Building subsidence recognition method based on spatio-temporal feature fusion of multi-source InSAR images

Abstract

The application discloses a multi-source InSAR image building subsidence recognition method based on space-time feature fusion, and belongs to the field of synthetic aperture radar interferometry and building deformation monitoring. The method comprises the following steps: S1, obtaining radar line-of-sight deformation observation values at each structural line endpoint of a building; S2, constructing a joint adjustment function model with the three-dimensional displacement of the endpoint as a parameter; S3, introducing a structural line length invariable constraint as an adjustment strong condition; S4, iteratively adjusting and solving the optimal three-dimensional displacement solution of each endpoint; and S5, recognizing uneven subsidence according to the displacement vector difference of adjacent endpoints. The multi-source InSAR image building subsidence recognition method based on space-time feature fusion introduces the rigid geometric structure constraint of a building, reliably reverses one-dimensional observation of a radar line-of-sight into three-dimensional deformation of a building structure key point, and realizes fine recognition and quantitative analysis on uneven subsidence in the building.

Description

Technical Field

[0001] This invention specifically relates to a method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion, belonging to the field of synthetic aperture radar interferometry and building deformation monitoring. Background Technology

[0002] With the acceleration of urbanization, a large number of high-rise and large buildings are located in complex geological environments. Due to foundation compression, underground engineering activities, and long-term loads, the interiors of these buildings often exhibit spatially inconsistent differential settlement characteristics. How to achieve precise identification and quantitative analysis of uneven settlement within individual buildings using the large-scale, high-precision deformation observation capabilities of InSAR without deploying a large number of ground sensors has become a key technology for remote sensing mapping and urban safety monitoring. Existing remote sensing settlement monitoring methods, such as the one disclosed in Chinese Patent Publication No. CN106204539B, involve inverting urban building settlement based on morphological gradients. This method mainly relies on the strong reflection and strong coherence characteristics of buildings in high-resolution SAR images. It filters high-quality point targets through coherence coefficient thresholds and amplitude thresholds, and combines this with the small baseline set method to calculate the point target settlement information. Then, it utilizes morphological gradients... Gradient extraction of building boundaries to eliminate non-building point targets; this method has certain advantages in point target screening and building area determination, improving the automation and reliability of building settlement identification; however, the above technical solutions still take the line-of-sight one-dimensional settlement results as the analysis object, focusing on the spatial screening and classification of point targets on the building surface, and cannot establish the inherent geometric connection between settlement observation and building structure, making it difficult to analyze the real differential settlement state between different parts inside the building; due to the lack of explicit modeling of the rigid structural characteristics of the building, this type of method cannot effectively integrate multi-track InSAR observations into physically consistent three-dimensional deformation results, nor can it distinguish between overall rigid body motion and non-uniform deformation inside the structure, making it difficult to accurately identify and quantify differential settlement, which has the greatest impact on building safety; this fundamental defect limits its application depth in high-precision building structure settlement analysis and risk assessment scenarios. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a multi-source InSAR imagery-based building settlement identification method based on spatiotemporal feature fusion. By introducing rigid geometric constraints on the building, the one-dimensional observation along the radar line of sight is reliably inverted into the three-dimensional deformation of key points of the building structure, thereby enabling precise identification and quantitative analysis of uneven settlement within the building.

[0004] The present invention provides a method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion, the method comprising the following steps:

[0005] S1. Obtain radar line-of-sight deformation observations at the endpoints of each structural line of the building: From the preprocessed multi-source multi-track InSAR time-series deformation map, extract the feature structural lines covering the target building, obtain the radar line-of-sight deformation time-series observations corresponding to the endpoint pixel positions of the feature structural lines, and construct the endpoint-level time-series observation vector; This step specifically extracts all feature structural lines covering the target building from the preprocessed multi-source multi-track InSAR time-series deformation map, including the outer contour edge lines and main load-bearing axes of the target building; The radar line-of-sight deformation time-series observations corresponding to the endpoint pixel positions, after atmospheric phase correction and time-series denoising processing, are used as the original input data for subsequent three-dimensional solution;

[0006] S2. Construct a joint adjustment function model using the three-dimensional displacement of the endpoints as parameters: Establish a multi-track joint observation equation for the endpoints of each characteristic structure line, associate the radar line-of-sight deformation observations with the three-dimensional displacement component parameters of the endpoints (east-north-height), and stack the observation equations of all endpoints to construct the objective function of the global joint adjustment model. The core of this step is to establish a rigorous mathematical inversion model, associate the line-of-sight observations obtained in the first step under multiple different track geometries with the unknown three-dimensional displacement component parameters of each endpoint (east-north-height), and systematically integrate the multi-source one-dimensional observation equations into a unified three-dimensional solution framework by constructing a joint adjustment function model based on the least squares criterion.

[0007] S3. Introduce the constraint of constant structural line length as a strong adjustment condition: For adjacent endpoint pairs on the same structural line, construct constraint equations with constant structural line length. Linearize the constraint equations and introduce them as strong conditions into the objective function of the joint adjustment model to obtain an enhanced objective function with rigid constraints. This step is the key point for in-depth exploration to overcome the problem. That is, in the adjustment function model of the second step, physical constraints reflecting the rigidity characteristics of the building are introduced. It is assumed that the physical length of each main structural line remains unchanged during the monitoring period. This geometric condition is constructed as a strong constraint equation and given a very high weight ratio, thereby significantly enhancing the stability and ill-conditioned nature of the equation system and forcing the calculated three-dimensional displacement to meet the actual structural characteristics of the building.

[0008] S4. Iterative Adjustment to Solve for the Optimal 3D Displacement Solution at Each Endpoint: The iterative weighted least squares algorithm is used to solve the enhanced objective function with rigid constraints. By iteratively updating the observation weight matrix and eliminating outlier observations, the time series of the 3D displacement components at each structural line endpoint is output. This step performs the specific numerical solution process. The iterative weighted least squares algorithm is used to solve the constrained adjustment model established in the second and third steps. By iteratively eliminating outlier observations and adjusting the weight matrix, the high-precision 3D displacement component time series with clear physical meaning at each structural line endpoint is finally stably output.

[0009] S5. Identify uneven settlement based on displacement vector difference between adjacent endpoints: Based on the three-dimensional displacement vector of each endpoint, calculate the three-dimensional relative displacement vector difference between adjacent endpoints on the same structural line. Identify and quantify the uneven settlement distribution, differential settlement amount, and development trend of the building based on the relative displacement vector difference. This step completes the final uneven settlement identification. Based on the fine three-dimensional displacement vector of each endpoint obtained in step four, calculate the relative displacement vector difference between adjacent endpoints on the same structural line in three-dimensional space. By analyzing the magnitude, direction, and spatiotemporal evolution mode of the vector difference, accurately identify and quantify the uneven settlement distribution, differential settlement amount, and development trend between different parts of the building, thereby realizing a complete analytical closed loop from observation to identification.

[0010] Furthermore, the specific process of constructing the endpoint-level time series observation vector in S1 is as follows:

[0011] S11. Based on the building vector outline and structural design drawings, determine the precise location of each characteristic structural line in the geographic coordinate system, project it into the radar coordinate system of each orbit InSAR image, and obtain the pixel-level corresponding position of the structural line endpoint in each interferometric deformation map through geometric inversion, forming a consistent endpoint observation index across sensors and orbits.

[0012] S12. For each endpoint position, extract the radar line-of-sight deformation observation value of the corresponding pixel along the time dimension, and convert the phase dimension into displacement through the wavelength conversion formula, and map it to the same reference epoch.

[0013] S13. Introduce a time synchronization resampling operator to perform linear interpolation reconstruction on observation sequences sampled at non-uniform times from different orbits, thereby eliminating the impact of inconsistent time axes of multi-source data on the stability of subsequent adjustment.

[0014] S14. Introduce consistency constraint filtering in the direction of the structural line. By comparing the deformation gradient of adjacent pixels on the same structural line, filter out abnormal observations that are inconsistent with the overall structural motion trend. Finally, obtain the endpoint-level multi-track line-of-sight deformation time series that satisfies spatial consistency and temporal synchronization.

[0015] Furthermore, the execution of S1 is not merely a data reading process, but a directional observation construction process designed around the solvability of subsequent 3D inversion. Its core objective is to provide spatially highly consistent and temporally strictly aligned one-dimensional observation input for the rigid geometric constraint inversion of buildings in a multi-source, multi-orbit InSAR temporal deformation field. In specific implementation, firstly, based on the building's vector profile and structural design information, the positions of the building's edge line and main load-bearing axis are accurately determined in the geographic coordinate system. These are then projected onto the radar coordinate system of each orbit's InSAR image. Through geometric inversion, the pixel-level corresponding positions of the structural line endpoints in each interferometric deformation map are obtained, thus forming a consistent endpoint observation index across sensors and orbits. For each endpoint position, the line-of-sight deformation observation value, after atmospheric phase screen modeling and temporal filtering, is extracted along the time dimension and uniformly mapped to the same reference epoch to construct an endpoint-level temporal observation vector. The mathematical expression of the endpoint-level temporal observation vector is:

[0016] ;

[0017] in, Indicates the first The endpoint of the structure line is at the... At each orbital moment Radar line-of-sight deformation observations To remove the residual interferometric phase after terrain phase and orbital error, To correspond to the radar system wavelength, the purpose of this formula is to unify the phase dimensions into displacement quantities that can directly participate in geometric inversion, laying a physical consistency foundation for multi-track joint solutions. To ensure the temporal comparability of the same endpoint in observations from different tracks, a time synchronization resampling operator is introduced to reconstruct the observation sequence with non-uniform time sampling, in the form of:

[0018] ;

[0019] in These are the resampled endpoint observations. The interpolation weighting factor is determined jointly by the orbital time coverage relationship and the coherence weight. For the first Each time element represents a unified observation time after time synchronization resampling. This processing effectively eliminates the impact of inconsistent time axes of multi-source data on the stability of subsequent adjustment. Finally, to reduce the interference of local noise and residual unstructured deformation on endpoint observations, a consistency constraint filter is introduced along the structure line direction. By comparing the deformation gradients of adjacent pixels on the same structure line, abnormal observations inconsistent with the overall structural motion are filtered out. The discriminant function for comparing the deformation gradients of adjacent pixels on the same structure line is:

[0020] ;

[0021] in For the first The absolute deviation value of each observation point is used to measure the degree of deviation between the endpoint observation and the overall deformation trend of its corresponding structural line. The introduction of this value ensures that the input observation conforms to the overall structural characteristics of the building in space. Through the above processing, a multi-track line-of-sight deformation time series that meets the requirements of spatial consistency, temporal synchronization and physical interpretability is finally obtained, which provides a strict, stable and highly reliable original observation input for the next step of constructing a joint adjustment model with the endpoint three-dimensional displacement as the unknown. For the first Radar line-of-sight deformation observations at the endpoints of each structural line; For the first The radar line-of-sight deformation observations at the endpoints of each structural line are as follows: Other observation points along the same structural line; This refers to the total number of observation points participating in the calculation along the same structural line; It is the average of all observations along the same structural line.

[0022] Furthermore, based on the endpoint-level multi-track line-of-sight temporal observations already constructed in S1, S2 further completes a rigorous mapping from the observation space to the parameter space. Its goal is to uniformly express the one-dimensional deformation information under different orbital geometric conditions as three-dimensional displacement parameters (east, north, and altitude) of the same endpoint in a unified geographic coordinate system. The multi-track joint observation equations established for the endpoints of each characteristic structure line are as follows:

[0023] Will With the endpoint three-dimensional displacement vector Perform geometric correlation, For the first The endpoints of the structure line at time... The three-dimensional displacement vector is a column vector. For this endpoint at time The eastward displacement component, For this endpoint at time The northward displacement component, For this endpoint at time The vertical displacement component, To transpose the vector, we obtain a column vector; this relationship is explicitly expressed using the unit vector of the radar line-of-sight direction as follows:

[0024] ;

[0025] in For the first The unit vector of the line-of-sight direction of each orbit. It refers to the three-dimensional spatial directions of east, north, and vertical, which are uniquely determined by the corresponding orbital incident angle and azimuth angle. The residual observation error after consistency constraint filtering is represented by this formula. The purpose of this formula is to embed multi-track line-of-sight observations into a unified parameterized expression, providing a rigorous mathematical foundation for the linear superposition of multi-source information. To avoid the instability caused by independent solutions for each endpoint and each track, the observation equations for all endpoints at the same epoch are stacked together to construct a global joint adjustment model. The objective function for constructing the global joint adjustment model is expressed as:

[0026] ;

[0027] in This is a set vector of three-dimensional displacement parameters for all endpoints. Based on observational stability index Jointly determined with orbital coherence, the introduction of this objective function enables multi-source observations to participate in the three-dimensional inversion under a unified trade-off mechanism, effectively suppressing the impact of single-orbit geometric degradation on the solution results. Furthermore, to enhance the responsiveness of the solution process to temporal continuity, a cross-epoch smoothing regularization term is introduced in the joint adjustment to constrain the rationality of displacement changes between adjacent times at the same endpoint. Its expression is:

[0028] ;

[0029] This is the time-smoothing regularization term over the entire time series. The total number of time epochs involved in the adjustment. For the same endpoint at the previous moment The three-dimensional displacement vector; the role of this regularization term is to suppress the cumulative amplification of high-frequency noise in the time dimension, so that the inversion results can maintain continuous and stable physical evolution characteristics while satisfying the observation constraints; through the construction of the above joint adjustment function model, the solution of the endpoint three-dimensional displacement is transformed into an optimization problem with a clear structure and complete constraints, which lays a direct mathematical framework for the next step of introducing the rigid geometric constraint of constant structure line length and improving the overall solution stability.

[0030] Furthermore, based on the joint adjustment function model already formed in S2, S3 further injects rigid geometric constraints directly derived from the physical properties of the building into the model. Its core objective is to fundamentally alleviate the rank deficiency and ill-conditioned problems of multi-track line observation in 3D inversion through the verifiable and quantifiable constraint of the unchanged structural line length. The specific enhanced objective function with rigid constraints is as follows:

[0031] For any pair of endpoints on the same structural line and The spatial coordinate difference vector at the reference epoch It has been determined by architectural geometry, and at time Three-dimensional displacement of the lower two endpoints and The instantaneous spatial length acting on this structural line, and the condition for its invariance, are rigorously stated as follows:

[0032] ;

[0033] This expression directly characterizes the influence of the endpoint displacement difference on the physical length of the structural line. Its purpose is to explicitly embed the overall rigidity characteristics of the building into the adjustment system, ensuring that the three-dimensional displacement solution satisfies the observation equations while strictly adhering to the actual structural geometry. To facilitate unified solution with the linear joint adjustment model in s2, the aforementioned nonlinear constraints are linearized to the first order at the current iteration displacement estimation, resulting in an equality constraint form that can directly participate in the adjustment:

[0034] ;

[0035] in From the direction of the initial structure line A uniquely determined direction coefficient vector; the effect of this linearization process is to seamlessly embed it into the optimization framework constructed by S2 while maintaining the physical meaning of the constraint; to express the dominant role of this constraint in the overall solution, it is introduced as a strong condition into the extended objective function and given a weight much higher than that of the observation term, which is expressed as follows:

[0036] ;

[0037] in, The objective function of the global joint adjustment model is... For the set of all endpoint pairs of structure lines, The purpose of this construction method is to prioritize the rigid geometric consistency of the building through weight ratio adjustment, thereby significantly improving the stability and physical reliability of the three-dimensional displacement solution. By systematically introducing the constraint of invariant structural line length into the adjustment model, the underdetermined problem that originally relied solely on line-of-sight observation is transformed into a structurally controlled solvable problem, laying a solid mathematical and physical foundation for the next step of using the iterative weighted least squares algorithm to stably solve the optimal three-dimensional displacement of each endpoint.

[0038] Furthermore, the iterative adjustment to find the optimal three-dimensional displacement solution for each endpoint is as follows:

[0039] S4, based on the strongly constrained joint adjustment model already constructed in S2 and S3, performs a specific numerical solution process. Its goal is to obtain the optimal solution for the three-dimensional displacements of the endpoints, satisfying both multi-track observation consistency and structural rigidity constraints, through a stable and convergent iterative mechanism. During the solution process, all endpoints are considered as unknowns at the same epoch. As a global parameter vector, the Lagrange multipliers are introduced to incorporate the strong condition of invariant structure line length into the normal equation system, in the first... In the next iteration, a linearized incremental equation is constructed, which is uniformly expressed as:

[0040] ;

[0041] in The design matrix is ​​obtained by linearizing the S2 observation equation. for The transpose of the matrix, For the first The observation matrix at the next iteration This is the coefficient matrix of the structural line strong constraint equation in S3. for The transpose of the matrix, To strengthen the constraint matrix, For the first The three-dimensional displacement increment vector at each iteration is the unknown quantity that needs to be solved in this iteration. After solving, it is used to update the displacement solution. For the first The observation residual vector at the next iteration; the purpose of this equation is to balance the observation fitting accuracy and structural geometric consistency in the same linear system.

[0042] Furthermore, to enhance the solution's ability to suppress outlier observations, the observation weight matrix is ​​adaptively updated based on the residual statistical characteristics after each iteration. The influence of observations from different orbits and endpoints is dynamically adjusted using a residual normalization metric. The update rule is defined as follows:

[0043] ;

[0044] in For the first In the nth iteration The endpoint The residuals of orbit observations To correspond to the stability scale parameters of the observations, this weight update mechanism aims to gradually reduce the interference of observations inconsistent with the overall model on the solution results, thus concentrating the solution towards highly consistent observations. During the iteration process, the convergence of the endpoint displacement increments is used as the termination criterion. When the overall displacement correction satisfies the following formula, the solution is considered to have reached a stable state, as follows:

[0045] ;

[0046] in, For the first The three-dimensional displacement increment vector at the next iteration The preset convergence threshold is used to determine the output when the settlement reaches a stable state. This represents the optimal three-dimensional displacement solution of each structural line endpoint in the current epoch. A complete time series result can be formed through epoch recursion. The endpoint-level three-dimensional displacement obtained through this iterative adjustment process satisfies the rigid constraints of the building in space and remains continuous and stable in time, providing a direct and reliable vector input basis for the next step of identifying and quantifying uneven settlement of the building based on the displacement difference between adjacent endpoints.

[0047] Furthermore, the identification of uneven settlement based on the displacement vector difference between adjacent endpoints is specifically as follows:

[0048] Based on the three-dimensional displacement time series of the endpoints obtained in S4, S5 completes the quantitative identification and spatial discrimination of uneven settlement. It treats the building as a whole composed of multiple rigid structural lines coupled together, directly characterizing the differential deformation features within the structure through the displacement difference between adjacent endpoints. In specific implementation, adjacent endpoints on the same structural line... and At any moment Construct the three-dimensional relative displacement vector:

[0049] ;

[0050] in and The optimal three-dimensional displacement solution obtained after convergence of iterative weighted least squares adjustment is derived from the iterative adjustment results of S4. This vector fully preserves the directionality and amplitude information of differential settlement in space, aiming to isolate the overall rigid body motion component from the absolute displacement at the endpoints, retaining only the effective signal reflecting the inconsistent response within the structure. To further distinguish between the dominant components of differential deformation and vertical settlement along the structural line direction, a unit vector along the structural line direction is introduced. By decomposing the relative displacement into directions, a scalar measure of the uneven settlement is obtained:

[0051] ;

[0052] This scalar measure directly reflects the relative settlement trend of adjacent endpoints along the structural line direction. Its magnitude corresponds to the differential settlement intensity, and its sign corresponds to the difference in settlement direction. The effect of this formula is to compress three-dimensional vector information into a core indicator that can be used for engineering interpretation. To characterize the spatiotemporal evolution of uneven settlement, [the following is implied:] ... By introducing rate of change analysis over time, an index for differential settlement development is defined:

[0053] ;

[0054] This index describes the growth rate and evolution trend of differential settlement between adjacent endpoints, enabling a unified expression of static spatial distribution and dynamic development within the same analytical framework; it is achieved through systematic calculations on all structural lines and endpoint pairs. , and It can construct a spatial distribution map and temporal evolution sequence of uneven settlement inside a building, thereby enabling accurate identification of the location, amplitude and development direction of differential settlement, completing a complete analytical loop from multi-source InSAR observation to the identification of uneven settlement in buildings, and providing direct quantitative basis for subsequent risk assessment and early warning decisions.

[0055] Compared with existing technologies, the multi-source InSAR image building settlement identification method based on spatiotemporal feature fusion of the present invention has the following advantages:

[0056] 1. By mapping multi-source, multi-track InSAR line-of-sight one-dimensional deformation observations to the three-dimensional displacement parameter space of the building structure line endpoints, the limitation of traditional InSAR methods that can only invert unidirectional or two-dimensional deformations is broken. This enables reliable analysis of the true three-dimensional non-uniform settlement of buildings without relying on external sensors, significantly improving the physical integrity and engineering interpretability of deformation results.

[0057] 2. Using the endpoints of the building's structural lines as the basic analysis unit, and explicitly introducing the building's edge lines and main load-bearing axes into the solution process, the inversion model directly serves the building structure itself, rather than relying on large-scale pixel statistical assumptions. This effectively avoids the impact of mixed ground features in complex urban environments on the accuracy of settlement identification, and improves the spatial targeting and refinement of building-level monitoring.

[0058] 3. By introducing the rigid geometric constraint of constant structural line length as a strong adjustment condition, the inherent physical structural characteristics of the building are deeply embedded into the mathematical inversion model, which fundamentally improves the ill-conditioned and unstable problems that are common in multi-track InSAR three-dimensional inversion, and makes the solution results mathematically stable and convergent and physically strictly in line with the actual deformation mechanism of the building.

[0059] 4. An iterative weighted least squares solution framework integrating multi-track observation weights, adaptive residual suppression, and strong geometric constraints was constructed. This framework can automatically weaken the impact of abnormal observations under complex noise backgrounds and uneven observation quality, and stably output continuous and consistent endpoint three-dimensional displacement time series, significantly improving the reliability and robustness of long-term monitoring results.

[0060] 5. By directly characterizing the differential deformation state inside the building structure through the three-dimensional displacement vector difference between adjacent endpoints, the overall rigid body motion and uniform settlement components are successfully separated, enabling accurate identification of the location, direction and amplitude of uneven settlement. This upgrades settlement analysis from traditional single-point or average judgment to quantitative identification of differential responses inside the structure.

[0061] 6. Within a unified technical framework, a complete closed-loop analysis process has been achieved, from multi-source InSAR observation construction, three-dimensional deformation inversion, rigid constraint enhancement to uneven settlement identification. It can directly provide highly reliable data support for building safety assessment, differential settlement risk early warning and refined management of urban infrastructure, and has good engineering applicability and promotion value. Attached Figure Description

[0062] Figure 1 This is a flowchart of the multi-source InSAR endpoint observation data extraction and preprocessing process of the present invention.

[0063] Figure 2 This is a schematic diagram of the joint adjustment function model construction process of the present invention.

[0064] Figure 3 This is a schematic diagram of the rigid geometric constraint integration process of the present invention.

[0065] Figure 4 This is a schematic diagram of the iterative weighted least squares solution process of the present invention.

[0066] Figure 5 This is a schematic diagram of the process for identifying and analyzing uneven settlement of buildings according to the present invention. Detailed Implementation

[0067] like Figures 1 to 5 The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion, as shown, includes the following steps:

[0068] S1. Obtain radar line-of-sight deformation observations at the endpoints of each structural line of the building: From the preprocessed multi-source multi-track InSAR time-series deformation map, extract the feature structural lines covering the target building, obtain the radar line-of-sight deformation time-series observations corresponding to the endpoint pixel positions of the feature structural lines, and construct the endpoint-level time-series observation vector; This step specifically extracts all feature structural lines covering the target building from the preprocessed multi-source multi-track InSAR time-series deformation map, including the outer contour edge lines and main load-bearing axes of the target building; The radar line-of-sight deformation time-series observations corresponding to the endpoint pixel positions, after atmospheric phase correction and time-series denoising processing, are used as the original input data for subsequent three-dimensional solution;

[0069] S2. Construct a joint adjustment function model using the three-dimensional displacement of the endpoints as parameters: Establish a multi-track joint observation equation for the endpoints of each characteristic structure line, associate the radar line-of-sight deformation observations with the three-dimensional displacement component parameters of the endpoints (east-north-height), and stack the observation equations of all endpoints to construct the objective function of the global joint adjustment model. The core of this step is to establish a rigorous mathematical inversion model, associate the line-of-sight observations obtained in the first step under multiple different track geometries with the unknown three-dimensional displacement component parameters of each endpoint (east-north-height), and systematically integrate the multi-source one-dimensional observation equations into a unified three-dimensional solution framework by constructing a joint adjustment function model based on the least squares criterion.

[0070] S3. Introduce the constraint of constant structural line length as a strong adjustment condition: For adjacent endpoint pairs on the same structural line, construct constraint equations with constant structural line length. Linearize the constraint equations and introduce them as strong conditions into the objective function of the joint adjustment model to obtain an enhanced objective function with rigid constraints. This step is the key point for in-depth exploration to overcome the problem. That is, in the adjustment function model of the second step, physical constraints reflecting the rigidity characteristics of the building are introduced. It is assumed that the physical length of each main structural line remains unchanged during the monitoring period. This geometric condition is constructed as a strong constraint equation and given a very high weight ratio, thereby significantly enhancing the stability and ill-conditioned nature of the equation system and forcing the calculated three-dimensional displacement to meet the actual structural characteristics of the building.

[0071] S4. Iterative Adjustment to Solve for the Optimal 3D Displacement Solution at Each Endpoint: The iterative weighted least squares algorithm is used to solve the enhanced objective function with rigid constraints. By iteratively updating the observation weight matrix and eliminating outlier observations, the time series of the 3D displacement components at each structural line endpoint is output. This step performs the specific numerical solution process. The iterative weighted least squares algorithm is used to solve the constrained adjustment model established in the second and third steps. By iteratively eliminating outlier observations and adjusting the weight matrix, the high-precision 3D displacement component time series with clear physical meaning at each structural line endpoint is finally stably output.

[0072] S5. Identify uneven settlement based on displacement vector difference between adjacent endpoints: Based on the three-dimensional displacement vector of each endpoint, calculate the three-dimensional relative displacement vector difference between adjacent endpoints on the same structural line. Identify and quantify the uneven settlement distribution, differential settlement amount, and development trend of the building based on the relative displacement vector difference. This step completes the final uneven settlement identification. Based on the fine three-dimensional displacement vector of each endpoint obtained in step four, calculate the relative displacement vector difference between adjacent endpoints on the same structural line in three-dimensional space. By analyzing the magnitude, direction, and spatiotemporal evolution mode of the vector difference, accurately identify and quantify the uneven settlement distribution, differential settlement amount, and development trend between different parts of the building, thereby realizing a complete analytical closed loop from observation to identification.

[0073] The specific process for constructing the endpoint-level time series observation vector in S1 is as follows:

[0074] S11. Based on the building vector outline and structural design drawings, determine the precise location of each characteristic structural line in the geographic coordinate system, project it into the radar coordinate system of each orbit InSAR image, and obtain the pixel-level corresponding position of the structural line endpoint in each interferometric deformation map through geometric inversion, forming a consistent endpoint observation index across sensors and orbits.

[0075] S12. For each endpoint position, extract the radar line-of-sight deformation observation value of the corresponding pixel along the time dimension, and convert the phase dimension into displacement through the wavelength conversion formula, and map it to the same reference epoch.

[0076] S13. Introduce a time synchronization resampling operator to perform linear interpolation reconstruction on observation sequences sampled at non-uniform times from different orbits, thereby eliminating the impact of inconsistent time axes of multi-source data on the stability of subsequent adjustment.

[0077] S14. Introduce consistency constraint filtering in the direction of the structural line. By comparing the deformation gradient of adjacent pixels on the same structural line, filter out abnormal observations that are inconsistent with the overall structural motion trend. Finally, obtain the endpoint-level multi-track line-of-sight deformation time series that satisfies spatial consistency and temporal synchronization.

[0078] The execution of S1 is not merely a data reading process, but a directional observation construction process designed around the solvability of subsequent 3D inversion. Its core objective is to provide spatially highly consistent and temporally strictly aligned one-dimensional observation input for the rigid geometric constraint inversion of buildings in a multi-source, multi-orbit InSAR temporal deformation field. Specifically, based on the building's vector profile and structural design information, the positions of the building's edge line and main load-bearing axis are accurately determined in the geographic coordinate system. These are then projected onto the radar coordinate system of each orbit's InSAR image. Through geometric inversion, the pixel-level corresponding positions of the structural line endpoints in each interferometric deformation map are obtained, thus forming a consistent endpoint observation index across sensors and orbits. For each endpoint position, the line-of-sight deformation observation value, after atmospheric phase screen modeling and temporal filtering, is extracted along the time dimension and uniformly mapped to the same reference epoch, constructing an endpoint-level temporal observation vector. The mathematical expression of the endpoint-level temporal observation vector is:

[0079] ;

[0080] in, Indicates the first The endpoint of the structure line is at the... At each orbital moment Radar line-of-sight deformation observations To remove the residual interferometric phase after terrain phase and orbital error, To correspond to the radar system wavelength, the purpose of this formula is to unify the phase dimensions into displacement quantities that can directly participate in geometric inversion, laying a physical consistency foundation for multi-track joint solutions. To ensure the temporal comparability of the same endpoint in observations from different tracks, a time synchronization resampling operator is introduced to reconstruct the observation sequence with non-uniform time sampling, in the form of:

[0081] ;

[0082] in These are the resampled endpoint observations. The interpolation weighting factor is determined jointly by the orbital time coverage relationship and the coherence weight. For the first Each time element represents a unified observation time after time synchronization resampling. This processing effectively eliminates the impact of inconsistent time axes of multi-source data on the stability of subsequent adjustment. Finally, to reduce the interference of local noise and residual unstructured deformation on endpoint observations, a consistency constraint filter is introduced along the structure line direction. By comparing the deformation gradients of adjacent pixels on the same structure line, abnormal observations inconsistent with the overall structural motion are filtered out. The discriminant function for comparing the deformation gradients of adjacent pixels on the same structure line is:

[0083] ;

[0084] in For the first The absolute deviation value of each observation point is used to measure the degree of deviation between the endpoint observation and the overall deformation trend of its corresponding structural line. The introduction of this value ensures that the input observation conforms to the overall structural characteristics of the building in space. Through the above processing, a multi-track line-of-sight deformation time series that meets the requirements of spatial consistency, temporal synchronization and physical interpretability is finally obtained, which provides a strict, stable and highly reliable original observation input for the next step of constructing a joint adjustment model with the endpoint three-dimensional displacement as the unknown. For the first Radar line-of-sight deformation observations at the endpoints of each structural line; For the first The radar line-of-sight deformation observations at the endpoints of each structural line are as follows: Other observation points along the same structural line; This refers to the total number of observation points participating in the calculation along the same structural line; It is the average of all observations along the same structural line.

[0085] Building upon the endpoint-level multi-track line-of-sight temporal observations established in S1, S2 further completes a rigorous mapping from the observation space to the parameter space. Its goal is to uniformly express the one-dimensional deformation information under different orbital geometric conditions as three-dimensional displacement parameters (east, north, and altitude) of the same endpoint in a unified geographic coordinate system. The multi-track joint observation equations established for the endpoints of each characteristic structure line are as follows:

[0086] Will With the endpoint three-dimensional displacement vector Perform geometric correlation, For the first The endpoints of the structure line at time... The three-dimensional displacement vector is a column vector. For this endpoint at time The eastward displacement component, For this endpoint at time The northward displacement component, For this endpoint at time The vertical displacement component, To transpose the vector, we obtain a column vector; this relationship is explicitly expressed using the unit vector of the radar line-of-sight direction as follows:

[0087] ;

[0088] in For the first The unit vector of the line-of-sight direction of each orbit. It refers to the three-dimensional spatial directions of east, north, and vertical, which are uniquely determined by the corresponding orbital incident angle and azimuth angle. The residual observation error after consistency constraint filtering is represented by this formula. The purpose of this formula is to embed multi-track line-of-sight observations into a unified parameterized expression, providing a rigorous mathematical foundation for the linear superposition of multi-source information. To avoid the instability caused by independent solutions for each endpoint and each track, the observation equations for all endpoints at the same epoch are stacked together to construct a global joint adjustment model. The objective function for constructing the global joint adjustment model is expressed as:

[0089] ;

[0090] in This is a set vector of three-dimensional displacement parameters for all endpoints. Based on observational stability index Jointly determined with orbital coherence, the introduction of this objective function enables multi-source observations to participate in the three-dimensional inversion under a unified trade-off mechanism, effectively suppressing the impact of single-orbit geometric degradation on the solution results. Furthermore, to enhance the responsiveness of the solution process to temporal continuity, a cross-epoch smoothing regularization term is introduced in the joint adjustment to constrain the rationality of displacement changes between adjacent times at the same endpoint. Its expression is:

[0091] ;

[0092] This is the time-smoothing regularization term over the entire time series. The total number of time epochs involved in the adjustment. For the same endpoint at the previous moment The three-dimensional displacement vector; the role of this regularization term is to suppress the cumulative amplification of high-frequency noise in the time dimension, so that the inversion results can maintain continuous and stable physical evolution characteristics while satisfying the observation constraints; through the construction of the above joint adjustment function model, the solution of the endpoint three-dimensional displacement is transformed into an optimization problem with a clear structure and complete constraints, which lays a direct mathematical framework for the next step of introducing the rigid geometric constraint of constant structure line length and improving the overall solution stability.

[0093] Based on the joint adjustment function model already established in S2, S3 further injects rigid geometric constraints directly derived from the physical properties of the building into the model. Its core objective is to fundamentally alleviate the rank deficiency and ill-conditioned problems of multi-track line observation in 3D inversion through the verifiable and quantifiable constraint of the unchanged structural line length. The enhanced objective function with rigid constraints is as follows:

[0094] For any pair of endpoints on the same structural line and The spatial coordinate difference vector at the reference epoch It has been determined by architectural geometry, and at time Three-dimensional displacement of the lower two endpoints and The instantaneous spatial length acting on this structural line, and the condition for its invariance, are rigorously stated as follows:

[0095] ;

[0096] This expression directly characterizes the influence of the endpoint displacement difference on the physical length of the structural line. Its purpose is to explicitly embed the overall rigidity characteristics of the building into the adjustment system, ensuring that the three-dimensional displacement solution satisfies the observation equations while strictly adhering to the actual structural geometry. To facilitate unified solution with the linear joint adjustment model in s2, the aforementioned nonlinear constraints are linearized to the first order at the current iteration displacement estimation, resulting in an equality constraint form that can directly participate in the adjustment:

[0097] ;

[0098] in From the direction of the initial structure line A uniquely determined direction coefficient vector; the effect of this linearization process is to seamlessly embed it into the optimization framework constructed by S2 while maintaining the physical meaning of the constraint; to express the dominant role of this constraint in the overall solution, it is introduced as a strong condition into the extended objective function and given a weight much higher than that of the observation term, which is expressed as follows:

[0099] ;

[0100] in The objective function of the global joint adjustment model is... For the set of all endpoint pairs of structure lines, The purpose of this construction method is to prioritize the rigid geometric consistency of the building through weight ratio adjustment, thereby significantly improving the stability and physical reliability of the three-dimensional displacement solution. By systematically introducing the constraint of invariant structural line length into the adjustment model, the underdetermined problem that originally relied solely on line-of-sight observation is transformed into a structurally controlled solvable problem, laying a solid mathematical and physical foundation for the next step of using the iterative weighted least squares algorithm to stably solve the optimal three-dimensional displacement of each endpoint.

[0101] The iterative adjustment process for finding the optimal three-dimensional displacement solution for each endpoint is as follows:

[0102] S4, based on the strongly constrained joint adjustment model already constructed in S2 and S3, performs a specific numerical solution process. Its goal is to obtain the optimal solution for the three-dimensional displacements of the endpoints, satisfying both multi-track observation consistency and structural rigidity constraints, through a stable and convergent iterative mechanism. During the solution process, all endpoints are considered as unknowns at the same epoch. As a global parameter vector, the Lagrange multipliers are introduced to incorporate the strong condition of invariant structure line length into the normal equation system, in the first... In the next iteration, a linearized incremental equation is constructed, which is uniformly expressed as:

[0103] ;

[0104] in The design matrix is ​​obtained by linearizing the S2 observation equation. for The transpose of the matrix, For the first The observation matrix at the next iteration This is the coefficient matrix of the structural line strong constraint equation in S3. for The transpose of the matrix, To strengthen the constraint matrix, For the first The three-dimensional displacement increment vector at each iteration is the unknown quantity that needs to be solved in this iteration. After solving, it is used to update the displacement solution. For the first The observation residual vector at the next iteration; the purpose of this equation is to balance the observation fitting accuracy and structural geometric consistency in the same linear system.

[0105] To enhance the solution's ability to suppress outlier observations, the observation weight matrix is ​​adaptively updated based on the residual statistical characteristics after each iteration. The influence of observations from different orbits and endpoints is dynamically adjusted using a residual normalization metric. The update rule is defined as follows:

[0106] ;

[0107] in For the first In the nth iteration The endpoint The residuals of orbit observations To correspond to the stability scale parameters of the observations, this weight update mechanism aims to gradually reduce the interference of observations inconsistent with the overall model on the solution results, thus concentrating the solution towards highly consistent observations. During the iteration process, the convergence of the endpoint displacement increments is used as the termination criterion. When the overall displacement correction satisfies the following formula, the solution is considered to have reached a stable state, as follows:

[0108] ;

[0109] in, For the first The three-dimensional displacement increment vector at the next iteration The preset convergence threshold is used to determine the output when the settlement reaches a stable state. This represents the optimal three-dimensional displacement solution of each structural line endpoint in the current epoch. A complete time series result can be formed through epoch recursion. The endpoint-level three-dimensional displacement obtained through this iterative adjustment process satisfies the rigid constraints of the building in space and remains continuous and stable in time, providing a direct and reliable vector input basis for the next step of identifying and quantifying uneven settlement of the building based on the displacement difference between adjacent endpoints.

[0110] The specific steps for identifying uneven settlement based on the displacement vector difference between adjacent endpoints are as follows:

[0111] Based on the three-dimensional displacement time series of the endpoints obtained in S4, S5 completes the quantitative identification and spatial discrimination of uneven settlement. It treats the building as a whole composed of multiple rigid structural lines coupled together, directly characterizing the differential deformation features within the structure through the displacement difference between adjacent endpoints. In specific implementation, adjacent endpoints on the same structural line... and At any moment Construct the three-dimensional relative displacement vector:

[0112] ;

[0113] in and The optimal three-dimensional displacement solution obtained after convergence of iterative weighted least squares adjustment is derived from the iterative adjustment results of S4. This vector fully preserves the directionality and amplitude information of differential settlement in space, aiming to isolate the overall rigid body motion component from the absolute displacement at the endpoints, retaining only the effective signal reflecting the inconsistent response within the structure. To further distinguish between the dominant components of differential deformation and vertical settlement along the structural line direction, a unit vector along the structural line direction is introduced. By decomposing the relative displacement into directions, a scalar measure of the uneven settlement is obtained:

[0114] ;

[0115] This scalar measure directly reflects the relative settlement trend of adjacent endpoints along the structural line direction. Its magnitude corresponds to the differential settlement intensity, and its sign corresponds to the difference in settlement direction. The effect of this formula is to compress three-dimensional vector information into a core indicator that can be used for engineering interpretation. To characterize the spatiotemporal evolution of uneven settlement, [the following is implied:] ... By introducing rate of change analysis over time, an index for differential settlement development is defined:

[0116] ;

[0117] This index describes the growth rate and evolution trend of differential settlement between adjacent endpoints, enabling a unified expression of static spatial distribution and dynamic development within the same analytical framework; it is achieved through systematic calculations on all structural lines and endpoint pairs. , and It can construct a spatial distribution map and temporal evolution sequence of uneven settlement inside a building, thereby enabling accurate identification of the location, amplitude and development direction of differential settlement, completing a complete analytical loop from multi-source InSAR observation to the identification of uneven settlement in buildings, and providing direct quantitative basis for subsequent risk assessment and early warning decisions.

[0118] Example 1:

[0119] Taking a 32-story reinforced concrete shear wall structure high-rise residential building as the monitoring object, the building has a total height of 98.6m and a plan dimension of 45m×18m. It is located along the city's Metro Line 10, and there are deep foundation pit construction activities in the surrounding area, posing a risk of uneven settlement. Long-term deformation monitoring and uneven settlement identification are required. This embodiment acquires multi-source, multi-track InSAR time-series deformation data, including: Sentinel-1A / B ascending orbit data (track number 12, time range from January 2023 to December 2024, a total of 48 scenes, VV polarization, spatial resolution 5m×20m) and Sentinel-1A / B descending orbit data (track number 91, time range from January 2023 to December 2024). The dataset includes 46 scenes from June 2023 to December 2024, with VV polarization and a spatial resolution of 5m×20m; 24 scenes from June 2023 to December 2024, with HH polarization and a spatial resolution of 1m×1m; and 8 structural lines from the building, including 4 outer contour edges, 2 main load-bearing shear wall axes along the building's length, and 2 main load-bearing shear wall axes along the building's width, corresponding to 16 structural line endpoints. The precise geographic coordinates of each endpoint were determined in the CGCS2000 coordinate system. The strong constraint weight factor was set to 1000 (initial weight of the observation term was 10); the strong constraint weight factor was set to 1200; the smoothing coefficient of the regularization expression was set to 0.1; and the convergence threshold was set to... m; the line-of-sight deformation observations are correlated with the three-dimensional displacement parameters of East (E), North (N), and Altitude (U) to establish observation equations and complete the setting of the orbit's incident angle and azimuth angle. In this embodiment, the Sentinel-1 ascending orbit incident angle is 39.2° and the azimuth angle is 10.3°; the Sentinel-1 descending orbit incident angle is 34.7° and the azimuth angle is −168.2°; the Gaofen-3 ascending orbit incident angle is 42.5° and the azimuth angle is 3.7°.

[0120] Results: After calculating the deformation identification method of the present invention for the 8 structural lines and 16 endpoints in this embodiment, a spatial distribution map and temporal evolution curve of the building's uneven settlement were constructed. The results show that the building as a whole exhibits a uniform settlement trend, with an average vertical settlement rate of -2.3 mm / year. The maximum differential settlement between adjacent endpoints at the southwest corner of the building is 8.6 mm, and the differential settlement rate is 0.6 mm / month, which is lower than the allowable value for differential settlement of high-rise buildings specified in the "Code for Design of Building Foundations" GB50007-2011, indicating that the structure is in a safe state. The differential settlement rate of the local endpoints at the northeast corner showed a slight increase from June to September 2024, which highly coincides with the excavation period of the surrounding deep foundation pit, verifying the accurate identification capability of this method for the spatiotemporal evolution characteristics of uneven settlement.

[0121] Example 2:

[0122] Taking a 24-story frame-core tube structure commercial office building as the monitoring object, the building has a total height of 86.2m and a plan dimension of 52m×26m. It is located in the Yangtze River floodplain landform unit, with complex geological conditions and a risk of uneven settlement of soft soil foundation. The monitoring period is from June 2022 to December 2024. The multi-source InSAR data used in this embodiment include: 56 scenes each of Sentinel-1A / B ascending and descending orbit data, with a time range from June 2022 to December 2024; and 22 scenes of ALOS-2 ascending orbit data, with a spatial resolution of 3m×3m, with a time range from July 2022 to November 2024. Based on the building structural drawings, 10 feature structural lines were extracted, including 6 outer contour edge lines and 4 core tube main load-bearing axes, for a total of 20 structural line endpoints. The strong constraint weight factor was set to 1200, the smoothing coefficient of the regularization expression was set to 0.15, and the convergence threshold was set to... m; other data are consistent with Example 1.

[0123] The final results of this embodiment show that the three-dimensional displacement solution obtained by this method, compared with the measured data from 12 GNSS monitoring points on site, has a vertical root mean square error of 1.9 mm and a horizontal root mean square error of 2.9 mm, which is highly consistent with the measured data. It successfully identified that the differential settlement rate at two adjacent endpoints on the north side of the building exceeded 1.2 mm / month, providing accurate data support for building safety assessment and foundation reinforcement treatment in a timely manner, and verifying the engineering practicality and reliability of this method.

[0124] The above embodiments are merely preferred embodiments of the present invention. Therefore, all equivalent changes or modifications made to the structure, features and principles described in the claims of the present invention are included within the scope of the present invention.

Claims

1. A method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion, characterized in that: The method includes the following steps: S1. Obtain radar line-of-sight deformation observations at the endpoints of each structural line of the building: Extract the feature structural lines covering the target building from the pre-processed multi-source multi-orbit InSAR time-series deformation map, obtain the radar line-of-sight deformation time-series observations corresponding to the endpoint pixel positions of the feature structural lines, and construct the endpoint-level time-series observation vector. S2. Construct a joint adjustment function model using the three-dimensional displacement of the endpoints as parameters: Establish a multi-track joint observation equation for the endpoints of each characteristic structure line, associate the radar line-of-sight deformation observation value with the three-dimensional displacement component parameters of the east-north-height of the endpoints, stack the observation equations of all endpoints as a whole, and construct the objective function of the global joint adjustment model. S3. Introduce the constraint of constant structural line length as a strong condition for adjustment: For adjacent endpoint pairs on the same structural line, construct constraint equations with constant structural line length. Linearize the constraint equations and introduce them as strong conditions into the objective function of the joint adjustment model to obtain an enhanced objective function with rigid constraints. S4. Iterative adjustment to solve the optimal three-dimensional displacement solution for each endpoint: The iterative weighted least squares algorithm is used to solve the enhanced objective function with rigid constraints. By iteratively updating the observation weight matrix and eliminating gross observations, the time series of the three-dimensional displacement components of each structural line endpoint is output. S5. Identify uneven settlement based on the displacement vector difference between adjacent endpoints: Based on the three-dimensional displacement vector of each endpoint, calculate the three-dimensional relative displacement vector difference between adjacent endpoints on the same structural line, and identify and quantify the uneven settlement distribution, differential settlement amount and development trend of the building based on the relative displacement vector difference. The enhanced objective function with rigid constraints is as follows: For any pair of endpoints on the same structural line and The spatial coordinate difference vector at the reference epoch It has been determined by architectural geometry, and at time Three-dimensional displacement of the lower two endpoints and The constraint equations that collectively affect the instantaneous spatial length of the structural line, while keeping the length of the structural line constant, are as follows: ； To facilitate a unified solution for the linear joint adjustment model, the constraint equations with invariant structural line lengths are linearized to the first order at the current iteration displacement estimation, resulting in an equality constraint form that can be directly used in the adjustment: ； in From the direction of the initial structure line A uniquely determined direction coefficient vector; to express the dominant role of this constraint in the overall solution, it is introduced as a strong condition into the extended objective function and given a weight much higher than that of the observation term. Its comprehensive expression is: ； in The objective function of the global joint adjustment model is... For the set of all endpoint pairs of structure lines, It is a strong constraint factor.

2. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 1, characterized in that: The characteristic structural lines in S1 include the outer contour edge line of the target building and the main load-bearing axis; the radar line-of-sight deformation time series observation values ​​are the observation values ​​after atmospheric phase correction and time-series denoising processing.

3. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 1, characterized in that: The specific process for constructing the endpoint-level time series observation vector in S1 is as follows: S11. Based on the building vector outline and structural design drawings, determine the precise location of each characteristic structural line in the geographic coordinate system, project it into the radar coordinate system of each orbit InSAR image, and obtain the pixel-level corresponding position of the structural line endpoint in each interferometric deformation map through geometric inversion, forming a consistent endpoint observation index across sensors and orbits. S12. For each endpoint position, extract the radar line-of-sight deformation observation value of the corresponding pixel along the time dimension, and convert the phase dimension into displacement through the wavelength conversion formula, and map it to the same reference epoch. S13. Introduce a time synchronization resampling operator to perform linear interpolation reconstruction on observation sequences sampled at non-uniform times from different orbits, thereby eliminating the impact of inconsistent time axes of multi-source data on the stability of subsequent adjustment. S14. Introduce consistency constraint filtering in the direction of the structural line. By comparing the deformation gradient of adjacent pixels on the same structural line, filter out abnormal observations that are inconsistent with the overall structural motion trend. Finally, obtain the endpoint-level multi-track line-of-sight deformation time series that satisfies spatial consistency and temporal synchronization.

4. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 3, characterized in that: The mathematical expression for the endpoint-level time series observation vector is: ； in Indicates the first The endpoint of the structure line is at the... At each orbital moment Radar line-of-sight deformation observations To remove the residual interferometric phase after terrain phase and orbital error, To correspond to the radar system wavelength and ensure time sequence comparability of the same endpoint in observations from different orbits, a time synchronization resampling operator is introduced to reconstruct the observation sequence with non-uniform time sampling, in the form of: ； in These are the resampled endpoint observations. The interpolation weighting factor is determined jointly by the orbital time coverage relationship and the coherence weight. For the first Each time element represents a unified observation time after time synchronization and resampling. The discriminant function for comparing the deformation gradients of adjacent pixels on the same structural line is: ； in For the first The absolute deviation value of each observation point is used to measure the degree of deviation between the endpoint observation and the overall deformation trend of its corresponding structural line. For the first Radar line-of-sight deformation observations at the endpoints of each structural line; For the first The radar line-of-sight deformation observations at the endpoints of each structural line are as follows: Other observation points along the same structural line; This refers to the total number of observation points participating in the calculation along the same structural line; It is the average of all observations along the same structural line.

5. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 4, characterized in that: The specific steps for establishing a multi-track joint observation equation for the endpoints of each characteristic structure line are as follows: Will With the endpoint three-dimensional displacement vector Perform geometric correlation, For the first The endpoints of the structure line at time... The three-dimensional displacement vector is a column vector. For this endpoint at time The eastward displacement component, For this endpoint at time The northward displacement component, For this endpoint at time The vertical displacement component, To transpose the vector, we obtain a column vector; this relationship is explicitly expressed using the unit vector of the radar line-of-sight direction as follows: ； in For the first The unit vector of the line-of-sight direction of each orbit. It refers to the three-dimensional spatial directions of east, north, and vertical, which are uniquely determined by the corresponding orbital incident angle and azimuth angle. The residual observation error is the result of consistency constraint filtering; the objective function for constructing the global joint adjustment model is expressed as: ； in This is a set vector of three-dimensional displacement parameters for all endpoints. Based on observational stability index In conjunction with orbital coherence, a cross-epoch smoothing regularization term is introduced into the joint adjustment to constrain the rationality of displacement changes between adjacent times at the same endpoint. Its expression is: ； This is the time-smoothing regularization term over the entire time series. The total number of time epochs involved in the adjustment. For the same endpoint at the previous moment The three-dimensional displacement vector.

6. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 5, characterized in that: The iterative adjustment process for finding the optimal three-dimensional displacement solution for each endpoint is as follows: When solving, all unknowns with endpoints at the same epoch are considered. As a global parameter vector, the Lagrange multipliers are introduced to incorporate the strong condition of invariant structure line length into the normal equation system, in the first... In the next iteration, a linearized incremental equation is constructed, which is uniformly expressed as: ； in The design matrix is ​​obtained by linearizing the S2 observation equation. for The transpose of the matrix, For the first The observation matrix at the next iteration This is the coefficient matrix of the structural line strong constraint equation in S3. for The transpose of the matrix, To strengthen the constraint matrix, For the first The three-dimensional displacement increment vector at each iteration is the unknown quantity that needs to be solved in this iteration. After solving, it is used to update the displacement solution. For the first The observed residual vector at the next iteration.

7. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 6, characterized in that: After each iteration, the observation weight matrix is ​​adaptively updated based on the residual statistical properties. The update rule is defined as follows: ； in For the first In the nth iteration The endpoint The residuals of orbit observations To correspond to the stability scale parameters observed, the convergence of the endpoint displacement increments is used as the termination criterion during the iteration process. The solution is considered to have reached a stable state when the overall displacement correction satisfies the following formula, as follows: ； in, For the first The three-dimensional displacement increment vector at the next iteration The preset convergence threshold is used to determine the output when the settlement reaches a stable state. This represents the optimal three-dimensional displacement solution of each structural line endpoint in the current epoch. A complete time series result can be formed through epoch recursion.

8. The method for identifying building settlement based on multi-source InSAR imagery using spatiotemporal feature fusion according to claim 7, characterized in that, The specific steps for identifying uneven settlement based on the displacement vector difference between adjacent endpoints are as follows: For adjacent endpoints on the same structural line and At any moment Construct the three-dimensional relative displacement vector: ； in and The optimal three-dimensional displacement solution was obtained after convergence of iterative weighted least squares adjustment. To further distinguish between the differential deformation along the structural line direction and the dominant vertical settlement component, a unit vector along the structural line direction was introduced. By decomposing the relative displacement into directions, a scalar measure of the uneven settlement is obtained: ； To characterize the spatiotemporal evolution of uneven settlement, By introducing rate of change analysis over time, an index for differential settlement development is defined: ； Through systematic calculations on all structural lines and endpoint pairs , and It can construct a spatial distribution map and temporal evolution sequence of uneven settlement inside a building, thereby enabling accurate identification of the location, amplitude, and development direction of differential settlement.

Images