Landslide area dynamic intelligent prediction method based on time-series insar and multi-feature clustering
By constructing a technical framework that deeply couples and collaboratively optimizes multiple features such as location, terrain, and velocity, and combining the improved DBSCAN clustering algorithm and Thiessen polygon calculation, the accuracy and rationality issues of landslide identification and area prediction are solved, realizing dynamic intelligent prediction and risk assessment of landslide area.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNIV OF GEOSCIENCES (WUHAN)
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-29
Smart Images
Figure CN121600053B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of dynamic intelligent prediction technology of landslide area, and in particular to a dynamic intelligent prediction method for landslide area based on temporal InSAR and multi-feature clustering. Background Technology
[0002] Synthetic aperture radar interferometry, especially PS-InSAR and SBAS-InSAR, has become a core tool for landslide deformation monitoring. It can retrieve large-scale temporal deformation fields based on multi-period SAR imagery and is widely used in geological disaster prevention and urban safety. Meanwhile, clustering analysis techniques such as K-means and DBSCAN, due to their unsupervised and high-efficiency characteristics, have been introduced into landslide identification and boundary extraction, enabling automatic classification based on features such as deformation and topography.
[0003] However, existing technical solutions still have obvious shortcomings, which restrict the accuracy and rationality of landslide identification and area calculation. The main shortcomings are: (1) Insufficient use of location information: The existing InSAR processing flow separates location information from velocity and slope, and treats it only as an auxiliary label for data. It does not play its core role of "spatial constraint" and "regional division", and does not realize the spatial correlation logic of "location-topography-velocity". However, landslides should have the inherent logic of continuity and aggregation of specific terrain units in space. (2) Insufficient spatial coupling of velocity-slope-location: In the existing technology, velocity reflects dynamic deformation, slope reflects static terrain, and location information reflects spatial correlation. Velocity, slope and location are input into the clustering algorithm as independent data, and spatial correlation verification is not performed, which makes it impossible to play the complementarity between features and not establish the spatial coupling relationship among the three. (3) Imbalance of spatial feature weights: In traditional clustering algorithms, the "dynamic significance of velocity" is overemphasized, while the "spatial constraint of location" is ignored. As a physical geological body, the spatial distribution of landslides exhibits "continuity," and the velocity only reflects "activity intensity." Location information is needed to ensure the "spatial rationality" of the clustering results. Summary of the Invention
[0004] To address the aforementioned challenges, this application provides a dynamic intelligent prediction method for landslide area based on temporal InSAR and multi-feature clustering. By constructing a technical framework that deeply couples and collaboratively optimizes multiple features such as location, terrain, and velocity, it overcomes the shortcomings of traditional methods, such as feature fragmentation, lack of spatial constraints, and weight imbalance, and significantly improves the accuracy, continuity, and spatiotemporal rationality of landslide identification and area prediction.
[0005] To achieve the above objectives, this application provides a dynamic intelligent prediction method for landslide area based on temporal InSAR and multi-feature clustering, comprising the following steps:
[0006] S1: Acquire multi-phase synthetic aperture radar images, digital elevation model data and atmospheric correction data of the target area, process them using the SBAS-InSAR algorithm, construct the model inversion deformation rate, and obtain deformation rate raster data.
[0007] S2: Based on deformation rate raster data and digital elevation model data, extract annual deformation rate, slope attributes and spatial coordinate location information, and construct coupled features;
[0008] S3: Standardize and reduce the dimensionality of the coupling features to obtain the optimized feature matrix;
[0009] S4: The improved DBSCAN clustering algorithm is used to perform cluster analysis on the feature matrix to obtain the landslide cluster labels;
[0010] S5: Filter landslide points based on cluster labels, construct Thiessen polygons, and calculate the landslide area;
[0011] S6: Repeat steps S1 to S5 to obtain landslide area data for multiple periods and perform time series analysis and trend prediction.
[0012] Preferably, S1 specifically includes:
[0013] S11: Preprocess synthetic aperture radar images;
[0014] S12: Based on the condition that the time baseline does not exceed the first threshold and the spatial baseline does not exceed the second threshold, candidate interference pairs are selected and the phase difference is calculated;
[0015] S13: The minimum cost flow algorithm is used for phase unwrapping, and the terrain phase and atmospheric delay phase are eliminated by using digital elevation model and atmospheric correction data, respectively;
[0016] S14: Based on the SBAS-InSAR algorithm, phase time series analysis of multi-period interferometric pairs is performed to construct a model for inverting deformation rate and output deformation rate raster data.
[0017] Preferably, the deformation rate inversion is expressed as:
[0018] ;
[0019] in, For deformation phase difference, For image time intervals, The wavelength of the synthetic aperture radar image;
[0020] The interference phase composition is represented as follows:
[0021]
[0022] in, For deformation phase, For terrain phase, For atmospheric delayed phase, This is the noise phase.
[0023] Preferably, S2 specifically includes:
[0024] Deformation rate turning point data are obtained based on deformation rate raster data;
[0025] Slope turning point data were obtained based on digital elevation model data;
[0026] Spatially connect the deformation rate turning point data and the slope turning point data, and associate the slope attribute value with the attribute table of the deformation rate point feature;
[0027] Spatial coordinate information is added to each spatially connected point element to form a coupled feature consisting of deformation rate, slope attribute and spatial coordinates.
[0028] Preferably, S3 specifically includes:
[0029] The coupling features are standardized using the Z-score standardization method.
[0030] Principal component analysis was performed on the standardized coupling features to reduce their dimensionality, resulting in an optimized feature matrix.
[0031] Preferably, the Z-score normalized representation is as follows:
[0032] ;
[0033] in, The standardized value. These are the original eigenvalues. The mean of the original features. denoted as the standard deviation of the original features.
[0034] Preferably, S4 specifically includes:
[0035] Optimize neighborhood radius and minimum number of points based on contour coefficient;
[0036] Dynamically adjust clustering parameters based on interannual data characteristics;
[0037] Clustering is performed on the optimized feature matrix to obtain landslide cluster labels.
[0038] Preferably, the contour coefficient is expressed as:
[0039] ;
[0040] in, For the sample The average distance to other samples in the same cluster, For the sample i The average distance to all samples in the nearest heterogeneous cluster.
[0041] The neighborhood is defined as:
[0042]
[0043] in, For the sample i and j Euclidean distance, is the neighborhood radius.
[0044] Preferably, the Tyson polygon in S5 is a polygon formed by perpendicular bisectors between adjacent points, with each landslide point as the center.
[0045] Preferably, the time series analysis and trend prediction in S6 include:
[0046] The area data of landslides from multiple periods were compiled into an area time series table;
[0047] By fitting an area change model using linear regression, and combining line graphs and fitted curves, the interannual trend is presented intuitively.
[0048] Output the interannual area change rate and goodness-of-fit index to achieve dynamic prediction of landslide area.
[0049] Therefore, the above-mentioned landslide area dynamic intelligent prediction method based on temporal InSAR and multi-feature clustering has the following beneficial effects:
[0050] (1) This application constructs a refined SBAS-InSAR processing link, filters out low-quality interfering pairs by using a coherence threshold (≥0.3), and combines minimum cost flow phase unwrapping and atmospheric correction techniques to effectively suppress spatiotemporal decoherence and atmospheric disturbances, and output high-precision annual deformation rate data.
[0051] (2) This application constructs a complete time-series monitoring system. By processing SBAS-InSAR data year by year and converting landslide points into affected areas based on Thiessen polygons, it achieves accurate quantitative statistics of landslide area. Furthermore, it uses linear regression analysis to analyze its interannual variation trend, extending the monitoring results from single state description to long-term trend prediction, providing a scientific quantitative basis for dynamic prevention and control and risk assessment of landslides.
[0052] (3) This application proposes a data-driven method for dynamically determining DBSCAN parameters. Based on the multi-feature distribution of “deformation rate-slope-coordinate”, the MinPts selection is optimized using the profile coefficient, and a time-adaptive mechanism is established. This effectively solves the adaptability defects of fixed parameters in interannual landslide monitoring and significantly improves the distinguishability between landslide clusters and noise points.
[0053] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0054] Figure 1 This is a flowchart illustrating a method for dynamic intelligent prediction of landslide area based on temporal InSAR and multi-feature clustering in this application.
[0055] Figure 2 This is a time-series analysis diagram of the landslide area in an embodiment of this application. Detailed Implementation
[0056] The following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0057] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning as understood by a person of ordinary skill in the art to which this application pertains.
[0058] The terms "comprising" or "including," as used in this application, mean that the element preceding the term encompasses the element listed after the term, and do not exclude the possibility of encompassing other elements as well. The terms "inner," "outer," "upper," and "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In this application, unless otherwise expressly specified and limited, the term "attached," etc., should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can refer to a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication of two elements or the interaction relationship between two elements. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0059] Example:
[0060] A dynamic intelligent prediction method for landslide area based on temporal InSAR and multi-feature clustering, such as Figure 1 As shown, it includes the following steps:
[0061] S1: Acquire multi-phase synthetic aperture radar images, digital elevation model data and atmospheric correction data of the target area, process them using the SBAS-InSAR algorithm, construct the model inversion deformation rate, and obtain deformation rate raster data.
[0062] S1 specifically includes:
[0063] S11: Preprocess synthetic aperture radar images;
[0064] S12: Based on the condition that the time baseline does not exceed the first threshold and the spatial baseline does not exceed the second threshold, candidate interference pairs are selected and the phase difference is calculated;
[0065] S13: The minimum cost flow algorithm is used for phase unwrapping, and the terrain phase and atmospheric delay phase are eliminated by using digital elevation model and atmospheric correction data, respectively;
[0066] S14: Based on the SBAS-InSAR algorithm, phase time series analysis of multi-period interferometric pairs is performed to construct a model for inverting deformation rate and output deformation rate raster data.
[0067] The deformation rate inversion is expressed as:
[0068] ;
[0069] in, For deformation phase difference, For image time intervals, The wavelength for synthetic aperture radar images can be C-band. .
[0070] The interference phase composition is represented as follows:
[0071]
[0072] in, For deformation phase, For terrain phase, For atmospheric delayed phase, This is the noise phase.
[0073] Specifically, Sentinel-1 images of the landslide area from 2020 to 2024, as well as 12.5m resolution ALOS DEM data and Gasco atmospheric correction data covering the study area and a 1km buffer zone, were collected. Then, SAR images were radiometrically calibrated, registered, and denoised. Interferometric pairs were constructed with a temporal baseline ≤36 days and a spatial baseline ≤150m, and the phase difference was calculated. Simultaneously, a coherence assessment and removal step was added, setting a coherence coefficient ≥0.3 as an effective threshold to remove low-quality interferometric pairs with coherence below the threshold, reducing the impact of spatiotemporal incoherence and noise interference at the source. The phase was then unwrapped using a minimum cost flow algorithm, and DEM data was imported to eliminate topographic phase, while Gasco data was imported to reduce atmospheric delay phase. Finally, the SBAS-InSAR algorithm was used to perform phase time-series analysis on multi-period interferometric pairs, constructing a model to invert deformation rates and outputting raster data.
[0074] S2: Based on deformation rate raster data and digital elevation model data, extract annual deformation rate, slope attributes and spatial coordinate location information, and construct coupled features;
[0075] S2 specifically includes:
[0076] Deformation rate turning point data are obtained based on deformation rate raster data;
[0077] Slope turning point data were obtained based on digital elevation model data;
[0078] Spatially connect the deformation rate turning point data and the slope turning point data, and associate the slope attribute value with the attribute table of the deformation rate point feature;
[0079] Spatial coordinate information is added to each spatially connected point element to form a coupled feature consisting of deformation rate, slope attribute and spatial coordinates.
[0080] In practical applications, based on SBAS-InSAR technology, the deformation of each year in the study area from 2020 to 2024 is first extracted. The average annual deformation rate for each year is then calculated, and the corresponding annual deformation rate raster data is output. Based on this, the annual deformation rate raster point conversion operation for the study area is performed. Using collected 12.5m DEM data, the slope is calculated and then the raster points are converted. In ArcGIS, the deformation rate conversion data and slope conversion data for the same year are loaded and spatially linked. The slope attribute value is associated with the attribute table of the deformation rate point feature. At this point, the attribute table of the feature includes not only the annual deformation rate and slope attribute, but also its own X and Y coordinates and other location information, achieving effective coupling of the three.
[0081] S3: Standardize and reduce the dimensionality of the coupling features to obtain the optimized feature matrix;
[0082] S3 specifically includes:
[0083] The coupling features are standardized using the Z-score standardization method.
[0084] The Z-score standard representation is:
[0085] ;
[0086] in, The standardized value. These are the original eigenvalues. The mean of the original features. denoted as the standard deviation of the original features.
[0087] Principal component analysis was performed on the standardized coupling features to reduce their dimensionality, resulting in an optimized feature matrix.
[0088] Covariance matrix calculation:
[0089] ;
[0090] in, The standardized feature matrix, For the sample size, is the mean vector of the characteristic matrix.
[0091] Principal component selection: Select the top k principal components with a cumulative variance contribution rate ≥ 85%, i.e. ;in, These are the eigenvalues of the covariance matrix. This represents the original feature dimension (where m=4).
[0092] Specifically, feature coupling is optimized for data quality through feature standardization and dimensionality reduction, providing more reliable input features for subsequent DBSCAN clustering. First, Z-score standardization is performed. Since the coupled features have significant dimensional differences, directly involving them in the analysis would lead to features with larger dimensions dominating the clustering results. Therefore, the mean and standard deviation of features such as annual deformation rate, slope, X-coordinate, and Y-coordinate are calculated separately to eliminate dimensional interference and bring the features to the same order of magnitude. After standardization, principal component analysis is performed for dimensionality reduction. The core purpose is to eliminate redundant information between features, reduce computational complexity, and retain key deformation and terrain features. The originally multidimensional features are condensed into a few independent composite features, which reduces the computational burden of subsequent clustering algorithms and avoids the interference of feature multicollinearity on the clustering results. The final low-dimensional principal component feature matrix can be directly used for training the DBSCAN clustering model.
[0093] S4: The improved DBSCAN clustering algorithm is used to perform cluster analysis on the feature matrix to obtain the landslide cluster labels;
[0094] S4 specifically includes:
[0095] Optimize neighborhood radius and minimum number of points based on contour coefficient;
[0096] The profile coefficient is expressed as:
[0097] ;
[0098] in, For the sample The average distance to other samples in the same cluster, For the sample i The average distance to all samples in the nearest heterogeneous cluster.
[0099] The neighborhood is defined as:
[0100]
[0101] in, For the sample i and j Euclidean distance, is the neighborhood radius.
[0102] Dynamically adjust clustering parameters based on interannual data characteristics;
[0103] Clustering is performed on the optimized feature matrix to obtain landslide cluster labels.
[0104] S5: Filter landslide points based on cluster labels, construct Thiessen polygons and calculate the landslide area; Thiessen polygons are polygons formed by perpendicular bisectors between adjacent points, with each landslide point as the center.
[0105] Specifically, the obtained cluster labels are associated with and assigned values to the corresponding landslide point elements. Then, the cluster labels corresponding to the landslides are determined by combining contemporaneous SAR deformation imagery, high-resolution optical imagery, and field survey data. For each landslide point, a Thiessen polygon is constructed, which uses each landslide point as the center and constructs polygons enclosed by the perpendicular bisectors between adjacent points, thus transforming points into polygons.
[0106] S6: Repeat steps S1 to S5 to obtain landslide area data for multiple periods and perform time series analysis and trend prediction.
[0107] Time series analysis and trend forecasting include:
[0108] The area data of landslides from multiple periods were compiled into an area time series table;
[0109] By fitting an area change model using linear regression, and combining line graphs and fitted curves, the interannual trend is presented intuitively.
[0110] Output the interannual area change rate and goodness-of-fit index to achieve dynamic prediction of landslide area.
[0111] Specifically, for the SBAS-InSAR processing results for each year from 2020 to 2024, the process of "clustering calculation - label assignment - label judgment - surface transformation" is repeated to ensure that the processing procedure is standardized each year and that the DBSCAN parameters are dynamically adjusted according to the actual situation each year. The total area of the landslide body is summarized annually, and the area data is organized into a time series table, such as... Figure 2 As shown, the linear regression method is selected based on the data characteristics to analyze the trend, and line graphs and fitted curves are drawn to present the trend intuitively.
[0112] Therefore, this application adopts the above-mentioned method for dynamic intelligent prediction of landslide area based on temporal InSAR and multi-feature clustering. By constructing an integrated technical system of "multi-source data fusion - deep feature coupling - intelligent clustering analysis - temporal dynamic prediction", it can effectively solve the key problems in the existing technology, such as insufficient utilization of location information, lack of multi-feature spatial coupling, blind selection of clustering parameters, and lack of timeliness and continuity of monitoring results.
[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of this application, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of this application.
Claims
1. A dynamic intelligent prediction method for landslide area based on temporal InSAR and multi-feature clustering, characterized in that, Includes the following steps: S1: Acquire multi-period synthetic aperture radar images, digital elevation model data, and atmospheric correction data of the target area. Process them using the SBAS-InSAR algorithm to construct the model inversion deformation rate and obtain deformation rate raster data. Among them, a coherence coefficient threshold of ≥0.3 is set to filter out low-quality interferometric pairs and output high-precision annual deformation rate. S2: Based on deformation rate raster data and digital elevation model data, extract annual deformation rate, slope attributes and spatial coordinate location information, and construct coupled features; S2 specifically includes: Deformation rate turning point data are obtained based on deformation rate raster data; Slope turning point data were obtained based on digital elevation model data; Spatially connect the deformation rate turning point data and the slope turning point data, and associate the slope attribute value with the attribute table of the deformation rate point feature; Add spatial coordinate information to each spatially connected point element to form a coupled feature consisting of deformation rate, slope attribute and spatial coordinates; S3: Standardize and reduce the dimensionality of the coupling features to obtain the optimized feature matrix; S3 specifically includes: The coupling features are standardized using the Z-score standardization method. Principal component analysis was performed to reduce the dimensionality of the standardized coupling features, and principal components with a cumulative variance contribution rate of ≥85% were selected to obtain the optimized feature matrix. S4: An improved DBSCAN clustering algorithm is used to perform cluster analysis on the feature matrix, and a parameter time-series adaptive adjustment mechanism based on the silhouette coefficient and interannual deformation evolution characteristics is established to obtain landslide body cluster labels. S4 specifically includes: Optimize neighborhood radius and minimum number of points based on contour coefficient; Dynamically adjust clustering parameters based on interannual data characteristics; Clustering is performed on the optimized feature matrix to obtain landslide cluster labels; The profile coefficient is expressed as: ； in, For the sample The average distance to other samples in the same cluster, For the sample i The average distance to all samples in the nearest heterogeneous cluster; The neighborhood is defined as: in, For the sample i and j Euclidean distance, The neighborhood radius; S5: Filter landslide points based on cluster labels, construct Thiessen polygons with each landslide point as the center, and calculate the landslide area; S6: Repeat steps S1 to S5 to obtain landslide area data for multiple periods and perform time series analysis and trend prediction. Time series analysis and trend forecasting in S6 include: The area data of landslides from multiple periods were compiled into an area time series table; By fitting an area change model using linear regression, and combining line graphs and fitted curves, the interannual trend is presented intuitively. Output the interannual area change rate and goodness-of-fit index to achieve dynamic prediction of landslide area.
2. The method for dynamic intelligent prediction of landslide area based on temporal InSAR and multi-feature clustering as described in claim 1, characterized in that, S1 specifically includes: S11: Preprocess synthetic aperture radar images; S12: Based on the condition that the time baseline does not exceed the first threshold and the spatial baseline does not exceed the second threshold, candidate interference pairs are selected and the phase difference is calculated; S13: The minimum cost flow algorithm is used for phase unwrapping, and the terrain phase and atmospheric delay phase are eliminated by using digital elevation model and atmospheric correction data, respectively; S14: Based on the SBAS-InSAR algorithm, phase time series analysis of multi-period interferometric pairs is performed to construct a model for inverting deformation rate and output deformation rate raster data.
3. The method for dynamic intelligent prediction of landslide area based on temporal InSAR and multi-feature clustering as described in claim 2, characterized in that, The deformation rate inversion is expressed as: ； in, For deformation phase difference, For image time intervals, The wavelength of the synthetic aperture radar image; The interference phase composition is represented as follows: in, For deformation phase, For terrain phase, For atmospheric delayed phase, This is the noise phase.
4. The method for dynamic intelligent prediction of landslide area based on temporal InSAR and multi-feature clustering as described in claim 3, characterized in that, The Z-score standard representation is: ； in, The standardized value. These are the original eigenvalues. The mean of the original features. denoted as the standard deviation of the original features.
5. The method for dynamic intelligent prediction of landslide area based on temporal InSAR and multi-feature clustering as described in claim 4, characterized in that, In S5, the Tyson polygon is a polygon formed by the perpendicular bisectors between adjacent points, with each landslide point as the center.