A rockfill dam deformation monitoring fusion method based on EPS-InSAR and particle filtering

Abstract

The application discloses a kind of based on EPS-InSAR and particle filtering's rockfill dam deformation monitoring fusion method, it is related to hydraulic structure monitoring technical field, the present application is aimed at traditional point monitoring space coverage insufficient, conventional InSAR measuring point sparse and conventional assimilation algorithm poor adaptability etc., in combination with enhanced time series InSAR technology and particle filtering algorithm realizes multi-source data fusion.The present application includes SAR image differential interference pretreatment, PS and DS target point selection, spatial dimension and resolution assimilation mapping and particle filtering-based multi-source data assimilation fusion etc..Through the method, construct multi-source data collaborative deformation perception process, improve dam surface measuring point density, greatly reduce the root mean square error of fusion data, overcome the defects of traditional algorithm, provide more scientific, reliable and applicable new method for rockfill dam deformation monitoring, can be used in pumped storage power station and complex water conservancy engineering scene.

Description

Technical Field

[0001] This invention relates to the field of hydraulic structure monitoring technology, specifically a fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering. Background Technology

[0002] Rockfill dams, with their advantages of strong adaptability, small engineering scale, and low cost, are widely used not only in conventional hydropower but have also become the mainstream dam type for newly built pumped storage power stations. However, with the continuous emergence of high rockfill dams exceeding 200m or even 300m in height, problems such as large deformation during operation and easy damage to the seepage prevention structure have become prominent, seriously threatening the overall safety and operational efficiency of the project. Therefore, large-scale, all-weather, and high-precision monitoring of the surface deformation of rockfill dams, and accurate modeling and simulation of their overall spatial deformation distribution characteristics, are of great significance.

[0003] Currently, surface deformation monitoring of rockfill dams mainly employs conventional techniques such as levels, total stations, tension lines, and global navigation satellite systems. While these traditional monitoring methods can provide high-precision deformation monitoring data for key parts of the dam, their discrete "point-to-surface" monitoring model has inherent limitations when dealing with large rockfill dams hundreds of meters long and hundreds of meters high. It struggles to accurately and comprehensively depict the overall spatial deformation distribution characteristics and cannot achieve rapid identification and thorough perception of early abnormal deformation areas. Furthermore, while Synthetic Aperture Radar Interferometry (InSAR) technology is increasingly being applied to dam deformation monitoring due to its advantages of wide coverage, high precision, and all-weather operation, many studies still rely primarily on traditional time-series analysis models such as PS-InSAR (Permanent Scatterer Interferometry) and SBAS-InSAR (Short Baseline Set Interferometry). These conventional methods perform well in urban areas with abundant strong scatterers, but in non-urban areas where most water conservancy projects are located, it is difficult to accurately identify a sufficiently dense number of observation points. As pointed out in "B. Osmanoğlu, F. Sunar, S. Wdowinski, et al., Time series analysis of InSAR data: Methods and trends, ISPRS Journal of Photogrammetry and Remote Sensing 115(2016)90-102", traditional time-series InSAR technology is limited by the surface scattering characteristics, resulting in low coherent target recognition rates in non-strong scattering areas and easily creating monitoring blind zones, thus limiting its application in large-scale deformation sensing of large-scale water conservancy projects. Furthermore, to address the issue of low single-point accuracy caused by atmospheric and spatiotemporal decorrelation in InSAR technology, some current studies employ data assimilation methods such as ensemble Kalman filtering (EnKF) to fuse large-scale InSAR observations with high-precision conventional leveling monitoring data. However, the deformation state variables of actual rockfill dams are mostly nonlinear and difficult to satisfy Gaussian distribution conditions, which greatly limits the information mining capabilities of Kalman filtering algorithms. For example, the study in "Ma Jianwen, Qin Sixian. A review of the current research status of data assimilation algorithms [J]. Advances in Earth Science, 2012, 27(7): 747-757" shows that when ensemble Kalman filtering is used to deal with the data assimilation problem of nonlinear systems, the filter divergence problem is prone to occur. As the assimilation time increases, the analysis value will get closer and closer to the background field, and eventually completely reject the actual observation data, resulting in the failure of multi-source data fusion. Summary of the Invention

[0004] To address the shortcomings of existing technologies, such as insufficient spatial coverage of traditional point-based monitoring, sparse monitoring points due to the limited coherent target recognition rate of conventional InSAR methods in non-urban areas, and poor adaptability of conventional data assimilation algorithms like Kalman filtering to nonlinear and non-Gaussian distributions leading to filter divergence, this invention provides a fusion method for rockfill dam deformation monitoring based on EPS-InSAR and particle filtering. This invention combines enhanced temporal InSAR technology with particle filtering algorithms to achieve large-scale, all-weather, high-precision deformation monitoring and cross-scale fusion reconstruction of multi-source data for rockfill dams.

[0005] The present invention specifically provides the following technical solutions to achieve the above objectives: The technical solution adopted in this invention is a fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering, specifically as follows: Step 1: Differential interferometric preprocessing and phase generation of SAR images; Step 2: Select target points in PS and DS and perform joint inversion to extract line-of-sight deformation; Step 3: Spatial dimension and resolution assimilation mapping; Step 4: Multi-source data assimilation and fusion based on particle filter algorithm.

[0006] The invention is further characterized in that: Step 1 is as follows: The image range was determined based on the geographical latitude and longitude coordinates and vector range of the study area, and single-view complex (SLC) SAR image data covering the area were collected. The master image was selected according to the spatiotemporal baseline criterion, and the remaining slave images were registered with it to form a connected image pair. Precise orbit information was introduced and combined with external DEM to simulate the terrain phase and generate a reference ellipsoid phase. Differential interferograms were generated by complex conjugate multiplication to remove the flat terrain effect and initially weaken the influence of terrain on the interferometric phase.

[0007] Step 2 is as follows: Step 2.1: Extraction and optimization of PS and DS points; Step 2.2: Construction of the joint target point network and phase unwrapping; Step 2.3: Deformation phase separation and geocoding.

[0008] Step 2.1 specifically involves: To calculate the stability of pixel amplitude in time-series images, pixels that meet the threshold condition are designated as permanent scatterer (PS) points, and their amplitude deviation index is calculated using the following formula:

[0009] In the formula: [0,1] represents the amplitude deviation index. The smaller the value, the stronger the temporal stability of the pixel amplitude, which is more in line with the characteristics of PS. is the standard deviation of the amplitude of a pixel in N time-series SAR images; This represents the average temporal amplitude of the pixel.

[0010] Next, the Kolmogorov-Smirnov Test is used to identify homogeneous pixels with the same statistical properties within the spatial window as candidate points for distributed scatterers (DS), as shown in the following formula:

[0011] In the formula: The Kolmogorov-Smirnov Test statistic is used to determine whether two pixels are statistically homogeneous (SUP) when the statistic is below a certain threshold. This means that when x belongs to The maximum difference of the cumulative distribution function; , The center pixel With neighboring pixels The cumulative distribution function of the amplitude values; N is the number of time-series images. After identifying the DS target, principal component analysis (PCA) is used to adaptively filter and extract the equivalent phase of the homogeneous pixel group.

[0012] Step 2.2 specifically involves: The selected DS points and PS points are combined to construct a hybrid target point network, and differential phase decomposition calculations are performed to separate the effective deformed phase. Specifically:

[0013] In the formula: Residual phase of the terrain; This refers to the phase of surface deformation. This is the atmospheric delayed phase; It represents random noise phase.

[0014] Step 2.3 specifically involves: Phase of surface deformation Divided into linear deformation With nonlinear deformation Separation is achieved by utilizing the differences in the spatiotemporal characteristics of the atmosphere and deformation, specifically:

[0015] In the formula: The wavelength is the radar wavelength; T is the time interval. The deformation rate is determined by the line-of-sight (LOS) axis. After determining the deformation, the deformation information in the SAR coordinate system is converted to the geographic coordinate system through geocoding to obtain the temporal deformation process and the deformation value along the radar line of sight in the study area.

[0016] Step 3 specifically involves: The deformation values ​​observed by InSAR technology along the radar line of sight are converted into vertical deformation, and the calculation formula is as follows:

[0017] In the formula: Line-of-sight deformation values ​​obtained from InSAR monitoring; This is the transformed vertical deformation value; This is the radar incident angle.

[0018] Subsequently, the Kriging spatial interpolation method was used to obtain the InSAR vertical deformation data of the corresponding positions of the discrete leveling points, so as to achieve spatial resolution assimilation and matching between the InSAR observation surface and the leveling observation points. The InSAR temporal deformation data was used as the background field, and the field leveling measurement data was used as the observation field.

[0019] Step 4 specifically involves: Step 4.1: Particle initialization and state prediction; Step 4.2: Weight Update; Step 4.3: Importance resampling; Step 4.4: State estimation and output.

[0020] Step 4.1 specifically involves: Using the InSAR data at time j=0 as a baseline, random noise following a Gaussian distribution is superimposed to generate an initial particle set of N particles, with a weight of 1 / N. A data-driven state transition model is then used to predict the particle state at the next time step. Update the particle set:

[0021] In the formula: This refers to the state at the previous moment; This refers to the temporal variations observed by InSAR; The process noise of the model is assumed to follow a Gaussian distribution with a mean of zero and a variance of Q.

[0022] Step 4.2 specifically involves: When the leveling observation value is obtained at time j At that time, the particle weights are updated using the probability density of the observed data and then normalized. The specific expression is as follows:

[0023] In the formula: Let be the weight of the nth particle at time j; In the state The observations obtained below The probability density.

[0024] Step 4.3 specifically involves: Determine the number of effective particles Is it below the set diversity threshold? ,if Then, SIR resampling is performed, copying particles with larger weights while discarding particles with smaller weights to avoid particle degradation, thus increasing the effective particle count. The calculation is as follows:

[0025] Step 4.4 specifically involves: By combining the particles and their weights, the optimal estimate of the particle filter at time j is calculated. The specific expression is as follows:

[0026] The assimilation time is calculated through the above iterations until the accuracy requirements are met and the optimal assimilation estimate is output, thus obtaining multi-source deformation monitoring fusion data with complementary advantages and high spatiotemporal resolution.

[0027] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a multi-source data collaborative deformation sensing workflow of "InSAR surface observation to ground leveling assimilation," featuring all-weather, large-scale, and high-precision capabilities, applicable to pumped storage power station high-face rockfill dams and complex hydraulic engineering scenarios. This method introduces and successfully applies Enhanced Time Series InSAR (EPS-InSAR) technology to the deformation monitoring of high-face rockfill dams. Compared to traditional PS-InSAR and SBAS-InSAR methods, this invention achieves a significant and leapfrog improvement in the density of measuring points on the dam surface (up to approximately 256% to 274%, with a measuring point density reaching 21,977 points / km). 2The spatial distribution of this invention has successfully transitioned from "discrete strips" to "area-like full coverage." Furthermore, by introducing a particle filter (PF) multi-source data assimilation model, the root mean square error (RMSE) of the fused data obtained in this invention has significantly decreased (with a maximum reduction of 83.37%), effectively overcoming the shortcomings of traditional ensemble Kalman filtering in handling nonlinear deformation and non-Gaussian distributions of rockfill dams, which tend to exhibit filter divergence and exclusion of actual observation data. This invention's method can accurately capture local settlement centers and anomalous fluctuations based on macroscopic continuous coverage, quantitatively revealing the corrective effect of multi-source data assimilation on reducing atmospheric and spatiotemporal uncorrelated noise. From the monitoring mechanism level, it establishes a complementary chain of "space-based large-scale area survey - high-precision calibration of ground discrete points," providing reliable data and theoretical support for explaining and controlling the complex deformation evolution of high rockfill dams.

[0028] 2. This invention addresses the shortcomings of conventional point-based monitoring, which often suffers from insufficient spatial coverage due to "point-to-area" approaches, and conventional time-series InSAR methods, which are limited by the coherent target recognition rate in non-urban water conservancy areas, resulting in sparse measurement points and easy monitoring blind spots. By jointly processing PS points and homogeneous DS points, this invention maximizes the extraction of effective information from SAR images and introduces a particle filter data assimilation system with an importance resampling (SIR) mechanism. This enables a more realistic and comprehensive perception of the overall spatial deformation characteristics, local settlement differences, and long-term evolution of rockfill dams. This invention can continuously characterize the overall spatial distribution and gradient features of dam deformation on a wide-area scale, and can also use high-precision leveling data for assimilation correction at a single-point scale to suppress local noise. It profoundly demonstrates the corrective effect of multi-source data fusion on the reconstruction of the deformation field on the surface of rockfill dams under spatial variability. This provides a more scientific, reliable, and applicable new deformation perception and data fusion method for deformation behavior analysis, engineering service performance evaluation, and structural safety early warning of major water conservancy projects such as pumped storage power stations. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the particle filtering principle of the present invention; Figure 2 This is a flowchart of the multi-source data assimilation technology of the present invention; Figure 3 This is a schematic diagram of the InSAR observation geometry and deformation decomposition of the present invention; Figure 4 This is an overview diagram of the dam and its monitoring system in the study area according to an embodiment of the present invention; Figure 5 This is a spatiotemporal baseline distribution map of SAR images in the study area according to an embodiment of the present invention; Figure 6 This is a monitoring overview diagram of different InSAR methods in the embodiments of the present invention; Figure 7 These are annual average vertical deformation rate diagrams for different InSAR technologies in embodiments of the present invention; Figure 8 This is a comparison diagram of dam longitudinal section leveling observation and InSAR monitoring in an embodiment of the present invention; Figure 9 This is a time-series vertical deformation curve of the monitoring points and InSAR coherence points of the dam cross section in an embodiment of the present invention; Figure 10 This is a comparison chart of errors between different InSAR techniques and leveling measurements in embodiments of the present invention; Figure 11 This is a diagram showing the data fusion result in an embodiment of the present invention. Detailed Implementation

[0030] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0031] This invention provides a fusion method for monitoring the deformation of rockfill dams based on EPS-InSAR and particle filtering, such as... Figure 2 As shown, the specific steps include: Step 1: Differential interferometric preprocessing and phase generation of SAR images; Step 1 is as follows: The image range was determined based on the geographical latitude and longitude coordinates and vector range of the study area, and single-view complex (SLC) SAR image data covering the area were collected. The master image was selected according to the spatiotemporal baseline criterion, and the remaining slave images were registered with it to form a connected image pair. Precise orbit information was introduced and combined with external DEM to simulate the terrain phase and generate a reference ellipsoid phase. Differential interferograms were generated by complex conjugate multiplication to remove the flat terrain effect and initially weaken the influence of terrain on the interferometric phase.

[0032] Step 2: Select target points in PS and DS and perform joint inversion to extract line-of-sight deformation; The selected DS and PS points are jointly processed, and the preliminary deformation rate is estimated based on a linear model inversion. This step introduces enhanced time-series InSAR technology into deformation monitoring. Step 2 is implemented as follows: Step 2.1: Extraction and optimization of PS and DS points; For PS candidate points with stable scattering characteristics, the amplitude deviation index of each target object is calculated through coherence calculation. Pixel points that meet the threshold condition are regarded as permanent scatterer (PS) points. The formula for calculating their amplitude deviation index is as follows:

[0033] In the formula: [0,1] represents the amplitude deviation index. The smaller the value, the stronger the temporal stability of the pixel amplitude, which is more in line with the characteristics of PS. is the standard deviation of the amplitude of a pixel in N time-series SAR images; This represents the average temporal amplitude of the pixel.

[0034] Next, the Kolmogorov-Smirnov Test is used to identify homogeneous pixels with the same statistical properties within the spatial window as candidate points for distributed scatterers (DS), as shown in the following formula:

[0035] In the formula: The Kolmogorov-Smirnov Test statistic is used to determine whether two pixels are statistically homogeneous (SUP) when the statistic is below a certain threshold. This means that when x belongs to The maximum difference of the cumulative distribution function; , The center pixel With neighboring pixels The cumulative distribution function of the amplitude values; N is the number of time-series images. After identifying the DS target, principal component analysis (PCA) is used to adaptively filter and extract the equivalent phase of the homogeneous pixel group.

[0036] Step 2.2: Construction of the joint target point network and phase unwrapping; The PS and DS points are merged to construct a joint target point sparse network. A three-dimensional phase unwrapping method is employed to address the phase ambiguity caused by terrain undulations and atmospheric delay by utilizing the phase gradient relationship in the time and spatial dimensions. Specifically, the selected DS and PS points are combined to construct a hybrid target point network, and differential phase decomposition calculations are performed to separate the effective deformable phase. Specifically:

[0037] In the formula: Residual phase of the terrain; This refers to the phase of surface deformation. This is the atmospheric delayed phase; It represents random noise phase.

[0038] Step 2.3: Deformation Phase Separation and Geocoding A linear deformation model is established, and the residual topographic phase, atmospheric delay phase, and nonlinear deformation phase are iteratively separated using the least squares method. Specifically, the surface deformation phase is... Divided into linear deformation With nonlinear deformation Separation is achieved by utilizing the differences in the spatiotemporal characteristics of the atmosphere and deformation, specifically:

[0039] In the formula: The wavelength is the radar wavelength; T is the time interval. The deformation rate is represented by the line-of-sight (LOS) axis. After obtaining the deformation, the deformation information in the SAR coordinate system is converted to the geographic coordinate system through geocoding to obtain the temporal deformation process and the deformation value along the radar line of sight in the study area.

[0040] Step 3: Spatial Dimension and Resolution Assimilation Mapping Because spaceborne SAR monitoring data differs from conventional point-based monitoring data (such as leveling points and GNSS points) in spatial distribution and resolution, spatial assimilation mapping is required. Step 3 specifically involves: The deformation values ​​observed by InSAR technology along the radar line of sight are converted into vertical deformation (see InSAR observation geometry and deformation decomposition diagram). Figure 3 (As shown), the calculation formula is as follows:

[0041] In the formula: Line-of-sight deformation values ​​obtained from InSAR monitoring; This is the transformed vertical deformation value; This is the radar incident angle.

[0042] Subsequently, the Kriging spatial interpolation method was used to obtain the InSAR vertical deformation data of the corresponding positions of the discrete leveling points, so as to achieve spatial resolution assimilation and matching between the InSAR observation surface and the leveling observation points. The InSAR temporal deformation data was used as the background field, and the field leveling measurement data was used as the observation field.

[0043] Step 4: Multi-source data assimilation and fusion based on particle filter algorithm (see particle filter principle reference) Figure 1 (As shown) To achieve all-weather, large-scale, and high-precision monitoring of surface deformation of rockfill dams in pumped storage power stations, and to overcome the limitations of single monitoring methods, a particle filtering algorithm is used to dynamically fuse large-scale space-based monitoring with high-precision ground-based point-based monitoring. Step 4 is implemented as follows: Step 4.1: Particle initialization and state prediction; Using the InSAR data at time j=0 as a baseline, random noise following a Gaussian distribution is superimposed to generate an initial particle set of N particles, with a weight of 1 / N. A data-driven state transition model is then used to predict the particle state at the next time step. Update the particle set:

[0044] In the formula: This refers to the state at the previous moment; This refers to the temporal variations observed by InSAR; The process noise of the model is assumed to follow a Gaussian distribution with a mean of zero and a variance of Q.

[0045] Step 4.2: Weight Update; When the leveling observation value is obtained at time j At that time, the particle weights are updated using the probability density of the observed data and then normalized. The specific expression is as follows:

[0046] In the formula: Let be the weight of the nth particle at time j; In the state The observations obtained below The probability density.

[0047] Step 4.3: Importance resampling; After a certain number of iterations, the weights of many particles gradually decrease. At this point, the particle set cannot fully represent the posterior probability density. To solve the particle degeneration problem and improve particle efficiency, the specific implementation method is: determining the number of effective particles. Is it below the set diversity threshold? ,if Then, SIR resampling is performed, copying particles with larger weights while discarding particles with smaller weights to avoid particle degradation, thus increasing the effective particle count. The calculation is as follows:

[0048] Step 4.4: State estimation and output; By combining the particles and their weights, the optimal estimate of the particle filter at time j is calculated. The specific expression is as follows:

[0049] The assimilation time is calculated through the above iterations. If the expected accuracy is not met, the parameter combination is optimized and assimilated again until the accuracy requirements are met and the optimal assimilation estimate is output, so as to obtain multi-source deformation monitoring fusion data with complementary advantages and high spatiotemporal resolution.

[0050] Example 1

[0051] In this embodiment, the upper reservoir of the Jurong pumped storage power station is taken as the research target. A data assimilation and fusion model based on EPS-InSAR and particle filtering is established. Sentinel-1A rising orbit single-look complex (SLC) SAR image data covering the area are collected, with image acquisition time from February 2023 to June 2025, totaling 50 rising orbit images. The image acquisition mode is interferometric wide swath (IW), the polarization mode is VV, the revisit period is 12 days, and the spatial resolution is 5m×20m. Copernicus digital elevation model data with a resolution of 30m is used to remove the terrain phase effect. Line-of-sight deformation is extracted according to steps 1 and 2 above, at which point the radar incident angle is... °. Following step 3, the line-of-sight deformation values ​​are converted to vertical deformation, and Kriging interpolation is used to obtain the corresponding location data of the ground leveling points. Then, following step 4, a particle filtering method is used to assimilate and fuse the leveling observations and InSAR data. In the fusion process of step 4, the number of particles is set to generate an initial particle set, which is iteratively updated and the effective particle count is determined for SIR resampling. The overview of the dams and their monitoring systems in the study area is provided in [reference needed]. Figure 4 As shown, the spatiotemporal baseline distribution reference of SAR images Figure 5 As shown.

[0052] (1) Comparative analysis of measurement point density and spatial coverage characteristics of different InSAR technologies The simulation method described above can demonstrate the monitoring capabilities of InSAR in complex surface environments. Figure 6 The measurement point distributions of four InSAR methods—SBAS, PS, ESBAS, and EPS—were compared. The figures show that: ① The measurement point density of PS-InSAR is approximately 1.8 times that of SBAS-InSAR, making it more suitable for man-made structures with dense strong scattering points; ② Enhanced algorithms (ESBAS and EPS) significantly improve measurement point density by combining PS and DS. The EPS method achieves a point density of 21977.369 points / km. 2 Compared to PS-InSAR, it improved by approximately 274%. The points transitioned from "discrete strips" to "area coverage," effectively filling voids in weakly coherent areas and significantly enhancing the continuous coverage capability of the reservoir bank platform and slope protection area.

[0053] (2) Comparative analysis of the spatial distribution of vertical deformation rate of different InSAR technologies Annual average settlement rates obtained from different InSAR techniques, such as Figure 7As shown in the figure, ① all four methods show a trend of larger settlement near the dam crest and smaller settlement at the dam site and joints; ② EPS-InSAR performs best in terms of spatial continuity, settlement center identification, and gradient transition, preserving the overall distribution characteristics of dam deformation while effectively suppressing local abnormal fluctuations; ③ in terms of settlement amplitude, the maximum settlement rate of EPS-InSAR reaches 28.48 mm / year, which is consistent with the settlement pattern during the actual engineering completion and impoundment period.

[0054] (3) Comparative analysis of vertical deformation processes of different InSAR technologies The accuracy of the time series was verified by selecting leveling points on typical longitudinal and transverse sections, such as... Figure 8 and Figure 9 As shown, the leveling settlement curves and the curves from various InSAR techniques show good consistency in overall trend. Quantitative error analysis is as follows: Figure 10 As shown, the RMSE value of EPS-InSAR technology is significantly lower than the other three methods, with smaller error fluctuations and the best accuracy and stability. This indicates that EPS-InSAR can accurately depict the settlement evolution process of pumped storage power station rockfill dams.

[0055] (4) Multi-source monitoring data fusion and optimization analysis The results of the data comparison before and after fusion are as follows Figure 11 As shown in the figure, ① before fusion, although the original InSAR results could reflect the overall subsidence trend, there were high-frequency fluctuations and amplitude deviations in local periods, with an average RMSE of 10.099 mm; ② after fusion, the subsidence trend of each measuring point was more consistent with the leveling measurement, with the consistency index increasing to a maximum of 0.974, and the average RMSE, MAE, and MAPE decreasing by 70.06%, 72.57%, and 61.14%, respectively, with the maximum RMSE reduction reaching 83.37%. ③ The assimilated values ​​effectively reduced local abnormal fluctuations and noise interference. In summary, this invention, by fusing leveling data, compensates for the observation errors of InSAR technology and improves the accuracy and reliability of deformation monitoring data.

[0056] This embodiment is merely a specific application example of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent transformations or improvements based on the technical concept of the present invention, such as changing the network architecture, adjusting the inversion algorithm, or extending to other geophysical methods, should fall within the scope of protection of the present invention.

Claims

1. A fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering, characterized in that: Includes the following steps: Step 1: Differential interferometric preprocessing and phase generation of SAR images; Step 2: Select target points in PS and DS and perform joint inversion to extract line-of-sight deformation; Step 3: Spatial dimension and resolution assimilation mapping; Step 4: Multi-source data assimilation and fusion based on particle filter algorithm.

2. The method for fusion monitoring of rockfill dam deformation based on EPS-InSAR and particle filtering according to claim 1, characterized in that, Step 1 specifically involves: determining the image range based on the geographical latitude and longitude coordinates and vector range of the study area, and collecting single-view complex SAR image data covering the area; selecting the master image based on the spatiotemporal baseline criterion, and registering the remaining secondary images with it to form a connected image pair; introducing precise orbital information and combining it with an external DEM to simulate the terrain phase and generate a reference ellipsoid phase; generating a differential interferogram through complex conjugate multiplication to remove the flat terrain effect and initially weaken the influence of terrain on the interferometric phase.

3. The method for fusion monitoring of rockfill dam deformation based on EPS-InSAR and particle filtering according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Extraction and optimization of PS and DS points; Step 2.2: Construction of the joint target point network and phase unwrapping; Step 2.3: Deformation phase separation and geocoding.

4. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 3, characterized in that, Step 2.1 specifically involves: calculating the stability of pixel amplitude in the time-series image, and designating pixels that meet the threshold condition as permanent scatterer points. The formula for calculating their amplitude deviation index is as follows: In the formula: ADI∈[0,1], is the amplitude deviation index. The smaller the value, the stronger the temporal stability of the pixel amplitude, and the more it conforms to the characteristics of PS. denoted as the standard deviation of the amplitude of a pixel in N temporal SAR images; α is the average temporal amplitude of the pixel. Next, the Kolmogorov-Smirnov Test is used to identify homogeneous pixels with the same statistical properties within the spatial window as candidate points for distributed scatterers. The formula is as follows: In the formula: The Kolmogorov-Smirnov Test statistic is used to determine whether two pixels are statistically homogeneous when the statistic is below a certain threshold. This indicates that the maximum difference of the cumulative distribution function is taken when x is a real number; , The center pixel With neighboring pixels The cumulative distribution function of the amplitude value; N is the number of time-series images; after identifying the DS target, principal component analysis is used to adaptively filter and extract the equivalent phase of the homogeneous pixel group.

5. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 3, characterized in that, Step 2.2 specifically involves: combining the selected DS points and PS points to construct a hybrid target point network, performing differential phase decomposition calculations, and separating the effective deformed phase. Specifically: In the formula: Residual phase of the terrain; This refers to the phase of surface deformation. This is the atmospheric delayed phase; It represents random noise phase.

6. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 3, characterized in that, Step 2.3 specifically involves: determining the surface deformation phase. Divided into linear deformation With nonlinear deformation Separation is achieved by utilizing the differences in the spatiotemporal characteristics of the atmosphere and deformation, specifically: In the formula: The wavelength is the radar wavelength; T is the time interval. The deformation rate of the structure is LOS. After determining the deformation, the deformation information in the SAR coordinate system is converted into the geographic coordinate system through geocoding to obtain the temporal deformation process and radar line-of-sight deformation value of the study area.

7. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 1, characterized in that, Step 3 specifically involves converting the deformation values ​​observed by InSAR technology along the radar line-of-sight direction into vertical deformation, calculated using the following formula: In the formula: Line-of-sight deformation values ​​obtained from InSAR monitoring; This is the transformed vertical deformation value; The radar incident angle; Subsequently, the Kriging spatial interpolation method was used to obtain the InSAR vertical deformation data of the corresponding positions of the discrete leveling points, so as to achieve spatial resolution assimilation and matching between the InSAR observation surface and the leveling observation points. The InSAR temporal deformation data was used as the background field, and the field leveling measurement data was used as the observation field.

8. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 1, characterized in that, Step 4 specifically includes: Step 4.1: Particle initialization and state prediction; Step 4.2: Weight Update; Step 4.3: Importance resampling; Step 4.4: State estimation and output.

9. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 8, characterized in that, Specifically, step 4.1 involves using the InSAR data at time j=0 as a baseline, superimposing random noise following a Gaussian distribution to generate an initial particle set containing N particles, with a weight of 1 / N; and then using a data-driven state transition model to predict the particle state at the next time step. Update the particle set: In the formula: This refers to the state at the previous moment; ε represents the temporal variation observed by InSAR; ε is the process noise of the model, which is assumed to follow a Gaussian distribution with zero mean and variance Q.

10. The fusion method for monitoring deformation of rockfill dams based on EPS-InSAR and particle filtering according to claim 8, characterized in that, Step 4.3 specifically involves determining the number of effective particles. Is it below the set diversity threshold? ,if Then, SIR resampling is performed, copying particles with larger weights while discarding particles with smaller weights to avoid particle degradation, thus increasing the effective particle count. The calculation is as follows: Step 4.4 specifically involves: By combining the particles and their weights, the optimal estimate of the particle filter at time j is calculated. The specific expression is as follows: The assimilation time is calculated through the above iterations until the accuracy requirements are met and the optimal assimilation estimate is output, thus obtaining multi-source deformation monitoring fusion data with complementary advantages and high spatiotemporal resolution.

Images