority of unwrapping paths by constructing quality maps (such as coherence maps, pseudo-correlation maps, etc.) or by using global optimization algorithms to minimize phase gradient errors, aiming to mathematically find the optimal combination of integer cycle slips to achieve continuous unwrapping of the wrapped phase.

[0003] However, in existing technologies, when large areas of low-coherence patches, abrupt coherence boundaries, or rapidly deformed local regions exist in the interferogram, the unwrapping process is prone to integer cycle slip misjudgments in low-reliability regions. These misjudgments can then propagate along the pre-defined unwrapping propagation path to adjacent high-coherence reliable regions, leading to systematic errors in the final unwrapped phase map, such as strip-like breaks, blocky shifts, or localized overall lifting and sinking. Existing methods primarily focus on optimizing the initial path or providing a one-time global solution, lacking automatic error propagation path localization mechanisms, dynamic backoff strategies for reliable boundaries, and the ability to redistribute integer cycle slips for local error propagation chains. This makes it difficult to guarantee the automated stability and local consistency of the unwrapping results in complex scenarios. Summary of the Invention

[0004] This application provides an automatic unwrapping method for InSAR based on coherence hierarchical constraints to solve the above problems. The method includes:

[0005] S1. Obtain the InSAR interferometric winding phase map and coherence coefficient map of the area to be processed, and determine the coherence statistical characteristics of each pixel or region unit based on the coherence coefficient map.

[0006] S2. Based on the coherence statistical characteristics, coherence spatial variation characteristics, residual point distribution characteristics, and entanglement phase gradient consistency characteristics, a coherence level map is generated. The coherence level map includes at least a high coherence confidence region, a medium coherence verification region, and a low coherence risk region.

[0007] S3. Perform initial unwrapping processing on the interference entanglement phase map based on the coherence level map to obtain an initial unwrapped phase map, and record the unwrapping propagation source relationship between each pixel or region unit during the initial unwrapping processing;

[0008] S4. Using each pixel or region unit as a node and the unwrapping propagation source relationship as a directed edge, construct a coherence level propagation graph and assign a propagation risk weight to the directed edge. The propagation risk weight is determined at least based on the coherence level difference between the nodes at both ends of the directed edge, the boundary coherence change, the boundary phase gradient change, and the residual point crossing situation.

[0009] S5. Based on the coherence level propagation map, perform propagation anomaly verification on the initial unwrapped phase map. The propagation anomaly verification includes at least two of the following: closed loop error verification, coherence level boundary cycle slip mutation verification, and high coherence confidence region abnormal intrusion verification.

[0010] S6. When it is determined that there is an abnormal propagation node or abnormal propagation edge according to the propagation anomaly verification, the region corresponding to the abnormal propagation node or abnormal propagation edge is used as the backtracking starting point, and the error propagation path is traced back along the directed edge of the coherence level propagation graph until the high coherence trust anchor point that meets the trust condition is reached.

[0011] S7. Using the highly coherent trusted anchor point as the backoff reference, perform backoff processing on the region corresponding to the error propagation path. The backoff processing includes retaining the unwrapped phase of the highly coherent trusted anchor point and its upstream trusted region, clearing the integer cycle jump allocation result of the region corresponding to the error propagation path, and increasing the propagation cost of abnormal propagation edges in the error propagation path or setting them as temporary blocking edges.

[0012] S8. Under the boundary constraints of the highly coherent and reliable anchor point, the candidate integer cycle jump combination is re-determined for the region after the back-off process, and the candidate integer cycle jump combination is screened according to the coherence level cost, phase gradient consistency cost, closed loop error cost and boundary continuity cost to obtain the local re-unwrapping result.

[0013] S9. Update the local re-unwrapping result to the initial unwrapping phase map to obtain the final unwrapping phase map, and output the unwrapping confidence mark map corresponding to the final unwrapping phase map.

[0014] The above technical solution enables explicit modeling and monitoring of the unwrapping path by constructing a coherence level propagation map, making error propagation localizable, traceable, and repairable. By using high-coherence regions as reliable anchor points and setting back endpoints, the spread of errors from low-coherence regions to high-coherence regions is effectively curbed. Furthermore, by performing local re-unwrapping only on areas affected by the error propagation chain instead of recalculating the entire map, computational overhead is significantly reduced. Based on this, the ability to automatically identify errors is enhanced through joint verification of three indicators: closed-loop error, boundary cycle jump abrupt changes, and high-coherence anomaly intrusion, reducing the need for manual intervention. More importantly, by outputting result layers with reliability markers, phase jumps caused by real geological deformation and pseudo-abrupt changes caused by low-coherence noise are distinguished, thus avoiding misjudging real landslide or subsidence signals as unwrapping errors and enhancing the reliability of the results.

[0015] Optionally, in step S1, the pixel or region unit is obtained by regular window partitioning, connected component partitioning, or superpixel segmentation.

[0016] For each of the aforementioned regional units, the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and winding phase gradient direction consistency are calculated, and the above calculation results are used as the coherence statistical characteristics, coherence spatial variation characteristics, residual point distribution characteristics, and winding phase gradient consistency characteristics.

[0017] Through the above technical solutions, region units are obtained by regular windows, connected components, or superpixel segmentation. A multi-dimensional coherence feature system is constructed by combining five complementary indicators: average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and consistency of the winding phase gradient direction. The average coherence coefficient and coherence coefficient variance are used together to evaluate the regional signal quality from both overall level and internal fluctuation dimensions, preventing local high values ​​from masking overall low signal quality or overall mean from masking internal abrupt changes. The coherence gradient magnitude and residual point density work synergistically; the former identifies boundary regions with abrupt changes in coherence, while the latter locates noise centers with phase discontinuities. Their combination can accurately pinpoint high-incidence areas of unwrapping errors. The consistency of the winding phase gradient direction further verifies the rationality of phase changes from a physical mechanism perspective, forming a triple verification of statistics, structure, and physics with the aforementioned statistical features.

[0018] Optionally, in step S2, a comprehensive confidence score is calculated for each of the aforementioned regional units. The comprehensive confidence score is obtained by weighting the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and the consistency of the winding phase gradient direction.

[0019] Based on the comprehensive confidence score, each regional unit is divided into a high coherence confidence zone, a medium coherence verification zone, a low coherence risk zone, and an untangleable shielding zone. The untangleable shielding zone is not used as an untangle propagation channel.

[0020] Through the aforementioned technical solution, and by leveraging the synergistic effect of comprehensive confidence score calculation and a four-level region division mechanism, a shift from passively tolerating noise to actively isolating risks is achieved. The weighted combination of five features—average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and consistency of the entanglement phase gradient direction—not only provides a refined basis for quality assessment but also adapts to varying imaging environments through dynamic weight adjustments. Based on this, the assessment results are mapped to a high-coherence confidence region, a medium-coherence verification region, a low-coherence risk region, and a non-unentangled shielding region, constructing a hierarchical unentanglement propagation topology. In particular, the introduction of the non-unentangled shielding region complements the high-coherence confidence region; the former acts as a physical barrier to block error sources, while the latter acts as a logical anchor point to ensure the correctness of the baseline. Together, they effectively reduce the probability of low-coherence patches contaminating the overall unentanglement result. With this hierarchical constraint strategy, the subsequent backtracking verification mechanism can accurately locate the error propagation path, performing local re-unentanglement only on affected low-risk regions, thereby significantly improving the algorithm's robustness and automated processing capabilities while ensuring unentanglement accuracy.

[0021] Optionally, in step S3, when performing the initial unwrapping process, the high coherence confidence region is taken as the priority unwrapping region, the medium coherence verification region is taken as the propagable region, and the low coherence risk region is taken as the restricted propagation region.

[0022] When the unwrapping phase of a certain region cell is obtained by propagation from an adjacent region cell, the adjacent region cell that provides the unwrapping reference is recorded as the parent node for the unwrapping propagation of that region cell, and the corresponding propagation direction, boundary phase difference, and integer cycle jump assignment value are recorded.

[0023] Through the aforementioned technical solutions and the synergistic effect of these technical features, the controllability and traceability of the initial unwrapping process are achieved. By setting the high-coherence reliability region as the priority unwrapping area, the starting reference for phase calculation is ensured to have the highest reliability, fundamentally suppressing the generation of errors. Simultaneously, different propagation constraints are applied to medium-coherence and low-coherence regions, optimizing the global topology of the unwrapping path and avoiding interference from noisy regions. Based on this, the unwrapping propagation parent node, propagation direction, boundary phase difference, and integer cycle slip assignment values ​​are recorded in real time. This not only fully preserves the logical chain of phase calculation but also transforms the implicit calculation process into explicit graph structure data. This combination of hierarchical constraint propagation and full-link information recording enables the system to accurately trace back along the recorded parent node relationships when subsequent closed-loop errors or boundary abrupt changes are detected, quickly locating the specific propagation edge or low-coherence source that caused the error. This supports local backtracking and re-unwrapping for erroneous paths, significantly improving the automation and robustness of InSAR phase unwrapping in complex scenarios.

[0024] Optionally, in step S4, the propagation risk weight of the directed edge is determined as follows:

[0025] The propagation risk weight is obtained by weighting and summing the coherence level crossings of the nodes at both ends of the directed edge, the difference in boundary coherence coefficients, the difference in boundary phase gradients, the number of residual points crossed, and whether or not a low-coherence-risk zone is crossed.

[0026] Specifically, when the directed edge points from a low-coherence-risk region to a medium-coherence-verification region or a high-coherence-trust region, the level crossing penalty value of the directed edge is increased.

[0027] Through the above technical solutions, a propagation risk assessment system with direction sensitivity and multi-dimensional perception capabilities was constructed. The comprehensive weighting of coherence level crossings, boundary coherence differences, phase gradient inconsistencies, residual point distribution, and low-coherence crossings ensures that the risk weights can fully reflect the complexity of the local environment; while the specific penalty for propagation from low to high coherence further enhances the model's ability to predict error propagation trends. The combination of these two aspects means that the coherence level propagation map not only records the connectivity between pixels but also embeds the probability distribution information of error propagation. When there are erroneous cycle slips caused by noise in the low-coherence region in the initial unwrapping result, these errors often propagate along paths with underestimated risks. This solution, by increasing the weight of the reverse propagation edges, makes such paths appear as high-cost channels in the map, thus enabling faster and more accurate location of the error propagation starting point during subsequent closed-loop error verification and reverse tracing. This mechanism effectively solves the problem of low-coherence errors spreading to the high-coherence core region caused by the uniform weighting standard in traditional methods, achieving the technical effect of accurately blocking the propagation of false errors while preserving true deformation mutations.

[0028] Optionally, in step S5, the closed-loop error verification includes:

[0029] In the coherence level propagation diagram, a closed propagation loop consisting of adjacent nodes is selected, and the integer cycle jump difference accumulated along the closed propagation loop is calculated;

[0030] When the cumulative integer cycle jump value of the closed propagation loop does not meet the closure consistency condition, the closed propagation loop is marked as an abnormal closed loop, and the directed edge that participates in the abnormal closed loop and has the highest propagation risk weight is determined as a candidate abnormal propagation edge.

[0031] By utilizing the topological characteristics of the coherence level propagation graph, the above technical solution enables automated detection of global unwrapping consistency. By selecting adjacent nodes to form a closed propagation loop and calculating the cumulative integer cycle jump difference, local errors violating the phase continuity principle can be detected from the perspective of geometric closed loop. When the cumulative value does not meet the closure consistency condition, candidate abnormal propagation edges are further screened using propagation risk weights, achieving a logical progression from anomaly detection to suspicious point location. This synergistic effect allows the system not only to identify erroneous regions but also to quickly pinpoint the most likely source of error based on prior risk assessment information. This provides accurate initial input for subsequent reverse tracing of error propagation paths along the coherence level propagation graph, effectively preventing the spread of integer cycle jump misjudgments from low-coherence regions to high-coherence confidence regions, significantly improving the reliability of the final unwrapped phase graph.

[0032] Optionally, in step S5, the coherence level boundary cycle slip abrupt change verification includes:

[0033] For directed edges that cross the boundaries between the high coherence confidence region, the medium coherence verification region, and the low coherence risk region, calculate the consistency between the unwrapping phase difference, the wrapping phase difference, and the integer cycle jump difference on both sides of the boundary.

[0034] When an integer cycle jump difference occurs abruptly in the spatial neighborhood on both sides of the boundary, and the abrupt change is not supported by a coherent gradient change or a tangled phase gradient change, the corresponding directed edge is marked as an anomalous propagation edge.

[0035] By employing the aforementioned technical solution, a rigorous logical judgment system is constructed by combining the consistency calculation of unwrapped phase difference, wrapped phase difference, and integer cycle jump difference with the coherence gradient and phase gradient characteristics within the spatial neighborhood. On one hand, the consistency calculation mathematically identifies the points of contradiction in the phase relationship; on the other hand, the gradient support judgment physically verifies the rationality of the causes of these contradictions. This collaborative mechanism enables the system to accurately locate erroneous propagation edges caused by low coherence noise but disguised as real deformations in complex InSAR interferograms. Furthermore, accurate marking of these anomalous propagation edges provides reliable triggering conditions for subsequent erroneous path backtracking, ensuring that the backtracking operation is only performed on genuine erroneous paths. This guarantees the global continuity of the unwrapping results while preserving the detailed information of real surface deformation to the greatest extent possible, significantly improving the robustness and reliability of the automatic unwrapping method in complex scenarios.

[0036] Optionally, in step S5, the anomaly intrusion verification of the high coherence trust region includes:

[0037] Determine whether the unentanglement propagation source of the target's high coherence confidence region passes through a low coherence risk region;

[0038] When the unwrapping propagation source of the target high coherence confidence region passes through a low coherence risk region, and the target high coherence confidence region has an integer multiple phase shift, a decrease in local phase continuity, or an increase in closed loop error relative to the adjacent high coherence confidence region, the target high coherence confidence region is identified as an abnormal intrusion region.

[0039] By employing the aforementioned technical solution, potential contamination paths can be identified in advance by determining whether the unwrapping propagation source of the target high-coherence confidence region passes through a low-coherence risk region. Furthermore, comprehensive verification is conducted by combining multi-dimensional characteristics such as integer multiple phase shifts, decreased local phase continuity, and increased closure loop errors, ensuring that anomalies are only identified when high-coherence regions are indeed affected by low-coherence error propagation. This synergistic effect not only overcomes the missed detection problem caused by traditional methods relying solely on local coherence coefficients while ignoring propagation history, but also effectively avoids misjudging real rapid deformations or geological boundaries as unwrapping errors, significantly improving the reliability and robustness of InSAR phase unwrapping results in complex surface environments.

[0040] Optionally, in step S6, the reverse tracing error propagation path includes:

[0041] Using the region corresponding to the abnormal propagation node or abnormal propagation edge as the backtracking starting point, backtrack level by level along the direction of the untangling propagation parent node;

[0042] In each backtracking process, the error contribution value is determined based on the propagation risk weight of the directed edge, the closed loop error contribution value of the closed propagation loop in which it is located, and the coherence level crossing situation.

[0043] The backtracking continues along the propagation direction with the largest error contribution value until a highly coherent trusted anchor point that meets the trusted conditions is reached. The trusted conditions include: the highly coherent trusted anchor point does not belong to an abnormal closed loop, was not obtained by direct propagation from a low-coherence risk region, and the integer cycle jump relationship between it and the adjacent highly coherent trusted region meets the consistency condition.

[0044] Using the aforementioned technical solution, a physical path model for error propagation is established by starting with an abnormal node and tracing back along its parent node. Based on this, an error contribution value evaluation system is constructed by introducing propagation risk weights, closed-loop error contribution values, and coherence level crossing scenarios. This transforms the backtracking process from a simple geometric deduction into an intelligent optimization based on statistical characteristics and topological constraints, significantly improving the accuracy of locating the true source of error in complex low-coherence environments. Furthermore, by setting multiple credibility conditions, including non-abnormal loops, non-low-coherence direct connections, and neighborhood consistency, high-coherence credibility anchor points are defined, ensuring the absolute reliability of the backtracking benchmark and preventing the use of the erroneous benchmark as the starting point for repair. Through this mechanism, the system can accurately remove and isolate error propagation chains caused by low-coherence regions while preserving a large area of ​​correct unwrapping results. This lays a solid foundation for subsequent local re-unwrapping targeting only the erroneous path region, thereby significantly improving the overall accuracy and credibility of the final unwrapped phase map while reducing computational load.

[0045] Optionally, in steps S8 and S9, when redetermining candidate integer cycle jump combinations for the region after backtracking, the unwrapped phase of the highly coherent and reliable anchor point is used as a boundary constraint, and the comprehensive cost of each candidate integer cycle jump combination is calculated respectively.

[0046] The overall cost is obtained by weighting the coherence level cost, phase gradient consistency cost, closed loop error cost, boundary continuity cost, and shielding zone crossing penalty cost.

[0047] Candidate integer cycle jump combinations whose comprehensive cost value meets the preset screening conditions are selected to generate local re-unwrapping results. Based on whether the local re-unwrapping results have undergone backtracking verification, whether they have passed the closed loop verification, and whether they originate from the low coherence risk area, a corresponding unwrapping confidence marker map is generated.

[0048] By employing the aforementioned technical solution, and using high-coherence reliability anchors as rigid boundary constraints, a comprehensive cost function is constructed by combining coherence level cost, phase gradient consistency cost, closed-loop error cost, boundary continuity cost, and shielding zone crossing penalty cost. This enables refined screening of candidate schemes for local re-unwrapping. Furthermore, an unwrapping reliability marker map is generated based on the historical trajectory of the solution process (whether it has backtracked), topology verification results (closed-loop consistency), and data source characteristics (whether it originates from a low-risk area). This synergistic mechanism not only ensures the generation of high-quality repair results after the discovery of error propagation paths, solving the technical challenge of potential new inconsistencies arising from local re-unwrapping, but also allows downstream users to intuitively identify the reliability distribution of data by outputting a visualized reliability layer. This enables flexible adjustments to data usage strategies in deformation inversion or disaster early warning, significantly improving the automation level and engineering practical value of the InSAR data processing workflow. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of an InSAR automatic unwrapping method based on coherence hierarchical constraints, provided as an embodiment of this application. Detailed Implementation

[0051] 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, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0052] Furthermore, the term "and / or" in this article 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 article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0053] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0054] Example 1

[0055] See Figure 1 The following is a flowchart of an InSAR automatic unwrapping method based on coherence hierarchical constraints, provided in an embodiment of this application. The method includes the following steps:

[0056] Step 1: Obtain the InSAR interferometric winding phase map and coherence coefficient map of the area to be processed, and determine the coherence statistical characteristics of each pixel or region unit based on the coherence coefficient map;

[0057] The area to be processed can refer to a specific geographical area requiring surface deformation inversion or DEM generation. Its input data includes complex interferograms generated from primary and secondary SAR images after registration, de-flattening, and filtering. Pixels or regional units are the basic processing units for coherence analysis. They can be obtained through regular window partitioning, connected component partitioning, or superpixel segmentation. Regional unit partitioning is preferred to avoid frequent triggering of subsequent backoff processing by individual noisy pixels. Coherence statistical features refer to a set of multidimensional indicators used to quantify the phase reliability of the regional unit, specifically including the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and consistency of the winding phase gradient direction. These features are obtained through local statistical calculations of the interferometric winding phase map and coherence coefficient map of the area to be processed. For example, for a regular grid area divided into 5×5 pixels, the arithmetic mean of the coherence coefficients of all pixels within the window is calculated as the average coherence coefficient, its standard deviation is calculated as the coherence coefficient variance, and the spatial rate of change of the coherence coefficient is extracted using the Sobel operator as the coherence gradient magnitude. The average coherence coefficient is used to characterize the overall signal strength of the region, the coherence coefficient variance reflects the signal stability, and the residual point density directly indicates the density of phase noise. By integrating the statistical characteristics of the above multiple dimensions, the edges of low coherence patches and regions of abrupt coherence changes can be identified more accurately than the single threshold method, providing a more refined basis for subsequent classification processing.

[0058] Step 2: Based on the coherence statistical characteristics, coherence spatial variation characteristics, residual point distribution characteristics, and entanglement phase gradient consistency characteristics, generate a coherence level map. The coherence level map includes at least a high coherence confidence region, a medium coherence verification region, and a low coherence risk region.

[0059] The coherence level map is a spatial distribution map that divides the area to be processed into different reliability levels. Its generation process is based on the weighted fusion and threshold determination of the calculated features. Specifically, firstly, a comprehensive reliability score is calculated for each region unit. This score is obtained by weighting the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and the consistency of the winding phase gradient direction. The weighting coefficients can be adaptively determined according to the sensor type or surface type. Then, based on the comprehensive reliability score, each region unit is divided into different levels: the high coherence reliability zone corresponds to a region with a high score, serving as the anchor point and backoff endpoint for unwrapping; the medium coherence verification zone corresponds to a region with a medium score, allowing unwrapping propagation but requiring boundary verification; the low coherence risk zone corresponds to a region with a low score, allowing limited propagation but marked as a risk source; additionally, it may include non-unwrappable shielding areas, such as water bodies or severely decoherent areas, which are not used as propagation channels. For example, three adaptive thresholds T1, T2, and T3 are set. When the overall confidence score of a certain region is greater than T1, it is classified as a high-coherence confidence region; when it is between T2 and T1, it is classified as a medium-coherence verification region; and when it is between T3 and T2, it is classified as a low-coherence risk region. This classification strategy not only utilizes coherence amplitude information but also introduces residual point density and phase gradient consistency, enabling the edges of low-coherence patches and local pseudo-continuum regions to be accurately identified as sources of risk propagation, thereby constructing a constraint system for untangling propagation.

[0060] Step 3: Perform initial unwrapping processing on the interferometric entanglement phase map based on the coherence level map to obtain the initial unwrapped phase map, and record the unwrapping propagation source relationship between each pixel or region unit during the initial unwrapping processing;

[0061] The initial untangling process refers to the process of recovering a continuous phase from an entangled phase using algorithms such as minimum cost flow, weighted least squares, branching, or quality graph-guided region growth. During this process, high-coherence reliability regions are designated as priority untangling regions, medium-coherence verification regions as propagable regions, and low-coherence risk regions as restricted propagation regions, ensuring that the untangling path prioritizes passing through high-reliability regions. The untangling propagation source relationship refers to the logical connection from which neighboring region units propagate or reference the untangled phase of the current region unit during the untangling process. Specifically, when the untangled phase of a region unit is obtained from propagation by a neighboring region unit, the neighboring region unit providing the untangling reference is recorded as the untangling propagation parent node of that region unit, and the corresponding propagation direction, boundary phase difference, and integer cycle slip assignment value are recorded simultaneously. For example, if the phase of region unit B is derived from its left-side neighboring high-coherence region unit A, then A is recorded as the parent node of B, and the propagation vector from A to B is stored. This recording process makes the implicit untangling path explicit, forming the initial propagation chain from parent node to child node, laying the data foundation for subsequent construction of the propagation graph and tracking of error paths.

[0062] Step 4: Using each pixel or region unit as a node and the unwrapping propagation source relationship as a directed edge, construct a coherence level propagation graph and assign a propagation risk weight to the directed edge. The propagation risk weight is determined at least based on the coherence level difference between the nodes at both ends of the directed edge, the boundary coherence change, the boundary phase gradient change, and the residual point crossing situation.

[0063] The coherence level propagation graph is a directed graph structure. Its node set corresponds to the pixels or regions defined above, and its edge set corresponds to the untangling propagation source relationships recorded above. Each directed edge represents the direction of untangling phase propagation from one node to another. The propagation risk weight is a numerical indicator that quantifies the probability of error occurring along the propagation path. It is determined by weighted summation of the coherence level crossing status of the nodes at both ends of the directed edge, the difference in boundary coherence coefficients, the difference in boundary phase gradients, the number of residual points crossed, and whether it crosses a low-coherence-risk region. Specifically, when a directed edge points from a low-coherence-risk region to a medium-coherence verification region or a high-coherence-confidence region, the level crossing penalty value of that directed edge is significantly increased because such reverse propagation easily introduces errors. For example, if an edge connects a low-coherence-risk region node and a high-coherence-confidence region node and crosses multiple residual points, its propagation risk weight will be assigned a maximum value, indicating that the path is extremely unreliable. By assigning risk weights, the graph not only records how the virus spreads but also quantifies the risks of its spread, enabling the precise identification of potential abnormal propagation edges based on their weights.

[0064] Step 5: Based on the coherence level propagation map, perform propagation anomaly verification on the initial unwrapped phase map. Propagation anomaly verification includes at least two of the following: closed loop error verification, coherence level boundary cycle slip mutation verification, and high coherence confidence region anomaly intrusion verification.

[0065] Among them, the propagation anomaly verification is a mechanism that uses physical consistency and logical constraints to detect whether there are errors in the initial unwrapping results. Closed-loop error verification involves selecting a closed propagation loop composed of adjacent nodes in the coherence level propagation graph, calculating the accumulated integer cycle jump difference along the loop, and marking the loop as an anomalous loop when the accumulated value does not meet the closure consistency condition (i.e., theoretically it should be zero but actually is not). The edge participating in the loop with the highest risk weight is identified as a candidate anomalous edge. Coherence level boundary cycle jump mutation verification involves calculating the consistency of the unwrapping phase difference, entanglement phase difference, and integer cycle jump difference on both sides of a directed edge crossing different coherence level boundaries. When the integer cycle jump difference on both sides of the boundary undergoes a sudden change in the spatial neighborhood and is not supported by coherence or gradient changes, it is marked as an anomaly. High coherence confidence region anomalous intrusion verification determines whether the unwrapping propagation source of the target high coherence confidence region passes through a low coherence risk region. If it does, and this region has an integer multiple phase shift or a decrease in continuity relative to the adjacent high coherence region, it is identified as an anomalous intrusion region. For example, in a city monitoring scenario, if the unwrapping path tracing of a highly coherent building area reveals its origin in a low-coherence construction water area, and an unnatural step in the phase within the building appears, then an abnormal intrusion is determined to have occurred. These three verification methods complement each other, enabling comprehensive capture of unwrapping errors from three dimensions: loop closure, boundary continuity, and source reliability.

[0066] Step Six: When an abnormal propagation node or edge is identified based on the propagation anomaly check, the region corresponding to the abnormal propagation node or edge is used as the backtracking starting point. The error propagation path is traced backward along the directed edges in the coherence level propagation graph until a high coherence trust anchor point that meets the trust conditions is reached. Tracing the error propagation path backward can refer to the process of starting from the detected anomaly location and searching for the error source in the reverse direction of untangling propagation. Specifically, the region corresponding to the abnormal propagation node or edge is used as the backtracking starting point, and backtracking is performed level by level along the direction of the untangling propagation parent node. During each level of backtracking, the error contribution value is calculated based on the propagation risk weight of the directed edge, the closure loop error contribution value of the closed propagation loop it belongs to, and the coherence level crossing situation. The backtracking continues along the propagation direction with the largest error contribution value. The high coherence trust anchor point is the termination point of the backtracking and must meet specific trust conditions, including: the anchor point belongs to a high coherence trust region, does not belong to an abnormal closed loop, was not directly propagated from a low coherence risk region, and the integer cycle jump relationship with adjacent high coherence regions meets the consistency condition. For example, if a closed-loop error is detected in a certain region, the system will start from that region and trace back to its parent node. If an edge with an extremely high risk weight is found and originates from a low-coherence region, the system will continue tracing along that edge until a stable region surrounded by high coherence and without any anomaly markers is found. This stable region is the high-coherence trusted anchor point. This step ensures the accuracy of error localization and avoids blindly recalculating the entire graph.

[0067] Step 7: Using the high coherence trusted anchor point as the backoff benchmark, perform backoff processing on the region corresponding to the error propagation path. The backoff processing includes retaining the unwrapped phase of the high coherence trusted anchor point and its upstream trusted region, clearing the integer cycle jump allocation results of the region corresponding to the error propagation path, and increasing the propagation cost of abnormal propagation edges in the error propagation path or setting them as temporary blocking edges.

[0068] The rollback process is a local error correction mechanism designed to break the error propagation chain and reset the state of the affected area. Specifically, it first locks and preserves the unwrapped phase of a highly coherent, reliable anchor point and all upstream feasible regions not marked as anomalous, treating them as the correct baseline. Next, it clears the integer cycle slip assignments for all regions along the error propagation path from this anchor point to the anomalous point, restoring these regions to an unconfirmed state. Simultaneously, to improve the robustness of subsequent re-unwrapping, the propagation cost of anomalous propagation edges in the error propagation path is increased, or they are directly set as temporary blocking edges to prevent the unwrapping algorithm from selecting the same erroneous path again. For example, if a path through water is determined to have caused a phase error in the preceding building area, the rollback process not only clears the cycle slip values ​​on the building area and water path but also sets the weight of the water path to infinity or marks it as impassable. This approach effectively isolates the error source, creating conditions for finding the correct propagation path again.

[0069] Step 8: Under the boundary constraints of the highly coherent and reliable anchor points, the candidate integer cycle jump combinations are re-determined for the region after the backtracking process. The candidate integer cycle jump combinations are then screened based on the coherence level cost, phase gradient consistency cost, closed loop error cost, and boundary continuity cost to obtain the local re-unwrapping result.

[0070] Local re-unwrapping involves resolving the optimal integer cycle slip within the backtracking region. This process uses the unwrapped phase of a high-coherence, reliable anchor point as a hard boundary constraint to ensure a smooth transition between the newly generated phase and the surrounding reliable region. Specifically, a candidate integer cycle slip set is first established within the backtracking region, and then the comprehensive cost is calculated for each candidate combination. This comprehensive cost is weighted by coherence level cost (preferring high-coherence paths), phase gradient consistency cost (preferring smooth gradients), loop closure error cost (preferring closure consistency), and boundary continuity cost (preferring continuity with anchor points), and may also include a penalty cost for crossing the shielded region. By minimizing this comprehensive cost, the optimal candidate integer cycle slip combination is selected to generate the local re-unwrapping result. For example, during re-unwrapping, the algorithm may tend to bypass previously marked low-coherence regions that were high-risk, instead choosing a path that passes through the medium-coherence verification region, even if the path is slightly longer, because its comprehensive cost is lower and its reliability is higher. This step achieves intelligent path planning and phase repair within a local scope.

[0071] Step 9: Update the local re-unwrapping results to the initial unwrapping phase map to obtain the final unwrapping phase map, and output the unwrapping confidence marker map corresponding to the final unwrapping phase map.

[0072] The update operation seamlessly integrates the obtained local correction results into the global phase map, replacing the original erroneous data to obtain a more accurate final unwrapped phase map. The unwrapping confidence marker map is a quality assessment layer output with the final phase map. Its generation is based on the experience of each pixel or region unit during the unwrapping process: whether it underwent backtracking verification, whether it passed loop closure verification, and whether it originated from a low-coherence-risk region. Confidence markers can be divided into multiple levels, such as Level 1 Confidence (originating from a high-coherence anchor point and not triggering backtracking), Level 2 Confidence (propagated through a medium-coherence verification region and passed verification), Level 3 Confidence (passed through a low-coherence-risk region but completed backtracking and re-unwrapping), and Level 4 Risk (failed verification, only the initial result is retained). For example, a region that has undergone backtracking and re-unwrapping and successfully passed loop closure verification is marked as Confidence after verification; a region that is still in low coherence and cannot find a reliable path is marked as Risk Retention. This output method not only provides corrected phase data, but also intuitively displays the reliability distribution of the data, providing important decision-making basis for subsequent applications such as surface deformation inversion and disaster early warning. Users can flexibly filter out data from risk areas based on the reliability label.

[0073] Example 2

[0074] In another optional embodiment, the method of dividing the processing unit in step S1 and the quantitative calculation process of coherence characteristics are also included.

[0075] Step 1: Pixels or region units are obtained through regular window partitioning, connected component partitioning, or superpixel segmentation.

[0076] Here, a pixel or region unit can refer to the basic computational unit for coherence statistics and unwrapping propagation in the InSAR interferogram to be processed. The source of this basic computational unit can be the original single pixel, but in order to suppress local noise interference and improve feature stability, it is preferable to use a region unit approach, that is, to obtain it by aggregating multiple adjacent pixels into a processing block. There are three main methods for obtaining regional units: First, regular window partitioning, which cuts the interferogram into non-overlapping rectangular grids according to fixed row and column sizes (such as 5×5, 7×7, or 9×9 pixels). This method is suitable for scenarios with relatively uniform land cover, and is computationally efficient and simple to implement. Second, connected component partitioning, which clusters spatially adjacent and similarly attributed pixels into irregularly shaped connected regions based on coherence coefficient thresholds or phase continuity. This method can adaptively preserve the complete structure of high-coherence patches and avoid forcibly cutting off the same land feature. Third, superpixel segmentation, which uses image segmentation algorithms (such as the SLIC algorithm) to generate superpixel blocks with boundaries that fit the edges of land features based on phase gradient and coherence similarity. This method has significant advantages when dealing with urban building edges or complex terrain, and can effectively reduce the mixing effect across land feature boundaries. For example, in urban land subsidence monitoring scenarios, densely built-up areas typically exhibit high coherence and contiguous distribution. Connected-domain partitioning can treat an entire building as a single regional unit, while the boundary between roads and green spaces can be precisely defined through superpixel segmentation, thus avoiding the boundary ambiguity caused by a single rule window. This diverse partitioning strategy allows for flexible selection of the optimal processing granularity based on different land surface types, providing a structured data foundation for subsequent feature extraction.

[0077] Step 2: Calculate the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and winding phase gradient direction consistency for each region unit. Use the above calculation results as coherence statistical characteristics, coherence spatial variation characteristics, residual point distribution characteristics, and winding phase gradient consistency characteristics.

[0078] The average coherence coefficient refers to the arithmetic mean of the coherence coefficients of all pixels within a region, used to characterize the overall signal quality and unwrapping reliability of the region. A higher value indicates more stable phase observations in the region. The coherence coefficient variance refers to the dispersion of the coherence coefficient distribution within a region, reflecting the uniformity of coherence within the region. A large variance indicates abrupt changes in coherence or the mixing of different types of ground features within the region. The coherence gradient magnitude is obtained by calculating the gradient magnitude of the coherence coefficient map in the horizontal and vertical directions, used to describe the drastic changes in coherence in space, and often appears in the transition zone between low-coherence patches and high-coherence areas. The residual point density refers to the number of residual points (i.e., phase closure loops and non-compliance points) per unit area. The number of points (in integer multiples) is used to quantify the level of phase noise and the difficulty of unwrapping. Areas with dense residual points usually indicate severe decoherence or rapid deformation. The consistency of the wrapped phase gradient direction refers to the degree of aggregation of the phase gradient vector directions of each pixel within a region, used to judge the physical continuity of phase changes. Good consistency indicates that the phase changes in the region conform to the actual terrain or deformation trend; conversely, poor consistency may be dominated by noise. These five indicators together constitute a multi-dimensional feature vector: the average coherence coefficient and coherence coefficient variance mainly serve as statistical features of coherence, reflecting the baseline quality of the region; the coherence gradient magnitude serves as a spatial variation feature of coherence, capturing edge information; the residual point density directly serves as a residual point distribution feature, identifying risk sources; and the consistency of the wrapped phase gradient direction serves as a consistency feature of the wrapped phase gradient, verifying the physical rationality of the phase field. For example, a vegetated area might have an average coherence coefficient of only 0.4 (low), but if its coherence coefficient variance is small and the phase gradient direction consistency is high, it indicates that although the signal is weak, the changes are stable, and it can still be considered a region that can be untangled. Conversely, if an area has an average coherence coefficient as high as 0.8, but the residual point density is extremely high and the coherence coefficient variance fluctuates drastically, it is very likely due to false high coherence caused by multipath effects or shading, and should be identified as a high-risk area. Through this multi-feature joint extraction mechanism, the untangling potential of regional units can be comprehensively characterized, avoiding misjudgments caused by relying solely on a single coherence threshold for classification, and providing detailed data support for the subsequent construction of a refined coherence level map.

[0079] Example 3

[0080] One embodiment also includes specific steps for calculating a comprehensive credibility score for each regional unit and dividing the region into four levels based on the score.

[0081] Step 1: Calculate the comprehensive confidence score for each region unit. The comprehensive confidence score is obtained by weighting the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and the consistency of the winding phase gradient direction.

[0082] The comprehensive reliability score is a scalar indicator used to quantitatively evaluate the unwrapping reliability of each regional unit. It is generated by linearly weighting and fusing multiple statistical features. Specifically, the average coherence coefficient reflects the stability of the scatterer within the region; a higher value indicates lower phase noise. The coherence coefficient variance characterizes the uniformity of coherence distribution in space; excessive variance usually indicates the presence of mixed pixels or edge effects. The coherence gradient magnitude is used to identify boundaries where coherence changes drastically; such boundaries are often potential sources of phase discontinuities. The residual point density directly measures the density of noise or real abrupt changes in the phase field; high-density regions significantly increase the difficulty of unwrapping. The consistency of the wrapped phase gradient direction describes the degree of agreement on phase change trends within the neighborhood; poor consistency suggests unwrapping ambiguity. The comprehensive reliability score is obtained by multiplying each of the above five features by a preset weighting coefficient and then summing them. For example, weighting coefficients can be set based on prior knowledge of sensor type (such as Sentinel-1 or TerraSAR-X) and land cover type (such as urban areas, vegetation, and water bodies). For urban built-up areas, the weight of the average coherence coefficient can be appropriately increased, while for deformation monitoring in mountainous areas, the weight of the consistency of the winding phase gradient direction can be increased. Through this weighted fusion of multi-dimensional features, the limitations of single indicators in complex scenarios can be overcome, enabling a comprehensive assessment of the quality of regional units.

[0083] Step 2: Based on the comprehensive confidence score, each region unit is divided into a high coherence confidence region, a medium coherence verification region, a low coherence risk region, and an untangleable shielded region. The untangleable shielded region is not used as an untangle propagation channel.

[0084] The region division is a classification operation based on the numerical range of the comprehensive confidence score, aiming to construct a hierarchical untangling constraint system. The high coherence confidence zone corresponds to the highest score range; this region has excellent phase quality and is defined as the starting anchor point and final verification benchmark of the untangling process. The medium coherence verification zone corresponds to the medium score range, allowing untangled phase propagation, but requiring strict boundary consistency checks. The low coherence risk zone corresponds to the lower score range; although limited propagation is allowed, it is considered a high-risk area for error propagation, and detailed propagation paths need to be recorded for subsequent backtracking. The non-untangling shielding zone corresponds to the lowest score range, specifically referring to water bodies, severely shadowed areas, overlapping areas, or completely incoherent areas; this region is explicitly marked as an untangling forbidden zone. Specifically, the division thresholds can adopt an adaptive strategy, for example, taking the 75th, 45th, and 20th percentiles of the comprehensive confidence score histogram of the entire image as the division points for each level to adapt to the overall quality fluctuations of different images. By establishing an unwrapable shielding region and prohibiting it from serving as a propagation channel, the spread of erroneous phase to high-quality regions can be blocked at the source, avoiding the overall phase field distortion caused by the forced unwrapping of low-quality regions in traditional methods.

[0085] Example 4

[0086] In one optional implementation, the method further includes implementing differentiated propagation strategies for different types of region units based on the coherence level map during the initial unwrapping process, and simultaneously constructing an unwrapping propagation source record.

[0087] Step 1: Select the high coherence confidence region as the priority untangling region, the medium coherence verification region as the propagation region, and the low coherence risk region as the restricted propagation region.

[0088] Among them, the high coherence reliability region refers to the set of regional units in the coherence level map that are judged to have the highest phase reliability, the lowest residual point density, and the best phase gradient consistency. Its role is to serve as the seed or starting anchor point for the initial unwrapping, ensuring that the unwrapping process starts from the region with the lowest error probability. The medium coherence verification region refers to the region with medium coherence, which may have local noise but has acceptable overall phase continuity. Its role is to serve as a transition channel connecting the high coherence region and the low coherence region, allowing the unwrapped phase to propagate normally in this region. The low coherence risk region refers to the region with low coherence coefficient and high phase noise due to the influence of vegetation, water bodies, or severe deformation. Setting it as a restricted propagation region means that in the unwrapping path search algorithm, the edges that cross this region are given a high cost weight, or it is restricted to only be the propagation endpoint and not a new propagation starting point, unless there is no better path to choose from. For example, in unwrapping algorithms based on minimum cost flow or region growing, all nodes within the high-coherence confidence region can be initialized and marked as unwrapped, and nodes in the medium-coherence check region adjacent to their boundaries can be added to a priority queue. The propagation path cost involving low-coherence risk regions can be set to 3 to 5 times that of medium-coherence paths, thus forcing the unwrapped wavefront to preferentially bypass low-coherence patches. This hierarchical constraint strategy effectively prevents unwrapped paths from prematurely entering low-quality regions, reducing the risk of integer cycle jump misjudgments spreading to high-quality regions from the source.

[0089] Step 2: When the unwrapping phase of a certain region cell is obtained by propagation from the adjacent region cell, record the adjacent region cell that provides the unwrapping reference as the parent node for the unwrapping propagation of that region cell, and record the corresponding propagation direction, boundary phase difference and integer cycle jump assignment value.

[0090] In this context, the untangling propagation parent node can refer to the upstream adjacent region unit that directly provides a phase reference for the current region unit during the untangling propagation process. Its source is dynamically determined by tracing the execution path of the untangling algorithm. The propagation direction can refer to the vector direction from the parent node to the current child node, representing the flow path of phase information. The boundary phase difference can refer to the entanglement phase difference between the parent node and the current child node at the common boundary, serving as the fundamental data for calculating integer cycle slips. The integer cycle slip allocation value can refer to the value used to eliminate... Ambiguity is accumulated based on the unwrapped phase of the parent node. value( (where is an integer) such that the untangled phase of the current child node is physically continuous with that of the parent node. Specifically, assume the region unit... untangling phase It consists of adjacent regional units What is obtained through transmission is recorded. Simultaneously store tuples ,in Indicates from regional unit Pointing to region unit The direction of dissemination The winding phase difference between the two unit boundaries. This is the calculated integer number of cycle jumps. For example, when the unwrapped wavefront propagates from a highly coherent building area (parent node) to an adjacent road edge area (child node), if the boundary entanglement phase difference between the two is... If the algorithm determines that no cycle jump needs to be crossed (i.e., k=0), then the integer cycle jump assignment value for that edge is recorded as 0; if subsequent verification finds that the edge actually exists... If a transition occurs, the correct k value can be re-derived using the recorded original boundary phase difference. By fully recording this propagation source information, an explicit untangled propagation tree or graph structure is constructed, providing an indispensable data foundation for subsequent steps such as constructing a coherent level propagation graph, locating erroneous propagation paths, and performing reverse backtracking, enabling any local phase anomaly to be traced back to a specific propagation source.

[0091] Example 5

[0092] In one embodiment, a process for determining the risk weights of directed edge propagation is provided, including:

[0093] Step 1: The propagation risk weight is obtained by weighting and summing the coherence level crossing status of the nodes at both ends of the directed edge, the difference in boundary coherence coefficient, the difference in boundary phase gradient, the number of residual points crossed, and whether the edge crosses a low coherence risk zone.

[0094] Among them, the propagation risk weight is the core indicator for quantifying the possibility of untangling errors spreading along a specific path. The coherence level crossing situation can refer to the hierarchical difference between the starting and ending nodes of a directed edge in the coherence level diagram (high coherence confidence region, medium coherence verification region, low coherence risk region), used to characterize the ease or difficulty of untangling information flowing from a high-quality region to a low-quality region or vice versa; the boundary coherence coefficient difference can refer to the absolute difference of the average coherence coefficients of the units on both sides of the directed edge, reflecting the magnitude of the sudden change in signal quality at the boundary; the boundary phase gradient difference can refer to the vector difference of the entangled phase gradients on both sides of the directed edge, used to measure whether the continuity of phase changes is disrupted; the number of residual points crossed can refer to the residual points directly crossed by the directed edge (i.e., phase closure loops and points that do not meet the requirements). The number of residual points (in integer multiples) indicates the probability of phase singularities in the path; more residual points mean a higher probability of phase singularities. Whether the path traverses a low-coherence-risk region is a binary or hierarchical marker used to identify whether the propagation path passes through areas of low reliability. Specifically, propagation risk weights... It can be done through formula The calculation yielded, where As a factor for level transition, , Representing regional units respectively and regional units The average coherence coefficient, This represents the difference in boundary coherence coefficients. , Representing regional units respectively and regional units phase gradient, The boundary phase gradient difference, For the number of residual points crossed, This is an indicator function for traversing the low-coherence-risk region. to These are the corresponding weighting coefficients. For example, when a directed edge connects two highly coherent regions with coherence coefficients of 0.85 and 0.82 respectively, and no residual points are crossed, the differences in its various parameters are small, resulting in a low calculated propagation risk weight, indicating high reliability of the path. Conversely, if a directed edge crosses the boundary where the coherence coefficient drops sharply from 0.9 to 0.3 and crosses three residual points, its calculated propagation risk weight will significantly increase. This multi-factor weighting mechanism transforms complex physical environment characteristics into a unified numerical metric, enabling subsequent error path identification algorithms to prioritize high-risk paths based on quantified risk values, thus avoiding the bias of a single indicator.

[0095] Step 2: When a directed edge points from a low-coherence-risk region to a medium-coherence-verification region or a high-coherence-trust region, the level crossing penalty value of that directed edge is increased.

[0096] The level-crossing penalty value is a nonlinear correction term embedded in the weighted summation formula, specifically designed to handle inverse level propagation scenarios. During InSAR unwrapping, the unwrapped phase should typically propagate from the high coherence confidence region (as the anchor point) to the low coherence risk region to minimize errors. If the propagation direction is reversed, i.e., from the low coherence risk region to the medium coherence verification region or the high coherence confidence region, it means that low-reliability phase information is intruding into the high-reliability region, easily leading to widespread integer jump errors. Therefore, this step artificially and significantly increases the propagation risk weight of such inverse edges by introducing an additional penalty coefficient or increasing the basic weight value. Specifically, when calculating the coherence level-crossing component, a directional function f(Dir) can be set. When the propagation direction is L→M (low to medium) or L→H (low to high), f(Dir) takes a magnification factor greater than 1 (e.g., 2.0 or 3.0), or a fixed high penalty constant can be directly superimposed. For example, suppose a directed edge points from a low-coherence-risk region with a coherence coefficient of 0.25 to a high-coherence-confidence region with a coherence coefficient of 0.88. Even if its boundary phase gradient changes smoothly and there are no residual points crossing it, its final propagation risk weight will be forcibly increased to a high level due to the triggering of the backpropagation penalty mechanism, making it appear as a high-impedance path in the coherence level propagation graph. This step complements the multi-factor weighting mentioned above. The above provides an objective risk assessment based on local features, while this step introduces prior constraints based on a global unwrapping strategy. The combination of the two can effectively suppress the contamination of high-quality unwrapping results by low-quality data. Therefore, when constructing the coherence level propagation graph and performing anomaly checks, the algorithm will automatically tend to avoid these high-weighted back edges, or prioritize locking such edges as suspicious sources when tracing back erroneous paths, significantly improving the system's sensitivity and interception capability for potential error propagation chains.

[0097] Example 6

[0098] In another alternative embodiment, specific steps are also included to perform closed-loop error verification in the coherence level propagation graph.

[0099] Step 1: Select a closed propagation loop consisting of adjacent nodes in the coherence level propagation diagram, and calculate the integer cycle jump difference accumulated along the closed propagation loop;

[0100] Here, a closed propagation loop can refer to a closed path in the coherence level propagation graph G=(V,E) consisting of a series of directed edges connected end-to-end. Specifically, the selection process involves starting from any unverified region cell node and traversing along the directed edges of the unwrapping propagation relationship. When the traversal path returns to the starting node, a closed propagation loop is formed. Integer cycle jump can refer to the integer number of cycles represented in the unwrapping phase ( The integer cycle jump difference accumulated along the closed propagation loop is the algebraic sum of the integer cycle jump changes corresponding to all directed edges on that loop. Physically, for an error-free continuous phase field, after one revolution along any closed loop, the total phase change should be zero; that is, the accumulated integer cycle jump difference should satisfy the closure consistency condition (usually zero or a very small noise margin). During the calculation, the system sequentially reads the boundary integer cycle jump difference recorded for each directed edge on the closed loop. And sum them up to get the total. For example, in a rectangular closed loop consisting of four regional units A, B, C, and D, if the cycle slip from A to B is 0, the cycle slip from B to C is +1, the cycle slip from C to D is 0, and the cycle slip from D to A is -1, then the cumulative value is 0, satisfying the closure condition. If the cycle slip from D to A is mistakenly judged as 0, then the cumulative value is 1, indicating the existence of an error. Through this traversal calculation method, the inconsistency of the unwrapping results in local loops can be quantitatively detected.

[0101] Step 2: When the cumulative value of integer cycle jump difference of the closed propagation loop does not meet the closure consistency condition, the closed propagation loop is marked as an abnormal closed loop, and the directed edge that participates in the abnormal closed loop and has the highest propagation risk weight is determined as a candidate abnormal propagation edge;

[0102] The closure consistency condition is a pre-set threshold used to distinguish normal phase noise from genuine unwrapping errors. When the absolute value of the calculated cumulative integer cycle jump difference exceeds this threshold, the closed propagation loop is determined to be an anomalous closed loop, meaning that there is at least one propagation edge within the loop that has misjudged an integer cycle jump. Candidate anomalous propagation edges are high-risk edges selected from all directed edges constituting the anomalous closed loop. The selection criterion is the propagation risk weight of each directed edge, which is determined during the construction of the coherence level propagation graph based on the coherence level difference between the two endpoints, boundary coherence changes, boundary phase gradient changes, and residual point crossings. Specifically, the system sorts the propagation risk weights of all directed edges within the anomalous closed loop or sets a high-weight threshold, marking the edge(s) with the highest weight as candidate anomalous propagation edges. This is because high propagation risk weights typically correspond to crossings from low to high coherence regions, large abrupt changes in phase gradients, or dense areas of residual points, where the probability of unwrapping errors is significantly higher than at other locations. For example, if an abnormal closed loop contains four edges, three of which lie within the high coherence confidence region (with relatively low weights), and the remaining edge crosses the boundary between the low coherence risk region and the high coherence confidence region (with the highest weight), then this high-weight edge is preferentially identified as a candidate anomaly propagation edge, serving as the starting point for subsequent error path backtracking. By combining the geometric constraints of the closed loop with the statistical characteristics of propagation risk weights, the key links most likely to cause error propagation can be accurately located, avoiding blindly checking all edges in the loop and improving the efficiency of error localization.

[0103] Example 7

[0104] In one optional implementation, the method further includes a specific process for performing consistency checks and anomaly marking on directed edges that cross the boundaries of regions with different coherence levels.

[0105] Step 1: For directed edges that cross the boundaries between the high coherence confidence region, the medium coherence verification region, and the low coherence risk region, calculate the consistency between the unwrapping phase difference, the wrapping phase difference, and the integer cycle jump difference on both sides of the boundary.

[0106] The high coherence confidence region, medium coherence verification region, and low coherence risk region are sets of regional units divided based on the coherence level map generated in the preceding steps. A directed edge can be a propagation path connecting two adjacent regional units belonging to different coherence levels, such as an edge connecting a node in the high coherence confidence region and a node in the medium coherence verification region, or an edge connecting a node in the medium coherence verification region and a node in the low coherence risk region. Computational consistency can refer to verifying whether the change in the unwrapping phase is strictly equal to the change in the wrapping phase plus an integer multiple thereof. Specifically, for any directed edge that crosses the coherence level boundary... Its starting point is the regional unit. The endpoint is the regional unit. First obtain and Untangling phase value in the initial untangling phase diagram and Calculate the unwrapping phase difference Simultaneously, the entanglement phase values ​​of the two regions in the original interferometric entanglement phase diagram are obtained. and Calculate the entanglement phase difference (Phase unwrapping process is required to eliminate) (The original difference before blurring); and then derive the theoretical integer cycle jump difference. This theoretical value is then compared with the integer cycle jumps allocated during the actual unwinding process. A comparison is performed. If the two match, it indicates that the phase jump at the boundary conforms to physical laws; if they do not match, it indicates a potential unwrapping error. For example, when a directed edge points from a low-coherence-risk region to a high-coherence-confidence region, if the calculated unwrapping phase difference is... The corresponding entanglement phase difference is only At this point, the theoretical integer cycle jump should be 1, but the actual recorded value may be incorrectly assigned as 2 or 0 due to noise interference. This inconsistency is identified as a failure of the consistency check. Through this multi-dimensional difference comparison, the rationality of the phase jump at the boundary can be quantified, providing a data basis for subsequent judgment of the authenticity of the abrupt change.

[0107] Step 2: When the integer cycle jump difference on both sides of the boundary undergoes an abrupt change within the spatial neighborhood, and this abrupt change is not supported by changes in coherent gradient or entangled phase gradient, the corresponding directed edge is marked as an anomalous propagation edge. The spatial neighborhood can refer to the set of adjacent region units centered on the current directed edge, encompassing a predetermined number (e.g., 8 neighborhoods or a 5×5 window) of neighboring regions. An abrupt change in the integer cycle jump difference within the spatial neighborhood can mean that, locally, the integer cycle jump value of the current edge exhibits a significant step change compared to the integer cycle jump values ​​of other edges in the neighborhood, exceeding the gradient range that could be caused by normal terrain deformation or atmospheric delay. The lack of support from changes in coherent gradient or entangled phase gradient can mean the absence of physical precursor features that would lead to drastic phase changes. In real-world rapid surface deformation (such as landslides and mining subsidence) or steep terrain, a sharp decrease in coherence coefficient (large coherence gradient) or violent oscillations in the entanglement phase (large phase gradient) is usually observed. Conversely, if an integer cycle jump occurs abruptly, but the coherence coefficient distribution near that location is gradual (small coherence gradient) and the entanglement phase changes smoothly (small phase gradient), it indicates that the abrupt change lacks physical basis and is highly likely due to mispropagation by the unentanglement algorithm in low-coherence regions. Specifically, the system calculates the magnitude of the coherence gradient at the current location of the directed edge. and the magnitude of the entangled phase gradient and respectively compared with the preset support threshold. and Compare them. If the rate of change of the integer cycle jump exceeds the mutation threshold, and at the same time satisfies... and If the mutation is deemed unsupported, the directed edge is marked as an anomalous propagation edge. For example, in a mining area monitoring scenario, if an integer cycle jump suddenly increases by two cycles at a directed edge crossing a boundary, but the area has stable vegetation cover, a coherence coefficient as high as 0.8 with no drastic changes, and the entanglement phase shows a linear and slow trend, then the cycle jump mutation can be determined to be a false signal, and the edge is marked as an anomaly. Thus, this step effectively distinguishes between phase discontinuities caused by real deformation and phase breaks caused by algorithm errors, avoiding the accidental deletion of genuine geological disaster signals and preventing erroneous cycle jumps from further propagating into high-coherence confidence regions.

[0108] Example 8

[0109] In another embodiment, an abnormal intrusion verification process for the high coherence trusted region is provided, including the following steps:

[0110] Step 1: Determine whether the unwrapping propagation source of the target's high coherence confidence region passes through a low coherence risk region;

[0111] The target high-coherence confidence region refers to a region cell in the coherence level diagram that is classified as a high-coherence confidence region (H region). It has a high average coherence coefficient and good phase gradient consistency, and is usually used as a reliable benchmark for the unwrapping result. The unwrapping propagation source can refer to the adjacent upstream region cells that provide integer cycle jump reference values ​​during the initial unwrapping process of this region cell, and the propagation path chain they form. The judgment process specifically includes: traversing the directed edges in the coherence level propagation diagram pointing to the node of the target high-coherence confidence region, backtracking step by step along the direction of the unwrapping propagation parent node, and checking whether all intermediate nodes on the backtracking path contain nodes marked as low-coherence risk regions (L regions). If there is at least one low-coherence risk region node in the backtracking path, it is determined that the unwrapping propagation source of the target high-coherence confidence region has passed through a low-coherence risk region. For example, when the unwrapping phase of a densely built area (high coherence) propagates from a construction bare soil area (low coherence), and there is no buffer between the two high-coherence regions, the condition of passing through a low-coherence risk region is met. This tracing mechanism can identify potential risk areas that, while having good local quality indicators, rely on unreliable paths for untangling results.

[0112] Step 2: When the unwrapping propagation source of the target high coherence confidence region passes through the low coherence risk region, and the target high coherence confidence region has an integer multiple phase shift, a decrease in local phase continuity, or an increase in the closure loop error relative to the adjacent high coherence confidence region, the target high coherence confidence region is identified as an abnormal intrusion region.

[0113] Here, an integer multiple of phase offset can refer to the phase difference between the unwrapped phase value within the target's high coherence confidence region and other high coherence confidence regions in the spatial neighborhood that are not affected by low coherence paths. An integer multiple of this value indicates a possible misjudgment of integer cycle jumps; a decrease in local phase continuity can mean that the smoothness of the phase gradient within the target region is significantly lower than the statistical mean of similar high-coherence regions, indicating a non-geologically caused phase break or abrupt change; an increase in closed-loop error can mean that the absolute value of the accumulated integer cycle jump difference in the local closed propagation loop containing the target region exceeds a preset threshold. Specifically, the system first calculates the average phase difference between the target high-coherence confidence region and multiple adjacent high-coherence confidence regions. If this difference falls within a certain range... Within the interval ( It is a non-zero integer. If the tolerance is set at a certain value, then an integer multiple of phase offset is determined to exist. Simultaneously, the variance of the phase gradient within this region is calculated; if the variance abnormally increases, then local continuity is determined to be decreased. Furthermore, a closed loop spanning the boundary of this region is selected; if the accumulated error... Greater than the threshold If the error increases, the closed loop error is determined to be larger. Through the combined constraints of the above multiple conditions, it is possible to effectively distinguish between real deformation abrupt changes and erroneous propagation caused by low coherence noise. Only when the propagation source is unreliable and accompanied by obvious phase inconsistency characteristics is it marked as an anomalous intrusion area, thereby avoiding the accidental deletion of real geological phenomena.

[0114] Example 9

[0115] In another optional embodiment, the process of tracing the error propagation path in reverse includes the following steps: Step 1: Using the region corresponding to the abnormal propagation node or abnormal propagation edge as the backtracking starting point, backtrack step by step along the direction of the untangling propagation parent node;

[0116] In this context, anomaly propagation nodes or edges refer to the regional units and their connecting edges that were marked as having closed-loop errors, boundary cycle slips, or anomalous intrusions into high-coherence regions during the aforementioned propagation anomaly verification. Unwrapping propagation parent nodes refer to the adjacent regional units that provide unwrapping phase reference values ​​to the current regional unit during the initial unwrapping process. Specifically, starting from the identified anomalous region, the system traces back along the recorded directed edges to find its previous level source, i.e., to identify who transmitted the unwrapping result to the current anomalous node. For example, if regional unit B5 is determined to be an anomalous intrusion region, and its unwrapping phase originates from the adjacent low-coherence-risk region B4, then B5 is the starting point for backtracking, and B4 is its first-level unwrapping propagation parent node. Through this step-by-step backtracking mechanism, the system can trace back along the error propagation trajectory until the source path causing the error is located, avoiding computational redundancy caused by blindly searching the entire graph.

[0117] Step 2: In each level of backtracking, determine the error contribution value based on the propagation risk weight of the directed edge, the closed loop error contribution value of the closed propagation loop it belongs to, and the coherence level crossing situation.

[0118] The error contribution value is a comprehensive indicator used to quantify the probability that a directed edge on the current backtracking path will cause an untangling error. The propagation risk weight reflects the inherent risks brought about by the difference in coherence levels between the two ends of the edge, changes in boundary coherence, and the crossing of residual points. The closed-loop error contribution value can refer to the proportion or influence of the cumulative error of integer cycle jumps on the total error of the closed propagation loop formed by the edge. The coherence level crossing situation describes whether the edge involves a reverse or abnormal crossing from a low-coherence risk region to a high-coherence confidence region. Specifically, the error contribution value can be calculated by weighted summation. Edges with high propagation risk weights, large closed-loop errors, and crossings from low-coherence regions to high-coherence regions are assigned higher error contribution values. For example, if an edge on a backtracking path... It connects the low-coherence-risk region and the medium-coherence-check region, and the cumulative integer cycle jump difference of the closed loop containing this edge is... If the phase gradient difference between the two ends of an edge is extremely large, then the error contribution of that edge is significantly higher than that of other edges with smooth transitions. By comprehensively considering these three dimensions, we can accurately identify the critical links most likely to cause error propagation, thereby guiding the backtracking process to prioritize the direction with the highest suspicion.

[0119] Step 3: Continue backtracking along the propagation direction with the largest error contribution value until a highly coherent trusted anchor point that meets the trusted conditions is reached. The trusted conditions include: the highly coherent trusted anchor point does not belong to an abnormal closed loop, was not obtained by direct propagation from a low-coherence risk region, and the integer cycle jump relationship between it and the adjacent highly coherent trusted region satisfies the consistency condition.

[0120] The highly coherent trusted anchor point serves as the termination node in the reverse tracing process, acting as a reliable benchmark for subsequent backtracking. The trusted condition involves multiple verification constraints on the anchor point's reliability: First, the anchor point cannot be located within any closed propagation loop marked as anomalous, ensuring local phase consistency; second, the unwrapped phase of the anchor point cannot directly originate from a low-coherence-risk region, cutting off the direct input source of error propagation; finally, the integer cycle jump relationship between the anchor point and other surrounding highly coherent trusted regions must satisfy spatial continuity or consistency conditions to prevent isolated highly coherent points from exhibiting overall offset. Specifically, during the backtracking process, the algorithm compares the error contribution values ​​of the corresponding edges of all upstream parent nodes of the current node, selects the direction with the largest value as the next hop path, and repeats the above calculation and judgment steps. Once a region cell is encountered, and it is determined that it does not belong to an anomalous closed loop, its parent node is not a low-coherence-risk region, and its phase is continuous with the neighboring highly coherent region, then the cell is determined to be a highly coherent trusted anchor point, and the backtracking stops. For example, when tracing an erroneous path extending from an urban construction zone (low coherence) to a stable building zone (high coherence), the backtracking will traverse multiple intermediate coherence transition zones until it reaches a building block completely surrounded by other stable building zones and without internal closed-loop errors. This block is then designated as the anchor point. Through this intelligent backtracking mechanism guided by error contribution values ​​and a rigorous anchor point selection mechanism, the starting boundary of error propagation can be accurately located, providing an accurate phase reference for subsequent local re-unwrapping and effectively avoiding secondary unwrapping failures caused by incorrect reference point selection.

[0121] Example 10

[0122] In one optional implementation, the process of screening candidate integer cycle jump combinations and generating a confidence marker map includes the steps of redetermining candidate integer cycle jump combinations in the backtracked region and generating an unwrap confidence marker map.

[0123] Step 1: When redetermining candidate integer cycle jump combinations for the region after rollback, the unwrapping phase of the high coherence confidence anchor point is used as the boundary constraint, and the comprehensive cost value of each candidate integer cycle jump combination is calculated respectively.

[0124] Here, a highly coherent, reliable anchor point can refer to a region unit determined during the reverse tracing of the aforementioned error propagation path, which meets the reliability conditions and whose integer cycle slip assignment results have not been cleared. Its unwrapped phase value remains fixed during local re-unwrapping and serves as the phase reference for the new solution region. A candidate integer cycle slip combination can refer to different integer multiples attempted for each region unit or pixel to be solved within the backtracking processing region. A set of phase offsets. Specifically, the system uses the boundary phase value of a highly coherent and reliable anchor point as a starting condition to construct multiple possible integer cycle slip allocation schemes within the back-off region. For example, for a back-off block containing N region cells, if each cell considers k possible integer cycle slip candidate values, then k... N There are several candidate combinations. In practical applications, dynamic programming or graph cut algorithms can be used to prune the branches and retain only a few candidate combinations that are topologically connected and have reasonable phase gradients.

[0125] The comprehensive cost is an indicator used to quantitatively evaluate the quality of each candidate integer cycle jump combination. The smaller the value, the more the combination conforms to physical reality and statistical regularity. This comprehensive cost is obtained by weighting coherence level cost, phase gradient consistency cost, closed loop error cost, boundary continuity cost, and shielding zone crossing penalty cost. Specifically, the coherence level cost is used to penalize the situation where cells in the candidate combination located in the low coherence risk region are given high confidence weights, prompting the solution results to tend to be consistent in the high coherence region; the phase gradient consistency cost is used to measure the difference between the absolute phase gradient generated by the candidate combination and the original entanglement phase gradient, and the smaller the difference, the lower the cost; the closed loop error cost is used to calculate whether the cumulative sum of integer cycle jumps in the closed loop formed by the candidate combination in the local region is zero, and a non-zero cumulative sum will result in a high penalty; the boundary continuity cost is used to evaluate the degree of phase smooth transition between the edge of the retreat region and the external fixed anchor point region, avoiding artificial steps at the boundary; the shielding zone crossing penalty cost applies an additional high penalty to the propagation path that crosses the non-untangled shielding region (such as water body, shadow region), preventing the untangling path from passing through the invalid data region. For example, the overall cost value is calculated using the following formula:

[0126] ;

[0127] in, to These are weighting coefficients that are adaptively adjusted based on land surface type. For the cost of coherence level, The cost of phase gradient consistency, As a cost for closed-loop error, As a cost of boundary continuity, The penalty for crossing the shielded zone is defined as follows: if a candidate combination causes the untangled path to cross the shielded zone marked as a water body, then... The item will increase significantly, thereby increasing the overall cost of the combination. It was eliminated because it far exceeded other combinations. By constructing this multi-dimensional cost function, we can deeply integrate coherence hierarchical information, geometric topological constraints, and physical prior knowledge to ensure that the selected local re-untangling results are mathematically optimal and physically reasonable.

[0128] Step 2: Select candidate integer cycle jump combinations whose comprehensive value meets the preset screening conditions to generate local re-unwrapping results, and generate corresponding unwrapping confidence marker maps based on whether the local re-unwrapping results have undergone backtracking verification, whether they have passed the closed loop verification, and whether they originate from low coherence risk areas.

[0129] The preset screening criteria can be that the comprehensive cost value is lower than a specific threshold, or that the comprehensive cost value is the smallest among all candidate combinations (i.e., globally optimal or locally optimal). The system compares the comprehensive cost values ​​of all candidate combinations, selects the set of integer cycle jump allocation schemes with the lowest cost as the final local re-unwrapping result, and fills this result into the corresponding back-off processing area in the initial unwrapping phase diagram, thereby completing the correction of the erroneous phase in that area.

[0130] The unwrapping reliability marker map is a quality evaluation layer that corresponds one-to-one with the spatial location of the final unwrapping phase map. It is used to identify the reliability of the unwrapping result for each region unit. The generation logic of this marker map is based on the historical trajectory and verification status of the local re-unwrapping result throughout the entire processing flow. Specifically, it includes: First, determining whether the region has undergone the aforementioned rollback processing step. If it has undergone rollback and successfully re-wrapped, it is marked as Level 2 reliable or reliable after verification, indicating that although the region once had risks, it has been repaired through multi-objective optimization. Second, determining whether the region has passed the closed loop verification after re-unwrapping, that is, checking whether the closed loop error formed by the newly generated local solution in its neighborhood meets the consistency condition. If it fails, it is marked as Level 3 risk or requires manual verification. Third, determining whether the unwrapping propagation source of the region directly originates from the low coherence risk area. If it mainly relies on the propagation of the low coherence area and has not undergone sufficient intermediate coherence area transition verification, even if it passes the cost screening, it is marked as Level 4 cautious use. For example, in densely built-up urban areas where construction interference caused unwrapping errors and which were corrected by this solution, a Level 1 or Level 2 trustworthy label can be generated because the area has undergone rigorous comprehensive cost screening and its boundaries are continuous with high-coherence anchor points. However, for remote mountainous areas with severe vegetation cover, where the comprehensive cost value is still at the critical value even after re-unwrapping, a Level 3 risk label is generated to indicate that the weight of the data in this area should be reduced or the data should be removed directly when applying deformation inversion downstream.

[0131] This step aims to ensure the global optimality of local repair results through a multi-objective optimization mechanism and provide users with transparent quality assessment criteria through explicit confidence marking. By comprehensively considering coherence, gradient, topological loop closure, and boundary constraints, it effectively avoids the problem of introducing new errors due to the lack of an evaluation system for local re-unwrapping in traditional methods. This significantly improves the robustness of InSAR automatic unwrapping in complex surface scenarios and provides a reliable data foundation for subsequent applications such as geological disaster monitoring and surface deformation analysis.

Claims

1. An automatic unwrapping method for InSAR based on coherence hierarchical constraints, characterized in that, include: S1. Obtain the InSAR interferometric winding phase map and coherence coefficient map of the area to be processed, and determine the coherence statistical characteristics of each pixel or region unit based on the coherence coefficient map. S2. Based on the coherence statistical characteristics, coherence spatial variation characteristics, residual point distribution characteristics, and entanglement phase gradient consistency characteristics, a coherence level map is generated. The coherence level map includes at least a high coherence confidence region, a medium coherence verification region, and a low coherence risk region. S3. Perform initial unwrapping processing on the interference entanglement phase map based on the coherence level map to obtain an initial unwrapped phase map, and record the unwrapping propagation source relationship between each pixel or region unit during the initial unwrapping processing; S4. Using each pixel or region unit as a node and the unwrapping propagation source relationship as a directed edge, construct a coherence level propagation graph and assign a propagation risk weight to the directed edge. The propagation risk weight is determined at least based on the coherence level difference between the nodes at both ends of the directed edge, the boundary coherence change, the boundary phase gradient change, and the residual point crossing situation. S5. Based on the coherence level propagation map, perform propagation anomaly verification on the initial unwrapped phase map. The propagation anomaly verification includes at least two of the following: closed loop error verification, coherence level boundary cycle slip mutation verification, and high coherence confidence region abnormal intrusion verification. S6. When it is determined that there is an abnormal propagation node or abnormal propagation edge according to the propagation anomaly verification, the region corresponding to the abnormal propagation node or abnormal propagation edge is used as the backtracking starting point, and the error propagation path is traced back along the directed edge of the coherence level propagation graph until the high coherence trust anchor point that meets the trust condition is reached. S7. Using the highly coherent trusted anchor point as the backoff reference, perform backoff processing on the region corresponding to the error propagation path. The backoff processing includes retaining the unwrapped phase of the highly coherent trusted anchor point and its upstream trusted region, clearing the integer cycle jump allocation result of the region corresponding to the error propagation path, and increasing the propagation cost of abnormal propagation edges in the error propagation path or setting them as temporary blocking edges. S8. Under the boundary constraints of the highly coherent and reliable anchor point, the candidate integer cycle jump combination is re-determined for the region after the back-off process, and the candidate integer cycle jump combination is screened according to the coherence level cost, phase gradient consistency cost, closed loop error cost and boundary continuity cost to obtain the local re-unwrapping result. S9. Update the local re-unwrapping result to the initial unwrapping phase map to obtain the final unwrapping phase map, and output the unwrapping confidence mark map corresponding to the final unwrapping phase map.

2. The method according to claim 1, characterized in that, In step S1, the pixel or region unit is obtained by regular window partitioning, connected component partitioning, or superpixel segmentation. For each of the aforementioned regional units, the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and winding phase gradient direction consistency are calculated, and the above calculation results are used as the coherence statistical characteristics, coherence spatial variation characteristics, residual point distribution characteristics, and winding phase gradient consistency characteristics.

3. The method according to claim 2, characterized in that, In step S2, a comprehensive confidence score is calculated for each of the aforementioned regional units. The comprehensive confidence score is obtained by weighting the average coherence coefficient, coherence coefficient variance, coherence gradient magnitude, residual point density, and the consistency of the winding phase gradient direction. Based on the comprehensive confidence score, each regional unit is divided into a high coherence confidence zone, a medium coherence verification zone, a low coherence risk zone, and an untangleable shielding zone. The untangleable shielding zone is not used as an untangle propagation channel.

4. The method according to claim 1, characterized in that, In step S3, when performing the initial unwrapping process, the high coherence confidence region is taken as the priority unwrapping region, the medium coherence verification region is taken as the propagable region, and the low coherence risk region is taken as the restricted propagation region. When the unwrapping phase of a certain region cell is obtained by propagation from an adjacent region cell, the adjacent region cell that provides the unwrapping reference is recorded as the parent node for the unwrapping propagation of that region cell, and the corresponding propagation direction, boundary phase difference, and integer cycle jump assignment value are recorded.

5. The method according to claim 4, characterized in that, In step S4, the propagation risk weight of the directed edge is determined as follows: The propagation risk weight is obtained by weighting and summing the coherence level crossings of the nodes at both ends of the directed edge, the difference in boundary coherence coefficients, the difference in boundary phase gradients, the number of residual points crossed, and whether or not a low-coherence-risk zone is crossed. Specifically, when the directed edge points from a low-coherence-risk region to a medium-coherence-verification region or a high-coherence-trust region, the level crossing penalty value of the directed edge is increased.

6. The method according to claim 1, characterized in that, In step S5, the closed-loop error verification includes: In the coherence level propagation diagram, a closed propagation loop consisting of adjacent nodes is selected, and the integer cycle jump difference accumulated along the closed propagation loop is calculated; When the cumulative integer cycle jump value of the closed propagation loop does not meet the closure consistency condition, the closed propagation loop is marked as an abnormal closed loop, and the directed edge that participates in the abnormal closed loop and has the highest propagation risk weight is determined as a candidate abnormal propagation edge.

7. The method according to claim 1, characterized in that, In step S5, the coherence level boundary cycle slip abrupt change verification includes: For directed edges that cross the boundaries between the high coherence confidence region, the medium coherence verification region, and the low coherence risk region, calculate the consistency between the unwrapping phase difference, the wrapping phase difference, and the integer cycle jump difference on both sides of the boundary. When an integer cycle jump difference occurs abruptly in the spatial neighborhood on both sides of the boundary, and the abrupt change is not supported by a coherent gradient change or a tangled phase gradient change, the corresponding directed edge is marked as an anomalous propagation edge.

8. The method according to claim 1, characterized in that, In step S5, the abnormal intrusion verification of the high coherence trusted region includes: Determine whether the unentanglement propagation source of the target's high coherence confidence region passes through a low coherence risk region; When the unwrapping propagation source of the target high coherence confidence region passes through a low coherence risk region, and the target high coherence confidence region has an integer multiple phase shift, a decrease in local phase continuity, or an increase in closed loop error relative to the adjacent high coherence confidence region, the target high coherence confidence region is identified as an abnormal intrusion region.

9. The method according to claim 1, characterized in that, In step S6, the reverse tracing error propagation path includes: Using the region corresponding to the abnormal propagation node or abnormal propagation edge as the backtracking starting point, backtrack level by level along the direction of the untangling propagation parent node; In each backtracking process, the error contribution value is determined based on the propagation risk weight of the directed edge, the closed loop error contribution value of the closed propagation loop in which it is located, and the coherence level crossing situation. The backtracking continues along the propagation direction with the largest error contribution value until a highly coherent trusted anchor point that meets the trusted conditions is reached. The trusted conditions include: the highly coherent trusted anchor point does not belong to an abnormal closed loop, was not obtained by direct propagation from a low-coherence risk region, and the integer cycle jump relationship between it and the adjacent highly coherent trusted region meets the consistency condition.

10. The method according to claim 1, characterized in that, In steps S8 and S9, when re-determining candidate integer cycle slip combinations for the region after backtracking, the unwrapped phase of the highly coherent and reliable anchor point is used as the boundary constraint, and the comprehensive cost value of each candidate integer cycle slip combination is calculated respectively. The overall cost is obtained by weighting the coherence level cost, phase gradient consistency cost, closed loop error cost, boundary continuity cost, and shielding zone crossing penalty cost. Candidate integer cycle jump combinations whose comprehensive cost value meets the preset screening conditions are selected to generate local re-unwrapping results. Based on whether the local re-unwrapping results have undergone backtracking verification, whether they have passed the closed loop verification, and whether they originate from the low coherence risk area, a corresponding unwrapping confidence marker map is generated.

Images