Multidimensional Deformation Monitoring and Prediction Method of Creep Landslide Time Series InSAR Technology

By combining multidimensional small baseline ensemble aperture radar interferometry and LSTM network with prior information, the problem of insufficient data for landslide three-dimensional deformation monitoring was solved, achieving high-precision landslide three-dimensional deformation monitoring and prediction, and improving the accuracy of landslide early warning.

CN117008128BActive Publication Date: 2026-11-10EAST CHINA UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310687932.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-12
Publication Date
2026-11-10
Estimated Expiration
2043-06-12

AI Technical Summary

Technical Problem

Existing technologies struggle to obtain three-dimensional deformation information of landslides when multi-source monitoring data is limited, and commonly used SAR platforms can only acquire two-dimensional deformation data, which restricts the accuracy of landslide deformation prediction and early warning.

Method used

By employing multidimensional small baseline ensemble aperture radar interferometry technology combined with prior information constraints and a long short-term memory neural network (LSTM), the three-dimensional deformation information of landslides is obtained by collecting various image data and DEM models, and future deformation is predicted.

Benefits of technology

It enables the acquisition of three-dimensional deformation information of landslides with less data, reduces data requirements, improves the accuracy of deformation monitoring and prediction, overcomes the problems of gradient vanishing and long time series dependence, and provides more accurate landslide early warning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117008128B_ABST
    Figure CN117008128B_ABST
Patent Text Reader

Abstract

This application relates to a multi-dimensional deformation monitoring and prediction method for creep-type landslides using InSAR technology, comprising the following steps: collecting ascending Sentinel-1A images, descending Sentinel-1A images, precise orbit data, and the Space Shuttle Radar Topography Mission Digital Elevation Model (SRTM) and the Copernicus Digital Elevation Model (COPM); obtaining one-dimensional landslide deformation based on small baseline ensemble aperture radar interferometry (LSTM); obtaining three-dimensional landslide deformation based on multi-dimensional small baseline ensemble aperture radar interferometry combined with prior information constraints; and predicting future landslide deformation based on a long short-term memory neural network (LSTM) to obtain the future temporal deformation results of landslide feature points. This invention can obtain three-dimensional landslide deformation information by adding reasonable prior information as constraints when multi-source monitoring data is limited, providing a reasonable basis for geological disaster prevention and control.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of data processing and neural network prediction technology in synthetic aperture radar interferometry, specifically to a multidimensional deformation monitoring and prediction method for creep-type landslides using time-series InSAR technology. Background Technology

[0002] Synthetic Aperture Radar (SAR) technology has been widely used in geological disasters such as earthquakes, volcanoes, glaciers, landslides and ground subsidence. After decades of development, Synthetic Aperture Radar Interferometry (InSAR) and its development of Temporal InSAR have been widely used to map and monitor large-scale landslide displacements due to their ability to measure minute surface deformations with high precision.

[0003] In the field of three-dimensional deformation monitoring of landslides, since SAR is not sensitive to deformation in the north-south direction, the deformation in the north-south direction is generally ignored and reduced to the two-dimensional direction. However, compared with two-dimensional deformation, three-dimensional deformation is more beneficial for interpreting landslide deformation fields, landslide prediction and risk management. To reconstruct three-dimensional deformation information, at least three observation results with significant imaging geometric differences are required. However, commonly used SAR platforms can generally only obtain two observation results with significant imaging geometric differences, and this limitation needs to be overcome.

[0004] The economic losses and subsequent costs of landslide control far exceed the expenses incurred in monitoring and predicting landslide deformation before instability. Therefore, landslide deformation prediction and early warning are essential. Geological disaster prevention and control work cannot be separated from monitoring and prediction. Combining the two can enable timely understanding of the current and future deformation characteristics of landslides, providing a reasonable basis for geological disaster prevention and control. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional deformation monitoring and prediction method for creep-type landslides using time-series InSAR technology. This method can obtain three-dimensional deformation information of landslides by adding reasonable prior information as constraints when there is limited multi-source monitoring data, thus providing a reasonable basis for geological disaster prevention and control.

[0006] The technical solution adopted in this invention is:

[0007] A multi-dimensional deformation monitoring and prediction method for creep-type landslides using time-series InSAR technology includes the following steps:

[0008] S1: Collect Sentinel-1A images of ascending orbit, Sentinel-1A images of descending orbit, precise orbit data, and the Space Shuttle Radar Terrain Mapping Mission Digital Elevation Model (SRTM) and the Copernicus Digital Elevation Model (Copernicus).

[0009] S2: Obtaining one-dimensional deformation of landslides based on small baseline ensemble aperture radar interferometry;

[0010] S3: Obtaining three-dimensional landslide deformation based on multi-dimensional small baseline ensemble aperture radar interferometry combined with prior information constraints;

[0011] S4: Predicting future landslide deformation based on a Long Short-Term Memory (LSTM) neural network. Each neuron in the LSTM network contains a forget gate, an input gate, and an output gate, which calculates the time-series deformation d of landslide feature points. N,E,U The long short-term memory neural network (LSTM) is input for iterative training to obtain the future temporal deformation results of landslide feature points.

[0012] Furthermore, the specific method of step S2 is as follows:

[0013] S201: Select control points, set appropriate spatiotemporal baseline thresholds, perform image registration on the ascending Sentinel-1A image and the descending Sentinel-1A image respectively, establish the coordinate mapping relationship between the main image and the sub-image, and perform coordinate transformation and pixel value resampling on the sub-image to complete SAR image registration.

[0014] S202: After the main and sub-images are registered, the pixel information of the image is represented by a complex number. The complex interferometric value of the pixel is obtained by multiplying the corresponding pixels of the main image and the resampled sub-image by a complex conjugate, and an interferogram is generated.

[0015] S203: Filter the generated interferogram;

[0016] S204: Perform phase unwrapping on the measured interferogram;

[0017] S205: The relationship between the deformation rate and the interference phase between adjacent acquisition times is obtained by using the singular value decomposition (SVD) method, thereby obtaining the velocity vector between adjacent acquisition times.

[0018] S206: Integrate the velocity vectors of adjacent acquisition times to obtain the deformation at the corresponding time, and then obtain the cumulative deformation of the pixels and the average deformation rate, that is, the deformation information of the radar line of sight to the ground surface.

[0019] Further, in step S206, the deformation characteristics of a single-track SAR image with an incident angle of θ and an azimuth angle of α are obtained using the SBAS method. The specific calculation formula is as follows:

[0020] AV LOS =Φ obs ;

[0021]

[0022] in, Indicates t iConstantly shifting the line of sight. Indicates t i+1 The line-of-sight displacement at any given time, matrix A is a matrix constructed from the time intervals between consecutive SAR data, V LOS The deformation rate Δt represents the line-of-sight (LOS) direction of the radar. i Represented as t i With t i+1 The time interval, Φ obs It is a vector of the observed interferogram values; Indicates t i The deformation rate of the radar line-of-sight (LOS) at any given time;

[0023] The generalized inverse matrix A of matrix A is obtained using the singular value decomposition (SVD) method. + The deformation rate V of the radar line-of-sight (LOS) can be obtained. LOS =A + Φ obs .

[0024] Furthermore, the specific method of step S3 is as follows:

[0025] S301: Calculate the one-dimensional deformation of the ascending and descending orbits; the line-of-sight deformation rate V of the ascending and descending orbit SAR image after SBAS processing. LOS It can be represented as:

[0026]

[0027] in, This indicates the line-of-sight deformation rate of the ascending orbit. V is expressed as the line-of-sight deformation rate of the descending orbit. E V represents the deformation rate in the horizontal direction. U V represents the deformation rate in the vertical direction. N θ represents the deformation rate in the north-south direction. asc α represents the angle of incidence of the ascending orbit. asc θ represents the azimuth angle of the ascending orbit. desc α represents the angle of incidence of the descending orbit. desc Indicates the azimuth angle of the descending orbit;

[0028] S302: With additional prior information, the surface parallel flow method is used to fuse rising and falling orbit SAR images to conduct three-dimensional deformation monitoring of creep-type landslides and calculate the three-dimensional deformation. The specific method is as follows:

[0029] Using the Copernicus Digital Elevation Model (DEM) as prior information, the first derivatives of pixels in the east-west and north-south directions are extracted from the Copernicus DEM. and Furthermore, functional relationships between deformation rates in the vertical, east-west, and north-south directions are constructed, i.e. Where H represents elevation; the three-dimensional deformation calculation is performed using the multidimensional small baseline set (MSABS) technique, and the specific formula is as follows:

[0030]

[0031]

[0032] G = [s N G,s E G,s U G]=[Gsinθcosα,-Gsinθcosα,Gcosθ];

[0033] Wherein, matrix G is a coefficient matrix, composed of the contributions of the deformation of adjacent SAR image time interval matrix G and radar line-of-sight (LOS) in the east-west, vertical, and north-south directions, s N s represents the unit vector of the line of sight in the north-south direction. E s represents the unit vector of the line of sight in the east-west direction. U Let H represent the unit vector perpendicular to the line of sight, and let H be the constraint condition for parallel flow on the Earth's surface. L represents the regularization matrix, and λ represents the Tikhonov regularization parameter. Indicates t i+1 The three-dimensional cumulative deformation at time t, Indicates t i The three-dimensional deformation rate at time t, where Φ represents the InSAR observation value of the ascending and descending orbits.

[0034] Furthermore, the formula for the forget gate or input gate is as follows:

[0035] C t =f t *C t-1 +i t *C t ;

[0036] The formula for the output gate is as follows:

[0037] h t =o t *tanh(C t );

[0038] Among them, C t-1 The information state at time t-1 is represented by the selective determination of information retention and forgetting within the cell; f t This is represented as the determination of information state C. t-1 The retention or removal of internal information; i t Let C be the candidate vector for determining the outcome. tThe section used to update the information status; C t Represented as a candidate vector for the newly added cell state; h t Represented as an output vector; o t This is represented as the determination of information state C. t The output section.

[0039] The beneficial effects of this invention are as follows:

[0040] (1) For translational landslides, the surface parallel flow method can be used to obtain prior information from the DEM as a constraint condition to obtain the three-dimensional deformation of the landslide. This can reduce the stringent requirements of three-dimensional deformation monitoring on data and achieve the goal of obtaining the three-dimensional deformation of the landslide by only using ascending and descending orbit InSAR observation.

[0041] (2) The LSTM model is used to predict the time series of landslide feature points. The model overcomes the problems of gradient vanishing or exploding and the inability to learn long-term dependencies in long time series. By setting the internal structure of the network, the network can learn the nonlinear correlation in long-term series data, further improving the prediction accuracy. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram illustrating the relationship between radar line-of-sight deformation and the actual three-dimensional deformation of the slope in an embodiment of the present invention.

[0045] Figure 3 This is a schematic diagram of the structure of the Long Short-Term Memory (LSTM) neural network in an embodiment of the present invention;

[0046] Figure 4 The three-dimensional deformation rate of the Shadong landslide;

[0047] Figure 5 This is a schematic diagram of the LSTM prediction results for the three-dimensional deformation of feature points of the Shadong landslide according to an embodiment of the present invention. Detailed Implementation

[0048] To better understand the above-described objects, features, and advantages of the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Many specific details are set forth in the following description to provide a thorough understanding of the invention; however, the invention may be practiced in other ways different from those described herein, and therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art described herein. The terms “first,” “second,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “connected” or “linked” and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms “upper,” “lower,” “left,” “right,” etc., are used only to indicate relative positional relationships, which change accordingly when the absolute position of the described object changes.

[0050] like Figures 1-5 As shown, the multidimensional deformation monitoring and prediction method of creep-type landslides using time-series InSAR technology includes the following steps:

[0051] S1: Collect Sentinel-1A images of ascending orbit, Sentinel-1A images of descending orbit, precise orbit data, and the Space Shuttle Radar Terrain Mapping Mission Digital Elevation Model (SRTM) and the Copernicus Digital Elevation Model (Copernicus).

[0052] S2: Obtaining one-dimensional deformation of landslides based on small baseline ensemble aperture radar interferometry;

[0053] S201: Select control points, set appropriate spatiotemporal baseline thresholds, perform image registration on the ascending Sentinel-1A image and the descending Sentinel-1A image respectively, establish the coordinate mapping relationship between the main image and the sub-image, and perform coordinate transformation and pixel value resampling on the sub-image to complete SAR image registration.

[0054] SAR image registration is the first step in interferometric processing. Its core idea is to calculate the coordinate mapping relationship of corresponding points in two SAR images that constitute an interferometric pair, and to sample the sub-image into the same pixel grid as the main image according to the mapping relationship, so that the corresponding points in the two images correspond to the same resolution cell on the ground.

[0055] S202: After the main and sub-images are registered, the pixel information of the images is represented by complex numbers. The complex interferometric values ​​of the pixels are obtained by multiplying the corresponding pixels of the main image and the resampled sub-image by complex conjugates, and an interferogram is generated.

[0056] S203: Filter the generated interferogram;

[0057] Because the imaging and interferometric processing of SAR systems results in a large amount of noise in the generated interferograms, which manifests as discontinuities, unclear phases, and unclear periodicity, it is necessary to filter the interferograms to reduce noise, make the phase more continuous, and facilitate subsequent unwrapping.

[0058] S204: Perform phase unwrapping on the measured interferogram;

[0059] The phase range of the interferogram obtained in actual measurement is [-π, π). If the measured phase exceeds the range, it is necessary to add an integer multiple of 2π or subtract the measured phase until it is wrapped within the range. The phase unwrapping process is to recover the true phase from the wrapped phase. Its basic idea is to integrate the difference value of the wrapped phase.

[0060] S205: The relationship between the deformation rate and the interference phase between adjacent acquisition times is obtained by using the singular value decomposition (SVD) method, thereby obtaining the velocity vector between adjacent acquisition times.

[0061] S206: Integrate the velocity vectors of adjacent acquisition times to obtain the deformation at the corresponding time, and then obtain the cumulative deformation of the pixels and the average deformation rate, that is, the deformation information of the radar line of sight to the ground surface.

[0062] The deformation characteristics of a single-track SAR image with an incident angle of θ and an azimuth angle of α are obtained using the SBAS method. The specific calculation formula is as follows:

[0063] AV LOS =Φ obs ;

[0064]

[0065] in, Indicates t i Constantly shifting the line of sight. Indicates t i+1 The line-of-sight displacement at any given time, matrix A is a matrix constructed from the time intervals between consecutive SAR data, V LOS The deformation rate Δt represents the line-of-sight (LOS) direction of the radar. i Represented as t i With t i+1 The time interval, Φ obs It is a vector of the observed interferogram values; Indicates t i The rate of deformation of the line-of-sight (LOS) in the radar line of sight at any given time.

[0066] The generalized inverse matrix A of matrix A is obtained using the singular value decomposition (SVD) method. + Then V LOS =A + Φ obs .

[0067] S3: Obtaining three-dimensional landslide deformation based on multi-dimensional small baseline ensemble aperture radar interferometry combined with prior information constraints;

[0068] S301: Calculate the one-dimensional deformation of the ascending and descending orbits; the line-of-sight deformation rate V of the ascending and descending orbit SAR image after SBAS processing. LOS It can be represented as:

[0069]

[0070] in, This indicates the line-of-sight deformation rate of the ascending orbit. V is expressed as the line-of-sight deformation rate of the descending orbit. E V represents the deformation rate in the horizontal direction. U V represents the deformation rate in the vertical direction. N θ represents the deformation rate in the north-south direction. asc α represents the angle of incidence of the ascending orbit. asc θ represents the azimuth angle of the ascending orbit. desc α represents the angle of incidence of the descending orbit. desc This indicates the azimuth angle of the descending orbit.

[0071] S302: The relationship between radar line-of-sight deformation and the actual three-dimensional deformation of the slope is as follows: Figure 2 As shown in the figure, θ represents the angle of incidence, α represents the azimuth angle, and D... N North-south deformation, D E Indicates east-west deformation, D U This represents vertical deformation. With additional prior information, a surface parallel flow method is used to fuse rising and falling orbit SAR imagery for three-dimensional deformation monitoring of creep-type landslides, calculating the three-dimensional deformation. The specific method is as follows:

[0072] To more closely approximate the current state of the landslide, the Copernicus Digital Elevation Model (DEM), which has a more recent data, was used as prior information. The first derivatives of the pixels in the east-west and north-south directions were extracted from the Copernicus DEM. and Furthermore, functional relationships between deformation rates in the vertical, east-west, and north-south directions are constructed, i.e. Where H represents elevation; the three-dimensional deformation calculation is performed using the multidimensional small baseline set (MSABS) technique, and the specific formula is as follows:

[0073]

[0074]

[0075] G = [s N G,s E G,s U G]=[Gsinθcosα,-Gsinθcosα,Gcosθ];

[0076] Wherein, matrix G is a coefficient matrix, composed of the contributions of the deformation of adjacent SAR image time interval matrix G and radar line-of-sight (LOS) in the east-west, vertical, and north-south directions, s N s represents the unit vector of the line of sight in the north-south direction. E s represents the unit vector of the line of sight in the east-west direction. U Let H represent the unit vector perpendicular to the line of sight, and let H be the constraint condition for parallel flow on the Earth's surface. L represents the regularization matrix, and λ represents the Tikhonov regularization parameter. Indicates t i+1 The three-dimensional cumulative deformation at time t, Indicates t i The three-dimensional deformation rate at time t, where Φ represents the InSAR observation value of the ascending and descending orbits.

[0077] In this embodiment of the invention, the Tikhonov regularization method is used to eliminate the rank deficiency problem of the equation, the three-dimensional deformation rate is obtained by the singular value decomposition method (SVD), and the three-dimensional displacement time series is obtained by numerical integration.

[0078] S4: Predicting future landslide deformation based on a Long Short-Term Memory (LSTM) neural network. Each neuron in the LSTM network contains a forget gate, an input gate, and an output gate, which calculates the time-series deformation d of landslide feature points. N,E,U The long short-term memory neural network (LSTM) is input for iterative training to obtain the future temporal deformation results of landslide feature points.

[0079] The formula for the forget gate or input gate is as follows:

[0080] C t =f t *C t-1 +i t *C t ;

[0081] The formula for the output gate is as follows:

[0082] ht =o t *tanh(C t );

[0083] Among them, C t-1 The information state at time t-1 is represented by the selective determination of information retention and forgetting within the cell; f t This is represented as the determination of information state C. t-1 The retention or removal of internal information; i t Let C be the candidate vector for determining the outcome. t The section used to update the information status; C t Represented as a candidate vector for the newly added cell state; h t Represented as an output vector; o t This is represented as the determination of information state C. t The output section.

[0084] This invention employs a 30m resolution SRTM DEM to remove errors in the interferometric processing, uses the Copernicus DEM as a priori information constraint, and combines the observation results of the riser and fallr orbit to obtain the three-dimensional deformation information of the landslide. This reduces the stringent requirements of three-dimensional deformation monitoring on data, and achieves the goal of obtaining the three-dimensional deformation of the landslide using only riser and fallr orbit InSAR observations. Furthermore, it uses a long short-term memory neural network LSTM to predict the future temporal deformation information of landslide feature points.

[0085] The experimental area was selected as the Jinsha River basin in Mindu Township, Gongjue County, Tibet. The Shadong landslide was used as the experimental object, and the collected data parameters are shown in Table 1.

[0086] Table 1 SAR Data Parameters

[0087]

[0088] Figure 4 , Figure 5The results of three-dimensional deformation monitoring of landslides using the method described in this embodiment of the invention show that the maximum deformation rates of the Shadong landslide in the east-west, vertical, and north-south directions are 211.44 mm / year, -169.87 mm / year, and 307.68 mm / year, respectively. Meanwhile, the three-dimensional deformation results of landslide feature points show that the deformation on the west side of the landslide is mainly in the east-west and north-south directions, while the vertical deformation is relatively small. At the leading edge feature points, the east-west deformation reaches approximately 600 mm, the north-south deformation reaches approximately 300 mm, and the vertical deformation is approximately 50 mm. The three-dimensional deformation of feature points in the middle and rear edges is within 200 mm. On the east side of the landslide, the deformation is mainly in the north-south and vertical directions. At the leading and middle feature points, the east-west deformation is within 200 mm, the north-south deformation exceeds 400 mm, and the vertical deformation exceeds 300 mm. The three-dimensional deformation of feature points at the rear edge is within 200 mm. The time-series prediction results show that the three-dimensional deformation of the feature points will continue to increase. The model can obtain appropriate prediction results for periodic and slowly deformed samples, while the results obtained for sample data with large deformation at the landslide front are consistent with its movement trend.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-dimensional deformation monitoring and prediction method for creep-type landslides using time-series InSAR technology, characterized in that, Includes the following steps: S1: Collect Sentinel-1A images of ascending orbit, Sentinel-1A images of descending orbit, precise orbit data, and the Space Shuttle Radar Terrain Mapping Mission Digital Elevation Model (SRTM) and the Copernicus Digital Elevation Model (Copernicus). S2: Obtaining one-dimensional deformation of landslides based on small baseline ensemble aperture radar interferometry; S3: Obtaining the three-dimensional deformation of landslides based on multi-dimensional small baseline ensemble aperture radar interferometry combined with prior information constraints; the specific method is as follows: S301: Calculation of one-dimensional deformation of the ascent and descent orbit; line-of-sight deformation rate of ascent and descent orbit SAR image after SBAS processing. It can be represented as: ; in, This indicates the line-of-sight deformation rate of the ascending orbit. Expressed as the line-of-sight deformation rate in the descending orbit. Represents the deformation rate in the horizontal direction. Represents the deformation rate in the vertical direction, θ represents the deformation rate in the north-south direction. asc α represents the angle of incidence of the ascending orbit. asc θ represents the azimuth angle of the ascending orbit. desc α represents the angle of incidence of the descending orbit. desc Indicates the azimuth angle of the descending orbit; S302: With additional prior information, the surface parallel flow method is used to fuse rising and falling orbit SAR images to conduct three-dimensional deformation monitoring of creep-type landslides and calculate the three-dimensional deformation. The specific method is as follows: Using the Copernicus Digital Elevation Model (DEM) as prior information, the first derivatives of pixels in the east-west and north-south directions are extracted from the Copernicus DEM. and This leads to the construction of functional relationships between deformation rates in the vertical, east-west, and north-south directions, i.e. Where H represents elevation; the three-dimensional deformation is calculated using the multidimensional small baseline set technique MSABS, and the specific formula is as follows: ; ; ; Among them, matrix The coefficient matrix consists of the contributions of the deformation of adjacent SAR image time interval matrix G and radar line-of-sight (LOS) in the east-west, vertical, and north-south directions. This represents the unit vector representing the line of sight in the north-south direction. This represents the unit vector representing the line of sight in the east-west direction. This represents the unit vector pointing vertically towards the line of sight. The constraint condition for surface parallel flow is, i.e. L represents the regularization matrix. This represents the Tikhonov regularization parameter. express The three-dimensional cumulative deformation at time t, express The three-dimensional deformation rate at time t, This represents InSAR observations from the ascending and descending orbits; S4: Predicting future landslide deformation based on a Long Short-Term Memory (LSTM) neural network. Each neuron in the LSTM network contains a forget gate, an input gate, and an output gate, which calculates the time-series deformation d of landslide feature points. N,E,U The long short-term memory neural network (LSTM) is input for iterative training to obtain the future temporal deformation results of landslide feature points.

2. The method for multidimensional deformation monitoring and prediction of creep-type landslides using time-series InSAR technology according to claim 1, characterized in that, The specific method for step S2 is as follows: S201: Select control points, set appropriate spatiotemporal baseline thresholds, perform image registration on the ascending Sentinel-1A image and the descending Sentinel-1A image respectively, establish the coordinate mapping relationship between the main image and the sub-image, and perform coordinate transformation and pixel value resampling on the sub-image to complete SAR image registration. S202: After the main and sub-images are registered, the pixel information of the image is represented by a complex number. The complex interferometric value of the pixel is obtained by multiplying the corresponding pixels of the main image and the resampled sub-image by a complex conjugate, and an interferogram is generated. S203: Filter the generated interferogram; S204: Perform phase unwrapping on the measured interferogram; S205: The relationship between the deformation rate and the interference phase between adjacent acquisition times is obtained by using the singular value decomposition method (SVD), and then the velocity vector between adjacent acquisition times is obtained. S206: Integrate the velocity vectors of adjacent acquisition times to obtain the deformation at the corresponding time, and then obtain the cumulative deformation of the pixels and the average deformation rate, that is, the deformation information of the radar line of sight to the ground surface.

3. The method for multidimensional deformation monitoring and prediction of creep-type landslides using time-series InSAR technology according to claim 2, characterized in that, In step S206, for the incident angle is And the azimuth angle is The deformation characteristics of single-track SAR images were obtained using the SBAS method. The specific calculation formula is as follows: ; ; in, express Constantly shifting the line of sight. express The line-of-sight displacement at any given time, matrix A is a matrix constructed from the time intervals between consecutive SAR data. This represents the deformation rate of the radar line-of-sight (LOS). Represented as and The time interval of time, It is a vector of the observed interferogram values; express The deformation rate of the radar line-of-sight (LOS) at any given time; The generalized inverse matrix of matrix A is obtained using the singular value decomposition (SVD) method. The deformation rate of the line-of-sight (LOS) in the radar line-of-sight direction can be obtained. .

4. The method for multidimensional deformation monitoring and prediction of creep-type landslides using time-series InSAR technology according to claim 3, characterized in that, The formula for the forget gate or input gate is as follows: ; The formula for the output gate is as follows: ; in, Represented as the information state at time t-1, the retention or forgetting of information is selectively determined within the cell; This indicates the determination of the information state. The retention or removal of internal information; Represented as the candidate vector The part used to update the information status; This is represented as a candidate vector for the newly added cell state; Represented as an output vector; This indicates the determination of the information state. The output section.

Citation Information

Patent Citations

  • Split and recombined passive micromixer with bridge-type structure

    CN104138728A

  • Systems for tsunami detection and warning

    CN107369297A