A method and system for correcting three-dimensional subsidence errors by fusing air and ground data

Abstract

The application discloses a kind of three-dimensional settlement error correction method and system of fusion air-ground data.First, initial radar point cloud is generated based on time sequence radar image, and combined with adaptive calibration deployment artificial corner reflector and its synchronous positioning data, decoupling is calculated out three-dimensional space error matrix;Subsequently, the absolute coordinate point cloud set containing time sequence deformation information is output through global space conversion model;Combined with error covariance and coherence coefficient, anisotropic dynamic error ellipsoid is constructed, and candidate entity nodes are extracted in three-dimensional digital model;Finally, the similarity of measured deformation sequence and equipment prior deformation sequence is calculated using dynamic time warping algorithm, and the unique confirmation of point cloud attribution and data mounting are completed.The application solves the problem of radar space projection distortion, and realizes the cross-modal fusion of large-scale microwave remote sensing data and three-dimensional entity model.

Description

Technical Field

[0001] This invention relates to the field of spatial information processing technology, specifically to a three-dimensional settlement error correction method and system that integrates air and ground data. Background Technology

[0002] With the continuous construction and expansion of large-scale hub substations and power transmission and transformation projects, higher precision requirements have been placed on the three-dimensional settlement monitoring of structures within the stations and the surrounding ground surface. Currently, Synthetic Aperture Radar Interferometry (InSAR) technology has become the mainstream technology for surface deformation monitoring due to its advantages of wide coverage, non-contact operation, and all-weather observation capabilities; at the same time, Global Navigation Satellite Systems (GNSS) are also widely used to provide absolute elevation and positioning benchmarks for the Earth's surface.

[0003] In typical open environments, existing technologies can generally perform deformation monitoring effectively using the two techniques mentioned above. However, directly applying existing methods to the unique high-density, complex three-dimensional environment of substations still faces the following technical bottlenecks: First, achieving refined spatial coordination of heterogeneous air-to-ground data presents significant challenges. Radar-acquired deformation sequences are natively based on the microwave line-of-sight (LOS) coordinate system, while ground-based satellite positioning systems provide absolute geodetic coordinates. Existing technologies for data fusion often employ conventional global translation or simple linear transformation models. While this approach works well in plains or simple terrain, it struggles to accurately separate orthogonal errors in the range, azimuth, and altitude directions in areas with complex micro-topography and electromagnetic interference, leading to residual local nonlinear coordinate distortions during three-dimensional spatial transformations.

[0004] Second, spatial matching of 3D point clouds in densely populated equipment areas presents multiple solutions. Substations are characterized by a high density of various facilities with significant spatial overlap. Due to the inherent asymmetry in resolution across different spatial directions caused by radar side-looking imaging mechanisms, traditional nearest neighbor search algorithms based on isotropic or purely geometric distances are prone to boundary intersections when dealing with such densely overlapping areas, making it difficult to uniquely determine the attribution of settlement points. Relying solely on spatial distance features, the system struggles to distinguish adjacent equipment with drastically different physical properties and deformation patterns, and occasional mismatches may reduce the reliability of subsequent construction safety early warning systems.

[0005] In summary, there is an urgent need for a settlement error correction method that can effectively integrate heterogeneous air-to-ground data and achieve accurate matching between radar point clouds and digital entities in complex 3D scenes. Summary of the Invention

[0006] The present invention aims to solve the technical problem that local nonlinear coordinate distortion is easily generated when existing heterogeneous air-ground data is fused with spatial reference; at the same time, it solves the technical problem that radar point clouds are prone to spatial mismatch and multiple matching solutions in three-dimensional scenes with dense facilities such as substations and spatial overlap due to the limitation of radar physical asymmetric resolution.

[0007] The present invention adopts the following technical solution.

[0008] Firstly, a three-dimensional settlement error correction method integrating air and ground data includes: The temporal synthetic aperture radar imagery of the target area and the synchronized ground absolute positioning data are collected, and an initial temporal radar point cloud is generated based on the temporal synthetic aperture radar imagery. Extract the true three-dimensional coordinates of the ground reference points in the ground absolute positioning data, and perform spatial registration and deviation analysis with the observed three-dimensional coordinates corresponding to the initial temporal radar point cloud set to construct a three-dimensional spatial error matrix. A global spatial transformation model is generated based on the three-dimensional spatial error matrix. The global spatial transformation model is used to perform global coordinate transformation on the initial temporal radar point set, and the output is an absolute coordinate point set containing absolute three-dimensional coordinates and time series deformation information. For the point cloud to be matched in the absolute coordinate point cloud set, a dynamic error ellipsoid is constructed with the absolute three-dimensional coordinates of the point cloud to be matched as the origin and the covariance data of the three-dimensional spatial error matrix; in the pre-imported three-dimensional digital model, a nearest neighbor search is performed on the internal space of the dynamic error ellipsoid to extract candidate entity nodes. Calculate the sequence similarity between the time-series deformation information of the point cloud to be matched and the preset allowable deformation features of each candidate entity node, and attach the time-series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity to obtain the settlement error correction result for the candidate entity node.

[0009] Preferably, generating an initial temporal radar point cloud based on the temporal synthetic aperture radar imagery includes: Image registration is performed on the temporal synthetic aperture radar image of the target area, and interferometric pairs are combined according to a preset spatiotemporal baseline threshold to generate an interferometric atlas covering the target area; The interferogram is subjected to differential processing, phase decoherence suppression, and temporal phase unwrapping in sequence to extract the phase evolution sequence of each pixel at multiple sampling times; By combining the external terrain parameters of the target area, the phase evolution sequence is converted into line-of-sight deformation values, and the line-of-sight deformation values ​​are mapped to a three-dimensional coordinate space according to the radar satellite orbit parameters of the target area, thus obtaining the initial time-series radar point set containing the observed three-dimensional coordinates and time-series deformation information.

[0010] Preferably, the ground reference point is an artificial corner reflector integrated with a satellite positioning device; before acquiring time-series synthetic aperture radar imagery of the target area and synchronized ground absolute positioning data, the method further includes an adaptive calibration and deployment step for the artificial corner reflector: Collect the radar wavelength, orbital inclination, and local latitude of the target area from the target radar satellite; Combining the preset radar backscattering intensity threshold with the radar wavelength, the optimal edge length of the artificial corner reflector is calculated based on the maximum radar cross section model. Based on the cosine theorem of a spherical right triangle, the true north azimuth angle of the artificial corner reflector is calculated using the orbital inclination angle and the local latitude, and the elevation angle is determined based on the geometric angle between the incident wave of the target radar satellite and the bottom surface of the artificial corner reflector. The artificial corner reflector is deployed according to the calculated optimal edge length, true north azimuth angle, and elevation angle, so as to synchronously collect image features and ground absolute positioning data through the artificial corner reflector and integrated satellite positioning equipment.

[0011] Preferably, the true three-dimensional coordinates of the ground reference points in the ground absolute positioning data are extracted, and spatial registration and deviation analysis are performed with the observed three-dimensional coordinates corresponding to the initial temporal radar point cloud set to construct a three-dimensional spatial error matrix, including: Separate the positioning two-dimensional coordinates and absolute elevation of the ground reference point from the actual three-dimensional coordinates; Extract the two-dimensional radar coordinates and interferometric radar height of the ground reference point in the radar image system from the observed three-dimensional coordinates; The difference between the positioning two-dimensional coordinates and the radar two-dimensional coordinates is calculated to obtain the range error component and the azimuth error component in the microwave line-of-sight plane. The difference between the absolute elevation and the interferometric radar height is calculated to obtain the height error component perpendicular to the microwave line-of-sight plane. The three-dimensional spatial error matrix is ​​constructed using the calculated range error component, azimuth error component, and altitude error component as orthogonal references.

[0012] Preferably, constructing a dynamic error ellipsoid using the covariance data of the three-dimensional spatial error matrix includes: Calculate the covariance data of the range error component, the azimuth error component, and the altitude error component in the three-dimensional spatial error matrix; Using the absolute three-dimensional coordinates of the point cloud to be matched as the geometric center, a local spatial retrieval coordinate system is established, which includes the microwave line-of-sight axis, the azimuth trajectory axis, and the vertical elevation axis. The covariance data of the range error component, the covariance data of the azimuth error component, and the covariance data of the altitude error component are mapped to the microwave line-of-sight axis, the azimuth trajectory axis, and the vertical elevation axis, respectively, and the initial length of the semi-axis of the dynamic error ellipsoid is calculated. Collect the coherence coefficients of the current time-series synthetic aperture radar image, and calculate the dynamic compensation weights based on the coherence coefficients; The initial length of the semi-axis is adaptively multiplicatively scaled using the dynamic compensation weights to generate a three-dimensional ellipsoidal envelope. The three-dimensional region enclosed by the three-dimensional ellipsoidal envelope is used as the internal space of the nearest neighbor search.

[0013] Preferably, calculating the sequence similarity between the time-series deformation information of the point cloud to be matched and the preset allowed deformation features of each candidate entity node includes: The displacement values ​​of the point cloud to be matched are collected in multiple consecutive sampling periods and arranged in chronological order to generate a measured deformation feature sequence. Based on the device category identifier of each candidate entity node, the corresponding preset deformation distribution logic is retrieved, and a baseline deformation feature sequence is generated according to the preset deformation distribution logic. The measured deformation feature sequence and the reference deformation feature sequence are normalized, and the cross-correlation coefficient or dynamic time-normalized distance between the measured deformation feature sequence and the reference deformation feature sequence is calculated as the sequence similarity.

[0014] Preferably, before attaching the time-series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity, the method further includes: Determine whether the similarity score corresponding to the candidate entity node with the highest sequence similarity is greater than a preset confidence threshold; If the similarity score is greater than the confidence threshold, the match is confirmed to be successful, and the time series deformation information of the point cloud to be matched is attached to the candidate entity node. If the similarity score is not greater than the confidence threshold, the geometric boundary of the dynamic error ellipsoid is enlarged by a preset expansion coefficient to expand the internal space, triggering the nearest neighbor search for secondary matching; if the highest similarity score after secondary matching is still not greater than the confidence threshold, the point cloud to be matched is determined to be a free noise point and an isolated point removal operation is performed.

[0015] Secondly, a three-dimensional settlement error correction system integrating air and ground data includes: The data acquisition module is used to acquire temporal synthetic aperture radar images of the target area and synchronized ground absolute positioning data, and generate an initial temporal radar point cloud based on the temporal synthetic aperture radar images. The error analysis module is used to extract the true three-dimensional coordinates of the ground reference points in the ground absolute positioning data, and perform spatial registration and deviation analysis with the observed three-dimensional coordinates corresponding to the initial time-series radar point cloud to construct a three-dimensional spatial error matrix. The coordinate transformation module is used to generate a global spatial transformation model based on the three-dimensional spatial error matrix, and to perform global coordinate transformation on the initial time-series radar point cloud using the global spatial transformation model, outputting an absolute coordinate point cloud containing absolute three-dimensional coordinates and time-series deformation information. The spatial retrieval module is used to construct a dynamic error ellipsoid for the point cloud to be matched in the absolute coordinate point cloud set, with the absolute three-dimensional coordinates of the point cloud to be matched as the origin and combined with the covariance data of the three-dimensional spatial error matrix; and to perform nearest neighbor search on the internal space of the dynamic error ellipsoid in the pre-imported three-dimensional digital model to extract candidate entity nodes. The attribute mounting module is used to calculate the sequence similarity between the time series deformation information of the point cloud to be matched and the preset allowable deformation features of each candidate entity node, and to mount the time series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity, so as to obtain the settlement error correction result for the candidate entity node.

[0016] Thirdly, a terminal includes a processor and a storage medium; the storage medium is used to store instructions; the processor is used to invoke and execute the instructions to implement the steps of the method as described in the first aspect.

[0017] Fourthly, a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method as described in any of the first aspects.

[0018] The beneficial effects of this invention are compared with those of the prior art: (1) This invention introduces an adaptively calibrated ground corner reflector as an absolute coordinate anchor point, orthogonally projects the spatial position residual between the radar and the ground, and decouples three independent orthogonal error components in the microwave line-of-sight plane and out of the vertical plane, namely the range error component, the azimuth error component, and the altitude error component. This mechanism overcomes the geometric projection distortion caused by the line-of-sight imaging of synthetic aperture radar from a physical source, and realizes accurate benchmark alignment between large-scale space-based microwave data and microscopic high-precision ground-based positioning data in a three-dimensional coordinate system.

[0019] (2) This invention constructs an anisotropic spatial retrieval ellipsoid by combining the accuracy differences of radar in different observation directions. At the same time, the data coherence coefficient is introduced as a dynamic compensation weight to achieve adaptive scaling of the retrieval envelope. This model overcomes the mismatch ambiguity caused by blind search of the traditional standard sphere, and while ensuring that targets in areas with poor signal quality are not missed, it effectively filters out the interference of irrelevant background nodes, and significantly reduces the computational cost of the underlying spatial optimization.

[0020] (3) When faced with overlapping devices in the same retrieval domain, this invention utilizes the heterogeneous deformation response law of solid materials to extract physical prior waveforms, and evaluates the similarity between measured deformation and prior waveforms through a dynamic time warping algorithm. This mechanism transforms the difficult-to-define overlapping three-dimensional spatial positions into highly recognizable time-series waveform differences, effectively solving the problem of confusion in point cloud attribution among dense devices, and realizing accurate cross-modal mapping of radar-observed deformation to specific three-dimensional digital components.

[0021] (4) To address the multipath effects and secondary scattering artifacts prevalent in complex electromagnetic environments, this invention designs a confidence-based authentication and adaptive secondary expansion logic for the retrieval range. When the system determines that the temporal characteristics of the data deviate from the set physical laws of the entity, it actively treats them as ionized noise and cleans them up. This mechanism effectively filters out false target displacements caused by environmental interference, avoids the contamination of the substation digital entity model by invalid distorted data, and improves the overall engineering usability of the monitoring and early warning system. Attached Figure Description

[0022] Figure 1 This is an overall flowchart of a three-dimensional settlement error correction method that integrates air and ground data provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0024] Example 1: like Figure 1 As shown, the present invention provides a three-dimensional settlement error correction method that integrates air and ground data, comprising: Step 1: Collect time-series synthetic aperture radar (SAR) images of the target area and synchronized ground absolute positioning data, and generate an initial time-series radar point cloud based on the SAR images.

[0025] The ground reference point is an artificial corner reflector integrated with satellite positioning equipment; before acquiring temporal synthetic aperture radar imagery of the target area and synchronized ground absolute positioning data, the method also includes an adaptive calibration and deployment step for the artificial corner reflector: Collect the radar wavelength, orbital inclination, and local latitude of the target area from the target radar satellite; Combining the preset radar backscattering intensity threshold with the radar wavelength, the optimal edge length of the artificial corner reflector is calculated based on the maximum radar cross section model. Based on the cosine theorem of a spherical right triangle, the true north azimuth angle of the artificial corner reflector is calculated using the orbital inclination angle and the local latitude, and the elevation angle is determined based on the geometric angle between the incident wave of the target radar satellite and the bottom surface of the artificial corner reflector. The artificial corner reflector is deployed according to the calculated optimal edge length, true north azimuth angle, and elevation angle, so as to synchronously collect image features and ground absolute positioning data through the artificial corner reflector and integrated satellite positioning equipment.

[0026] Specifically, the adaptive calibration and deployment process for the artificial corner reflector is as follows: (1) Parameter acquisition and optimal edge length calculation Obtain the orbit and physical parameters of the target radar satellite, for example, using the C-band Sentinel-1A satellite, whose radar wavelength... Approximately 0.055m, inclination of the sun-synchronous orbit. The latitude is approximately 98.18°, and the local latitude of the target substation area is extracted. .

[0027] In substations, where there are numerous metal structures and complex electromagnetic scattering backgrounds, to ensure that the artificial corner reflector appears as a coherent point with a high signal-to-noise ratio in radar images, a preset radar backscattering intensity threshold, i.e., the target RCS value, is used. The physical model is derived by working backwards from the maximum radar cross section model. For a standard triangular trihedral reflector, the derived physical model is as follows:

[0028] in, This refers to the side length of the inner edge of the artificial corner reflector. This is achieved by setting a setting that satisfies the background suppression contrast requirements. The optimal edge length can be adaptively solved by finding the value. This effectively avoids the problem of blindly selecting the size, which could lead to the corner reflector being too large and deformed by wind loads, or too small and causing the signal to be drowned out by the background noise of the substation.

[0029] (2) Spatial attitude calculation To ensure that the maximum scattering direction of the artificial corner reflector is precisely aligned with the satellite's microwave line of sight, the orbital inclination angle is used, based on the cosine theorem of a spherical right triangle. with local latitude The azimuth angle of the satellite's flight direction is calculated, and then the true north azimuth angle of the radar side-looking observation is derived. Subsequently, the nominal incident angle of the target radar satellite at that latitude is extracted, such as approximately 30°~45° for Sentinel-1. Combined with the geometric angle between the bottom surface of the artificial corner reflector and the ground plane, the optimal elevation angle is calculated to make the corner reflector's axis of symmetry parallel to the radar incident beam. .

[0030] (3) Integrated deployment and data synchronization The optimal edge length calculated above Custom-made artificial corner reflectors, following the stated true north azimuth. and elevation angle Attitude calibration is performed. The device is rigidly fixed to a stable base in the target area, such as the concrete base of a substation, and a high-precision satellite positioning device, such as a GNSS antenna, is integrated at the phase center of the corner reflector or the vertex of a known offset position. Through this deployment, image features and absolute three-dimensional ground coordinates are simultaneously acquired within the same epoch of satellite overpass imaging, achieving source unification of air-to-ground data on a spatiotemporal reference.

[0031] Next, an initial temporal radar point cloud is generated based on the temporal synthetic aperture radar image, including: Image registration is performed on the temporal synthetic aperture radar image of the target area, and interferometric pairs are combined according to a preset spatiotemporal baseline threshold to generate an interferometric atlas covering the target area; The interferogram is subjected to differential processing, phase decoherence suppression, and temporal phase unwrapping in sequence to extract the phase evolution sequence of each pixel at multiple sampling times; By combining the external terrain parameters of the target area, the phase evolution sequence is converted into line-of-sight deformation values, and the line-of-sight deformation values ​​are mapped to a three-dimensional coordinate space according to the radar satellite orbit parameters of the target area, thus obtaining the initial time-series radar point set containing the observed three-dimensional coordinates and time-series deformation information.

[0032] Specifically, the process of generating an initial point cloud based on radar imagery is as follows: (1) Interference Pair Combination and Atlas Generation First, among the acquired multi-scene temporal synthetic aperture radar images, based on the temporal and spatial baseline distribution and the difference in Doppler centroid frequency, the image with the best comprehensive correlation is selected as the main image, and the remaining images are used as auxiliary images.

[0033] Subsequently, all auxiliary images are registered with the main image at the subpixel level. To minimize noise caused by spatiotemporal decorrelation, preset spatiotemporal baseline thresholds are set, for example, controlling the spatial vertical baseline to be less than 150 meters and the temporal baseline to be less than 45 days. Only image pairs that meet these threshold conditions are combined by conjugate multiplication to generate a high-quality interferometric atlas covering the target substation area.

[0034] (2) Phase processing and unwrapping extraction Since the phase of the original interferogram includes flat terrain phase, topographic phase, and noise phase, it needs to be differentially processed and filtered.

[0035] In this embodiment, an external digital elevation model (DEM) is introduced as an external terrain parameter to simulate and remove the terrain phase caused by flat terrain effect and terrain undulation. Subsequently, an adaptive filtering algorithm, such as Goldstein filtering, is used to decoherently suppress the interference phase, smooth random noise, and enhance the clarity of the interference fringes.

[0036] Because the phase in the interferogram is enclosed in... Within the principal value range, this step employs a three-dimensional temporal phase unwrapping algorithm, such as the minimum cost flow (MCF) algorithm, to perform phase unwrapping in both spatial and temporal dimensions, restoring the wrapped phase to the true absolute cumulative phase, thereby extracting the phase evolution sequence of each pixel at multiple consecutive sampling times.

[0037] (3) Deformation solution and three-dimensional space mapping After obtaining the unwrapped phase evolution sequence, atmospheric delay error and residual orbital error are corrected by incorporating external topographic parameters of the target area. Subsequently, based on the physical correspondence between microwave wavelength and phase, the phase evolution sequence is converted into millimeter-level microwave line-of-sight deformation values.

[0038] Finally, the three-dimensional spatial projection stage begins. Geocoding is performed strictly based on the target radar satellite's orbital state vector (containing the satellite's three-dimensional position and velocity parameters) at each imaging moment and the radar line-of-sight geometry equation (i.e., the range-Doppler model). The pixels originally located in the two-dimensional radar coordinate system (range-azimuth direction) and their corresponding line-of-sight deformation values ​​are inversely projected and mapped to the three-dimensional geographic coordinate space, ultimately obtaining the initial temporal radar point cloud. Each discrete point in this point cloud carries uncorrected observation three-dimensional coordinates. And rich time-series deformation information. Among them, The longitude direction position estimated by radar. The latitude position estimated by radar. The altitude is calculated by the radar using interferometric phase.

[0039] Step 2: Extract the true three-dimensional coordinates of the ground reference points in the ground absolute positioning data, and perform spatial registration and deviation analysis with the observed three-dimensional coordinates corresponding to the initial time-series radar point cloud set to construct a three-dimensional spatial error matrix.

[0040] The true three-dimensional coordinates of the ground reference points in the ground absolute positioning data are extracted, and spatial registration and deviation analysis are performed with the observed three-dimensional coordinates corresponding to the initial temporal radar point cloud set to construct a three-dimensional spatial error matrix, including: Separate the positioning two-dimensional coordinates and absolute elevation of the ground reference point from the actual three-dimensional coordinates; Extract the two-dimensional radar coordinates and interferometric radar height of the ground reference point in the radar image system from the observed three-dimensional coordinates; The difference between the positioning two-dimensional coordinates and the radar two-dimensional coordinates is calculated to obtain the range error component and the azimuth error component in the microwave line-of-sight plane. The difference between the absolute elevation and the interferometric radar height is calculated to obtain the height error component perpendicular to the microwave line-of-sight plane. The three-dimensional spatial error matrix is ​​constructed using the calculated range error component, azimuth error component, and altitude error component as orthogonal references.

[0041] Specifically, this step aims to decouple the various systematic biases of the radar system in three-dimensional space by comparing the true values ​​with the observed values. The specific process is as follows: (1) Construction of the true value benchmark Obtain the absolute coordinates output by the GNSS device integrated on the artificial corner reflector. Its planar position information As a two-dimensional coordinate system for positioning, geodetic elevation information This serves as the absolute elevation, which is used as the true value benchmark for spatial correction. The longitude position of the ground reference point in the real geodetic coordinate system. This indicates the position in the latitude direction.

[0042] (2) Feature extraction of observations From the initial temporal radar point cloud set, radar point cloud features corresponding to the physical location of the artificial corner reflector are identified based on the spatial proximity matching principle. The observation position of this point under radar geometric mapping is then extracted. As a two-dimensional radar coordinate system, the measured interference height is extracted. ,by As a system observation with errors.

[0043] (3) Spatial Dimension Error Decoupling Analysis The true values ​​are compared with the observed values ​​using a three-dimensional spatial decomposition. On one hand, the difference between the two-dimensional coordinates in the plane is calculated to obtain the planar position residual. Since radar imagery is based on satellite flight trajectory (azimuth) and microwave transmission direction (range), the plane position residual needs to be orthogonally projected onto the radar's microwave line-of-sight plane to decompose and obtain the range error component within the microwave line-of-sight plane. and azimuth error components .

[0044] On the other hand, the difference operation is performed on the height dimension, i.e., calculation. The height error component perpendicular to the microwave line-of-sight plane was calculated. .

[0045] (4) Assembly of three-dimensional spatial error matrix The calculated error components in the three orthogonal directions are assembled to construct a three-dimensional spatial error matrix. In this embodiment, the error feature vector is used as the basis for the analysis. The transformation matrix is ​​constructed based on this. The range error component is described below. The line-of-sight deviation caused by atmospheric delay and radar ranging system was quantified; the azimuth error component The azimuth deviation caused by satellite orbital clock drift or Doppler shift was quantified; the altitude error component The vertical mapping bias introduced by terrain residuals was quantified. This three-dimensional spatial error matrix... The underlying offset mechanisms of the radar system in three-dimensional spatial mapping were precisely quantified, providing core prior error parameters for the subsequent generation of a global spatial transformation model.

[0046] Step 3: Generate a global spatial transformation model based on the three-dimensional spatial error matrix, and use the global spatial transformation model to perform global coordinate transformation on the initial time-series radar point cloud, outputting an absolute coordinate point cloud containing absolute three-dimensional coordinates and time-series deformation information.

[0047] Specifically, this step aims to eliminate spatial distortion across the entire radar point cloud system and achieve global uniformity of data benchmarks. The specific implementation process is as follows: (1) Construction of global space transformation model Extract the three-dimensional spatial error matrix constructed in step 2 Based on this error matrix, a three-dimensional affine transformation equation is constructed to generate a global spatial transformation model. In practical engineering implementation, the range, azimuth, and altitude error components in the error matrix are projected into three-dimensional translation correction vectors in a standard geographic coordinate system (such as WGS84 or CGCS2000) using a Jacobian matrix projection. :

[0048] in, This is the absolute translation correction amount in the longitude direction. This represents the absolute translation correction in the latitudinal direction. This represents the absolute translation correction in the elevation direction. Based on this three-dimensional translation correction vector... The global space transformation model is thus constructed.

[0049] (2) Batch conversion of global coordinates Using the global spatial transformation model, a batch traversal calculation is performed on the initial temporal radar point cloud set generated in step 1. The original observation three-dimensional coordinates of each discrete point cloud are then calculated. As the input matrix, perform spatial transformation operations:

[0050] in, ; This is the absolute position vector output by the global spatial transformation model. Through the aforementioned global coordinate transformation operation, the nonlinear geometric distortion caused by the line-of-sight projection is effectively eliminated, and the absolute three-dimensional coordinates of each point cloud are calculated. ,Right now .

[0051] (3) Output of absolute coordinate point set While performing coordinate system upsizing and correction on the spatial location of point clouds, the process fully preserves and inherits the millimeter-level time-series deformation information (i.e., phase evolution characteristics) originally carried by each point cloud. The final output set of absolute coordinate point clouds possesses both absolute spatial location with the same reference as ground GNSS equipment and rich deformation time-series attributes. This breaks down the barrier that traditional microwave remote sensing data is difficult to directly integrate with substation digital entity models (such as BIM models), providing a reliable underlying coordinate system for subsequent accurate 3D spatial retrieval.

[0052] Step 4: For the point cloud to be matched in the absolute coordinate point cloud set, construct a dynamic error ellipsoid with the absolute three-dimensional coordinates of the point cloud to be matched as the origin and the covariance data of the three-dimensional spatial error matrix; in the pre-imported three-dimensional digital model, perform nearest neighbor search on the internal space of the dynamic error ellipsoid to extract candidate entity nodes.

[0053] Constructing a dynamic error ellipsoid using the covariance data of the aforementioned three-dimensional spatial error matrix includes: Calculate the covariance data of the range error component, the azimuth error component, and the altitude error component in the three-dimensional spatial error matrix; Using the absolute three-dimensional coordinates of the point cloud to be matched as the geometric center, a local spatial retrieval coordinate system is established, which includes the microwave line-of-sight axis, the azimuth trajectory axis, and the vertical elevation axis. The covariance data of the range error component, the covariance data of the azimuth error component, and the covariance data of the altitude error component are mapped to the microwave line-of-sight axis, the azimuth trajectory axis, and the vertical elevation axis, respectively, and the initial length of the semi-axis of the dynamic error ellipsoid is calculated. Collect the coherence coefficients of the current time-series synthetic aperture radar image, and calculate the dynamic compensation weights based on the coherence coefficients; The initial length of the semi-axis is adaptively multiplicatively scaled using the dynamic compensation weights to generate a three-dimensional ellipsoidal envelope. The three-dimensional region enclosed by the three-dimensional ellipsoidal envelope is used as the internal space of the nearest neighbor search.

[0054] Specifically, although a set of absolute coordinate points with a unified spatial reference has been obtained after the global coordinate transformation in step 3, in actual engineering scenarios such as substations, physical equipment such as transformer bodies, bushing insulators, and overhead high-voltage busbars usually present a highly dense, three-dimensionally interwoven state. If rigid point-to-point distance matching is performed in the three-dimensional digital model based solely on the single absolute three-dimensional coordinates of the point cloud, it is highly susceptible to errors due to minor observational disturbances remaining in the radar system. This can easily lead to incorrect association of settlement data with adjacent equipment that are spatially very close but have not actually undergone deformation, resulting in serious data ambiguity.

[0055] To overcome this technical bottleneck, this step introduces the statistical characteristics of the underlying error into the spatial matching stage, designing a dynamic spatial retrieval mechanism based on covariance and coherence. This mechanism aims to scientifically define the potential physical attribution search boundary for each radar point cloud. The specific implementation process is as follows: (1) Covariance analysis and initial length calculation of semi-axis The three-dimensional spatial error matrix obtained in step 2 Statistical analysis was performed to extract the covariance data of the error components within a specific spatiotemporal window. Due to the inherent limitations of radar side-looking imaging mechanisms, its ranging accuracy in different directions exhibits significant asymmetry; for example, the azimuth resolution is usually lower than that in the range direction.

[0056] For any point cloud to be matched in the absolute coordinate point cloud set, its absolute three-dimensional coordinates after correction in step 3 are used. Establish a local spatial retrieval coordinate system that conforms to the physical characteristics of the radar, with the geometric center as the reference point. Integrate the range error components... Azimuth error components and height error components These are respectively mapped onto the microwave line-of-sight axis, azimuth trajectory axis, and vertical elevation axis. The direction with the larger error variance is used to calculate the initial length of the semi-axis of the dynamic error ellipsoid. .in, This is the length of the error half-axis corresponding to the microwave line-of-sight axis (range direction). This corresponds to the error half-axis length of the azimuth trajectory axis. The initial length of the error semi-axis corresponds to the vertical elevation axis. This initial length is obtained by multiplying the standard deviation of each component in the three-dimensional spatial error matrix by a preset confidence coefficient. The larger the error variance, the longer the corresponding initial length of the semi-axis. This allows for the construction of an anisotropic spatial asymmetric ellipsoid based on the covariance principle, overcoming the limitations of blind searching using a standard sphere.

[0057] (2) Adaptive multiplicative scaling and dynamic ellipsoid construction Extract the coherence coefficient of the point cloud to be matched in the current radar image. The value typically ranges from 0 to 1, representing the stability of the signal. The dynamic compensation weights are calculated based on the following formula. :

[0058] in, This is the preset scaling factor.

[0059] In this embodiment, the dynamic compensation weight is utilized. Initial length of half axis Adaptive breathing-style scaling is performed to calculate the actual dynamic half-axis length. Its computational logic is as follows:

[0060] in, Let be the specific component of the vector in the i-th direction, where when i = r, a, h, it represents the dynamic range half-axis length, dynamic azimuth half-axis length, and dynamic altitude half-axis length, respectively. When the coherence coefficient... At extremely high levels, dynamic compensation weights The ellipsoid shrinks as coherence decreases, resulting in rigorous and precise spatial matching; dynamic weight compensation is applied when coherence is reduced by environmental disturbances. Increase the size of the ellipsoid, moderately expand it, and widen the search boundary to prevent real targets from being missed. Based on Generate the final three-dimensional ellipsoidal envelope.

[0061] (3) Extraction of candidate nodes inside the three-dimensional digital model A pre-imported 3D digital model of the substation, such as a BIM solid model, is used as the base map reference. A 3D spatial indexing algorithm, such as an octree or KD-Tree algorithm, is used to perform nearest neighbor retrieval within the envelope of the dynamic ellipsoid.

[0062] All BIM model entity nodes falling within this space, such as the closely spaced transformer casing and the insulator above it, are extracted and used as candidate entity nodes for subsequent temporal feature verification. This dynamic retrieval mechanism based on covariance and coherence ensures that real nodes are not missed while significantly filtering out irrelevant distant background interference, greatly reducing the computational cost of subsequent matching.

[0063] Step 5: Calculate the sequence similarity between the time series deformation information of the point cloud to be matched and the preset allowable deformation features of each candidate entity node, and attach the time series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity to obtain the settlement error correction result for the candidate entity node.

[0064] Calculating the sequence similarity between the time-series deformation information of the point cloud to be matched and the preset allowed deformation features of each candidate entity node includes: The displacement values ​​of the point cloud to be matched are collected in multiple consecutive sampling periods and arranged in chronological order to generate a measured deformation feature sequence. Based on the device category identifier of each candidate entity node, the corresponding preset deformation distribution logic is retrieved, and a baseline deformation feature sequence is generated according to the preset deformation distribution logic. The measured deformation feature sequence and the reference deformation feature sequence are normalized, and the cross-correlation coefficient or dynamic time-normalized distance between the measured deformation feature sequence and the reference deformation feature sequence is calculated as the sequence similarity.

[0065] Specifically, this step introduces a spatiotemporal multidimensional verification mechanism to address the challenge of determining the unique attribution of a point cloud solely based on spatial geometric distance when multiple overlapping devices exist within the error ellipsoid. The specific implementation process is as follows: (1) Construction of measured deformation feature sequence Extract historical deformation data of the point cloud to be matched within the radar monitoring period (e.g., 24 images from the past year). Concatenate the data in chronological order according to timestamps to construct a one-dimensional sequence of measured deformation features. :

[0066] in, The number of sampling periods for the radar image. Indicates at the sampling time The specific displacement value obtained.

[0067] (2) Generation of prior benchmark deformation feature sequence For the multiple candidate entity nodes extracted in step 4, their equipment category identifiers (ID and attribute labels) attached to the 3D digital model, such as the BIM model, will be read. For equipment with different physical attributes, a pre-established equipment prior deformation model will be invoked.

[0068] For example, if the category is concrete foundation, its preset deformation distribution logic is usually a monotonically gradually changing linear or logarithmic settlement curve; if the category is flexible overhead cables or metal structures, its preset deformation distribution logic is a periodic sinusoidal fluctuation curve strongly correlated with seasonal temperature. Based on this, a baseline deformation characteristic sequence corresponding to the observation period is generated. .

[0069] (3) Temporal similarity calculation and data mounting Since radar observations represent absolute displacement, while preset benchmarks may only reflect trends, Z-score normalization or Min-Max normalization algorithms are needed to adjust the values. and To eliminate the interference of amplitude differences on shape matching, the units are standardized. After normalization, the similarity between the two sequences is calculated: under ideal synchronization, the cross-correlation coefficient can be used to assess linear correlation; considering that deformation response may have time lag (phase shift) in real-world scenarios, this embodiment preferably uses the Dynamic Time Warping (DTW) algorithm. The DTW algorithm can find the optimal alignment path between two time series by elastically stretching or compressing the time axis, thereby accurately quantifying their inherent waveform similarity distance.

[0070] Finally, the similarity scores of all candidate entity nodes are compared. The system determines which device's prior rules best match the actual physical behavior of the point cloud, thus confirming its true affiliation. The millimeter-level deformation data of this point cloud is then formally mounted and written into the data interface of the candidate entity node with the highest sequence similarity (i.e., the shortest DTW distance or the largest correlation coefficient), completing the cross-modal accurate mapping from radar line-of-sight coordinates to the 3D digital entity, and outputting the final settlement error correction result.

[0071] Before attaching the time-series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity, the method further includes: Determine whether the similarity score corresponding to the candidate entity node with the highest sequence similarity is greater than a preset confidence threshold; If the similarity score is greater than the confidence threshold, the match is confirmed to be successful, and the time series deformation information of the point cloud to be matched is attached to the candidate entity node. If the similarity score is not greater than the confidence threshold, the geometric boundary of the dynamic error ellipsoid is enlarged by a preset expansion coefficient to expand the internal space, triggering the nearest neighbor search for secondary matching; if the highest similarity score after secondary matching is still not greater than the confidence threshold, the point cloud to be matched is determined to be a free noise point and an isolated point removal operation is performed.

[0072] Specifically, this step is a closed-loop fault-tolerant mechanism to enhance the robustness of the settlement error correction system, aiming to avoid incorrect mounting and filter out radar false targets caused by multipath effects or electromagnetic interference. The specific implementation process is as follows: (1) Confidence authentication and formal mounting After obtaining the highest similarity score in step 5, extract the highest similarity score from the results. For example, the minimum DTW distance or the maximum cross-correlation coefficient. Instead of directly forcing the mounting, a confidence-based authentication is introduced to determine the highest similarity score. Does it meet the system's preset strict confidence threshold? .like This indicates that the temporal fluctuation behavior of the point cloud to be matched is highly consistent with the prior physical laws of the selected entity, confirming a successful match. The system then performs the formal data mounting and attribute overwriting operations, writing the settlement time series data into the substation's three-dimensional digital model.

[0073] (2) Search space expansion and quadratic matching like This indicates that the initial extracted local spatial retrieval coordinate system does not contain the true physical affiliation of the point cloud. This is usually due to the initial evaluation's error covariance being too conservative. At this point, the system triggers a self-healing search mechanism, adjusting the search according to a preset dilation coefficient. For example, setting the coefficient to 1.5, the lengths of the three semi-axis of the original dynamic error ellipsoid... The length of the expanded semi-axis is calculated by performing proportional or adaptive scaling. ,Right now:

[0074] by Expand the internal retrieval space to generate a new three-dimensional envelope surface, and re-execute nearest neighbor search and temporal similarity calculation within this range to complete the secondary matching.

[0075] (3) Removal of free noise points If, after expanding the search scope a second time, the highest similarity score still fails to surpass the confidence threshold... If the radar point cloud does not belong to any valid monitoring entity within the substation, it is highly likely to be a false target caused by secondary scattering from temporarily parked maintenance vehicles or metal scaffolding, or severe atmospheric phase noise.

[0076] For this type of data, the system marks it as a free noise point and directly performs an isolated point removal operation on the absolute coordinate point cloud. By setting up a complete closed loop of confidence interception, range expansion, secondary retry, and noise cleaning, the system reduces the contamination of the substation's 3D digital model and subsequent construction early warning system by invalid data while ensuring matching accuracy.

[0077] Example 2: A three-dimensional settlement error correction system integrating air and ground data includes: The data acquisition module is used to acquire temporal synthetic aperture radar images of the target area and synchronized ground absolute positioning data, and generate an initial temporal radar point cloud based on the temporal synthetic aperture radar images. The error analysis module is used to extract the true three-dimensional coordinates of the ground reference points in the ground absolute positioning data, and perform spatial registration and deviation analysis with the observed three-dimensional coordinates corresponding to the initial time-series radar point cloud to construct a three-dimensional spatial error matrix. The coordinate transformation module is used to generate a global spatial transformation model based on the three-dimensional spatial error matrix, and to perform global coordinate transformation on the initial time-series radar point cloud using the global spatial transformation model, outputting an absolute coordinate point cloud containing absolute three-dimensional coordinates and time-series deformation information. The spatial retrieval module is used to construct a dynamic error ellipsoid for the point cloud to be matched in the absolute coordinate point cloud set, with the absolute three-dimensional coordinates of the point cloud to be matched as the origin and combined with the covariance data of the three-dimensional spatial error matrix; and to perform nearest neighbor search on the internal space of the dynamic error ellipsoid in the pre-imported three-dimensional digital model to extract candidate entity nodes. The attribute mounting module is used to calculate the sequence similarity between the time series deformation information of the point cloud to be matched and the preset allowable deformation features of each candidate entity node, and to mount the time series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity, so as to obtain the settlement error correction result for the candidate entity node.

[0078] Example 3: A terminal includes a processor and a storage medium; the storage medium is used to store instructions; the processor is used to invoke and execute the instructions to implement the steps of the method described in Embodiment 1.

[0079] Example 4: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described in any one of Examples 1.

[0080] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0081] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0082] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0083] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A three-dimensional settlement error correction method integrating air and ground data, characterized in that, include: The temporal synthetic aperture radar imagery of the target area and the synchronized ground absolute positioning data are collected, and an initial temporal radar point cloud is generated based on the temporal synthetic aperture radar imagery. The true three-dimensional coordinates of the ground reference point in the absolute ground positioning data are extracted, and spatial registration and deviation analysis are performed with the corresponding observed three-dimensional coordinates in the initial time-series radar point cloud to construct a three-dimensional spatial error matrix. This includes: separating the positioning two-dimensional coordinates and absolute elevation of the ground reference point from the true three-dimensional coordinates; extracting the radar two-dimensional coordinates and interferometric radar height of the ground reference point in the radar image system from the observed three-dimensional coordinates; performing a subtraction operation between the positioning two-dimensional coordinates and the radar two-dimensional coordinates to calculate the range error component and azimuth error component in the microwave line-of-sight plane; performing a subtraction operation between the absolute elevation and the interferometric radar height to calculate the height error component perpendicular to the microwave line-of-sight plane; and constructing the three-dimensional spatial error matrix using the calculated range error component, azimuth error component, and height error component as orthogonal references. A global spatial transformation model is generated based on the three-dimensional spatial error matrix. The global spatial transformation model is used to perform global coordinate transformation on the initial temporal radar point set, and the output is an absolute coordinate point set containing absolute three-dimensional coordinates and time series deformation information. For the point cloud to be matched in the absolute coordinate point cloud set, a dynamic error ellipsoid is constructed using the absolute three-dimensional coordinates of the point cloud to be matched as the origin and the covariance data of the three-dimensional spatial error matrix. This includes: calculating the covariance data of the range error component, the azimuth error component, and the altitude error component in the three-dimensional spatial error matrix; establishing a local spatial retrieval coordinate system containing the microwave line-of-sight axis, the azimuth trajectory axis, and the vertical elevation axis, using the absolute three-dimensional coordinates of the point cloud to be matched as the geometric center; and combining the covariance data of the range error component, the azimuth error component, and the altitude error component. The initial length of the semi-axis of the dynamic error ellipsoid is calculated by mapping the images to the microwave line-of-sight axis, the azimuth trajectory axis, and the vertical elevation axis, respectively. The coherence coefficients of the current time-series synthetic aperture radar image are collected, and dynamic compensation weights are calculated based on these coefficients. The initial length of the semi-axis is adaptively scaled using these dynamic compensation weights, which increase as the coherence coefficients decrease, generating a three-dimensional ellipsoidal envelope. The three-dimensional region enclosed by this envelope is used as the internal space for nearest neighbor search. In the pre-imported three-dimensional digital model, nearest neighbor search is performed on the internal space of the dynamic error ellipsoid to extract candidate entity nodes. Calculating the sequence similarity between the time-series deformation information of the point cloud to be matched and the preset allowable deformation features of each candidate entity node includes: retrieving the corresponding preset deformation distribution logic based on the device category identifier of each candidate entity node, generating a benchmark deformation feature sequence according to the preset deformation distribution logic, calculating the cross-correlation coefficient or dynamic time warping distance between the measured deformation feature sequence and the benchmark deformation feature sequence; and attaching the time-series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity to obtain the settlement error correction result for the candidate entity node.

2. The three-dimensional settlement error correction method merging air-ground data according to claim 1, characterized in that, The ground reference point is an artificial corner reflector integrated with satellite positioning equipment; before acquiring temporal synthetic aperture radar imagery of the target area and synchronized ground absolute positioning data, the method also includes an adaptive calibration and deployment step for the artificial corner reflector: Collect the radar wavelength, orbital inclination, and local latitude of the target area from the target radar satellite; Combining the preset radar backscattering intensity threshold with the radar wavelength, the optimal edge length of the artificial corner reflector is calculated based on the maximum radar cross section model. Based on the cosine theorem of a spherical right triangle, the true north azimuth angle of the artificial corner reflector is calculated using the orbital inclination angle and the local latitude, and the elevation angle is determined based on the geometric angle between the incident wave of the target radar satellite and the bottom surface of the artificial corner reflector. The artificial corner reflector is deployed according to the calculated optimal edge length, true north azimuth angle, and elevation angle, so as to synchronously collect image features and ground absolute positioning data through the artificial corner reflector and integrated satellite positioning equipment.

3. The three-dimensional settlement error correction method based on fused air-ground data according to claim 1, characterized in that, Based on the aforementioned temporal synthetic aperture radar imagery, an initial temporal radar point cloud is generated, including: Image registration is performed on the temporal synthetic aperture radar image of the target area, and interferometric pairs are combined according to a preset spatiotemporal baseline threshold to generate an interferometric atlas covering the target area; The interferogram is subjected to differential processing, phase decoherence suppression, and temporal phase unwrapping in sequence to extract the phase evolution sequence of each pixel at multiple sampling times; By combining the external terrain parameters of the target area, the phase evolution sequence is converted into line-of-sight deformation values, and the line-of-sight deformation values ​​are mapped to a three-dimensional coordinate space according to the radar satellite orbit parameters of the target area, thus obtaining the initial time-series radar point set containing the observed three-dimensional coordinates and time-series deformation information.

4. The three-dimensional settlement error correction method based on fused air-ground data according to claim 1, characterized in that, Calculating the sequence similarity between the time-series deformation information of the point cloud to be matched and the preset allowed deformation features of each candidate entity node includes: The displacement values ​​of the point cloud to be matched are collected in multiple consecutive sampling periods and arranged in chronological order to generate a measured deformation feature sequence. The measured deformation feature sequence and the reference deformation feature sequence are normalized, and the cross-correlation coefficient or dynamic time-normalized distance between the measured deformation feature sequence and the reference deformation feature sequence is calculated as the sequence similarity.

5. The three-dimensional settlement error correction method based on fused air-ground data according to claim 1, characterized in that, Before attaching the time-series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity, the method further includes: Determine whether the similarity score corresponding to the candidate entity node with the highest sequence similarity is greater than a preset confidence threshold; If the similarity score is greater than the confidence threshold, the match is confirmed to be successful, and the time series deformation information of the point cloud to be matched is attached to the candidate entity node. If the similarity score is not greater than the confidence threshold, the geometric boundary of the dynamic error ellipsoid is enlarged by a preset expansion coefficient to expand the internal space, triggering the nearest neighbor search for secondary matching; if the highest similarity score after secondary matching is still not greater than the confidence threshold, the point cloud to be matched is determined to be a free noise point and an isolated point removal operation is performed.

6. A three-dimensional settlement error correction system integrating air and ground data, running a three-dimensional settlement error correction method integrating air and ground data as described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to acquire time-series synthetic aperture radar images of the target area and synchronized ground absolute positioning data, and generate an initial time-series radar point cloud based on the time-series synthetic aperture radar images. The error analysis module is used to extract the true three-dimensional coordinates of the ground reference points in the ground absolute positioning data, and perform spatial registration and deviation analysis with the observed three-dimensional coordinates corresponding to the initial time-series radar point cloud to construct a three-dimensional spatial error matrix. The coordinate transformation module is used to generate a global spatial transformation model based on the three-dimensional spatial error matrix, and to perform global coordinate transformation on the initial time-series radar point cloud using the global spatial transformation model, outputting an absolute coordinate point cloud containing absolute three-dimensional coordinates and time-series deformation information. The spatial retrieval module is used to construct a dynamic error ellipsoid for the point cloud to be matched in the absolute coordinate point cloud set, with the absolute three-dimensional coordinates of the point cloud to be matched as the origin, and in combination with the covariance data of the three-dimensional spatial error matrix. In the pre-imported 3D digital model, a nearest neighbor search is performed on the interior space of the dynamic error ellipsoid to extract candidate entity nodes; The attribute mounting module is used to calculate the sequence similarity between the time series deformation information of the point cloud to be matched and the preset allowable deformation features of each candidate entity node, and to mount the time series deformation information of the point cloud to be matched to the candidate entity node with the highest sequence similarity, so as to obtain the settlement error correction result for the candidate entity node.

7. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; the processor is used to invoke and execute the instructions to implement the steps of the method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-5.

Images