Slope tiny deformation dynamic monitoring method and system based on multi-source remote sensing data

Abstract

The invention provides a slope tiny deformation dynamic monitoring method and system based on multi-source remote sensing data, and the method comprises the steps: collecting multi-source contour data, and carrying out the fracture repair through weighted fusion and cosine similarity calculation; constructing a topological model, calculating section continuity, and performing interpolation reconstruction on a region with insufficient continuity; constructing a three-dimensional model based on the reconstruction data, calculating the gradient, curvature and roughness of a local curved surface, and smoothing a sudden change area; aligning the multi-time-point data to calculate a deformation feature vector, judging a deformation area and quantifying the deformation quantity and the rate; a support vector machine is used to classify high-risk sections, and a neural network model is combined to predict a deformation trend. According to the application, a multi-environment variable optimization model is fused, the monitoring precision is improved, and dynamic monitoring of slope tiny deformation and high-risk section early warning are realized through multi-source data collaboration and multi-step processing.

Description

Technical Field

[0001] This application relates to the field of slope deformation monitoring technology, and in particular to a method and system for dynamic monitoring of minor slope deformation based on multi-source remote sensing data. Background Technology

[0002] Slope deformation monitoring, as a crucial field in geological disaster prevention and control, bears the critical mission of ensuring the safety of infrastructure and the lives and property of people. With the acceleration of urbanization and the increasing frequency of extreme weather events, slope stability issues are becoming increasingly prominent, placing higher demands on the accuracy and real-time performance of monitoring technologies. Accurately grasping the dynamics of minute slope deformations is not only fundamental to preventing disasters such as landslides, but also an important basis for engineering design and maintenance.

[0003] Chinese Patent, Publication No. CN106441174B, Publication Date: May 28, 2019, discloses a method and system for monitoring the deformation of high slopes. The method eliminates gross errors in carrier observations, eliminates residuals using unbiased estimation, calculates the geometric distance of single differences, performs carrier single difference calculations relative to the reference station for each monitoring station, establishes the carrier single difference expressions of all synchronous observation satellites, performs difference calculations on the carrier single difference matrices of two adjacent epochs, establishes the matrix equations of the baseline vector change and the single difference receiver clock difference change between epochs, and calculates the three-dimensional deformation of the high slope using the least squares method.

[0004] The shortcomings of the above technical solutions are: GPS signals are easily blocked by vegetation on steep slopes / canyons, and a single data source cannot capture small surface deformations, making it difficult to monitor small deformations in complex terrain. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to provide a method and system for dynamic monitoring of minor deformations of slopes based on multi-source remote sensing data. This method can achieve complementarity of multi-source remote sensing data and quantify deformation characteristics, thereby enabling the capture of minor deformations on the slope surface.

[0006] To achieve the above objectives, this application adopts the following technical solution: This application provides a method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data. The method includes the following steps: S101, collect multi-source contour data of different sections of the slope; S102. After weighted fusion processing of multi-source contour data, the contour feature similarity between each section is calculated. For areas with insufficient contour feature similarity, fracture repair is performed to obtain the slope contour data after fracture repair. S103, construct a topological data model of the slope based on the slope profile data after fracture repair, and calculate the continuity between each segment in the topological data model; S104. Determine the fracture area based on the section with insufficient continuity, calculate the Euclidean distance between adjacent data points in the fracture area, and perform interpolation reconstruction on the data points with insufficient Euclidean distance to obtain the reconstructed slope profile data. S105. Based on the reconstructed slope profile data, construct a three-dimensional slope model, calculate the geometric features of each section in the three-dimensional slope model, and if the geometric features of a certain area change abruptly, smooth the area through a local weighted regression algorithm to obtain the smoothed slope profile data. S106: Align the smoothed slope profile data at different time points, calculate the deformation feature vector through the geometric features of each section, determine the deformation area based on the change range of the deformation feature vector, and calculate the deformation amount and deformation rate of the deformation area. S107 uses the support vector machine algorithm to classify the deformation area based on the deformation amount and deformation rate of the deformation area, identifies high-risk sections, and uses time series prediction method to calculate the deformation trend of the high-risk sections of the slope in the future.

[0007] As a preferred technical solution, in step S102, calculating the contour feature similarity between each segment and repairing the fracture in areas with insufficient contour feature similarity includes: extracting the main contour direction vector of each segment based on the multi-source contour data after weighted fusion processing; calculating the contour similarity between segments using a cosine similarity algorithm; and repairing the fracture by using a spline interpolation algorithm if the contour similarity between segments is less than a preset threshold.

[0008] As a preferred technical solution, step S103, the construction of the topological data model of the slope includes: gridding the data point cloud of the overall geometric features of the slope to form the topological data model of the slope; step S103, the calculation of the continuity between each segment in the topological data model includes: dividing the slope into N segments, constructing an N×N adjacency matrix M, where the corresponding matrix element in the adjacency matrix M is 1 when two segments share a boundary, otherwise the matrix element is 0; calculating the number of adjacent segments for each segment based on the adjacency matrix M; extracting the length of the shared boundary of each pair of adjacent segments; and calculating the continuity between each segment by calculating the ratio of the length of the shared boundary of the segment to the perimeter of the segment.

[0009] As a preferred technical solution, step S104, the interpolation reconstruction of data points with insufficient Euclidean distance, includes: selecting two boundary points in the fracture area, calculating the weighted average coordinates of the two boundary points as new insertion points; correcting the coordinates of the new insertion points using a Gaussian filtering method; generating supplementary data points using a non-uniform rational B-spline algorithm; and connecting the new insertion points, supplementary data points, and original data points.

[0010] As a preferred technical solution, step S105, calculating the geometric features of each section in the three-dimensional slope model, includes: based on the three-dimensional slope model, constructing a local surface centered on the target data point for each section using the moving least squares method; constructing the covariance matrix of the target data point and neighboring data points within the local surface, extracting the eigenvector corresponding to the smallest eigenvalue of the covariance matrix as the normal vector of the local surface, calculating the angle between the normal vector of the local surface and the vertical direction to obtain the slope of the local surface; solving for the curvature of the local surface using the second-order partial derivatives of the surface; extracting all data points within the local surface, fitting a local reference plane using the least squares method, calculating the distance from each data point within the local surface to the local reference plane and calculating the standard deviation to obtain the surface roughness of the local surface.

[0011] As a preferred technical solution, in step S105, if the geometric features of a certain region change abruptly, the region is smoothed by a local weighted regression algorithm, which includes: if the slope, curvature, or surface roughness of a certain region changes abruptly, the abnormal region is marked and the data point set of the abnormal region is obtained; a Gaussian kernel function is used to generate the spatial weight of each data point in the abnormal region; a quadratic regression polynomial of the abnormal region is established, and a weighted least squares solution is performed based on the spatial weight of each data point and the quadratic regression polynomial of the abnormal region to output the smoothed data point set of the abnormal region.

[0012] As a preferred technical solution, in step S106, determining the deformation region based on the change amplitude of the deformation feature vector includes: if the change amplitude of the deformation feature vector exceeds a preset threshold, then using principal component analysis algorithm to extract the main deformation direction and obtain the deformation trend distribution; based on the deformation trend distribution, using spatial interpolation algorithm to perform gridding processing on the deformation values ​​to generate a deformation distribution heat map; and based on the deformation distribution heat map, using clustering analysis algorithm to divide the deformation region.

[0013] As a preferred technical solution, in step S107, the step of using a time series prediction method to calculate the deformation trend of the high-risk section of the slope in the future includes: integrating the historical deformation feature vectors of the high-risk section; establishing a neural network deep learning model based on the historical deformation feature vectors of the high-risk section, and outputting the deformation feature vectors in the future.

[0014] As a preferred technical solution, in step S101, environmental variable data of multiple environmental factors are collected simultaneously; in step S107, the neural network deep learning model is established based on the environmental variable data of multiple environmental factors and the historical deformation feature vector of the high-risk section; the deviation between the predicted value and the measured value of the deformation feature vector is analyzed using confidence intervals; if the deviation between the predicted value and the measured value of the deformation feature vector exceeds a preset deviation, a data correction mechanism is triggered, and interpolation correction is performed by combining the historical deformation feature vector and the predicted value of the neural network deep learning model; the contribution of each factor is analyzed using the entropy method, the interference weight of environmental factors on the deformation feature vector is calculated, the environmental variable data is fused using a weighted fusion algorithm, and the neural network deep learning model is updated using the fused environmental variable data.

[0015] This application also provides a dynamic monitoring system for minor slope deformation based on multi-source remote sensing data. The system includes: a data acquisition module for acquiring multi-source contour data of different sections of the slope; and a data processing module comprising: a weighted fusion unit, a contour repair unit, and a topology modeling unit. The weighted fusion unit performs weighted fusion processing on the multi-source contour data. The contour repair unit calculates the contour feature similarity between sections, repairs fractures in areas with insufficient contour feature similarity, and obtains fracture-repaired slope contour data. The topology modeling unit constructs a topology model of the slope based on the fracture-repaired slope contour data, calculates the continuity between sections in the topology model, identifies fracture areas based on sections with insufficient continuity, calculates the Euclidean distance between adjacent data points in the fracture area, and performs interpolation reconstruction on data points with insufficient Euclidean distance to obtain the reconstructed slope contour. The system includes a data processing module; a 3D analysis module comprising a geometric modeling unit, a deformation monitoring unit, and a risk classification unit; the geometric modeling unit constructs a 3D model of the slope based on the reconstructed slope outline data, calculates the geometric features of each section in the 3D model, and if the geometric features of a certain area change abruptly, smoothing is performed on that area using a local weighted regression algorithm to obtain smoothed slope outline data; the deformation monitoring unit aligns the smoothed slope outline data at different time points and calculates deformation feature vectors based on the geometric features of each section; the risk classification unit determines the deformation area based on the change magnitude of the deformation feature vector, calculates the deformation amount and deformation rate of the deformation area, and uses a support vector machine algorithm to classify the deformation area based on the deformation amount and deformation rate to identify high-risk sections; and a prediction and early warning module, which uses a time series prediction method to calculate the deformation trend of high-risk sections of the slope over a future period.

[0016] Compared with the prior art, the beneficial effects of this application are as follows: This application overcomes the limitations of single GPS data by fusing multi-source remote sensing data, solving the problem of slopes being easily obscured by vegetation; by weighted fusion, fracture repair, topology modeling and smoothing of slope contour data, the accuracy of the three-dimensional slope model is improved, thereby improving the accuracy of monitoring minor slope deformations; and by predicting the time series of slope deformation areas, geological disaster early warning is achieved. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the dynamic monitoring method for minor slope deformation based on multi-source remote sensing data proposed in this application. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in specific embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0019] like Figure 1 As shown, this application provides a method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data, including the following steps: S101, collect multi-source contour data of different sections of the slope; S102. After weighted fusion processing of multi-source contour data, the contour feature similarity between each section is calculated. For areas with insufficient contour feature similarity, fracture repair is performed to obtain the slope contour data after fracture repair. S103, construct a topological data model of the slope based on the slope profile data after fracture repair, and calculate the continuity between each segment in the topological data model; S104. Determine the fracture area based on the section with insufficient continuity, calculate the Euclidean distance between adjacent data points in the fracture area, and perform interpolation reconstruction on the data points with insufficient Euclidean distance to obtain the reconstructed slope profile data. S105. Based on the reconstructed slope profile data, construct a three-dimensional slope model, calculate the geometric features of each section in the three-dimensional slope model, and if the geometric features of a certain area change abruptly, smooth the area through a local weighted regression algorithm to obtain the smoothed slope profile data. S106: Align the smoothed slope profile data at different time points, calculate the deformation feature vector through the geometric features of each section, determine the deformation area based on the change range of the deformation feature vector, and calculate the deformation amount and deformation rate of the deformation area. S107 uses the support vector machine algorithm to classify the deformation area based on the deformation amount and deformation rate of the deformation area, identifies high-risk sections, and uses time series prediction method to calculate the deformation trend of the high-risk sections of the slope in the future.

[0020] This application collects multi-source contour data (such as optical remote sensing, InSAR, lidar, etc.) from different sections of the slope and uses weighted fusion processing to effectively integrate the advantages of spatial resolution, temporal resolution and coverage of different data sources. This makes up for the deficiencies of a single data source in terms of information integrity, accuracy or timeliness, and provides a more comprehensive and reliable raw data foundation for subsequent analysis.

[0021] This application addresses the contour breakage problem that may occur after multi-source data fusion (such as breakage caused by gaps in coverage of different sensor data or noise). By calculating the similarity of contour features and repairing the broken areas, and combining topological data models to quantify the continuity between sections, it solves the problem of morphological misjudgment caused by data breakage in traditional methods, ensuring the integrity and structure of slope contour data, and laying the foundation for subsequent high-precision modeling.

[0022] This application performs Euclidean distance discrimination and interpolation reconstruction on data points in sections with insufficient continuity, filling local data gaps; furthermore, it uses a local weighted regression algorithm to smooth areas with abrupt changes in geometric features, effectively suppressing the interference of noise or outliers on the model, improving the geometric accuracy and stability of the slope profile, and avoiding the misleading effect of local abrupt changes on deformation analysis.

[0023] This application achieves high-sensitivity detection of minute deformations by aligning smooth contour data at multiple time points and calculating deformation feature vectors; it determines the deformation region based on the dynamic changes in deformation amount and deformation rate, and accurately identifies high-risk sections by combining support vector machine classification technology, thus overcoming the limitations of subjective deformation threshold setting and coarse risk section division in traditional methods.

[0024] This application uses time series forecasting methods to quantitatively predict the future deformation trend of high-risk sections, extending monitoring from "current status description" to "trend prediction", providing a more forward-looking decision-making basis for slope safety management (such as taking reinforcement measures in advance), and significantly improving the initiative and practicality of monitoring.

[0025] Furthermore, in step S101, the multi-source contour data of different sections of the slope includes: lidar point cloud data, ground-penetrating radar data, and image data.

[0026] In this application, a LiDAR sensor and a multispectral camera are carried on a drone. The LiDAR sensor collects lidar point cloud data (a total of 1 million points, with a point cloud density of 1,000 points per square meter), and the multispectral camera collects image data (with a resolution of 0.1 meters and a data volume of approximately 500 MB per square kilometer). The ground-penetrating radar data provides 10 reflected waves per meter.

[0027] Furthermore, to address challenges in complex terrain, such as data loss due to steep slopes, RTK-GPS assisted positioning achieves an accuracy of 0.01 meters. Combined with SLAM algorithm for point cloud registration, the error is controlled within 0.05 meters.

[0028] Furthermore, the multi-source contour data of the slope section is processed: a pre-established standardized process is used to convert the format of the collected raw data to obtain an initial dataset with a uniform format; a preset noise filtering rule is used to obtain a denoised intermediate dataset.

[0029] In this application, lidar point cloud data was converted to LAS format, image data to GeoTIFF format, and ground-penetrating radar data to CSV format. Format unification was achieved by calling the GDAL library via a Python script, with a conversion time of approximately 10 seconds per GB of data. Noise filtering for the lidar point cloud data used the Brant-Wiener filtering algorithm to remove outliers, setting a threshold of 0.2 meters to filter out 95% of noise points. Gaussian filtering with a 3x3 kernel size was used for the image data to reduce the impact of illumination variations, improving the signal-to-noise ratio to 30dB. The initially integrated dataset was stored in a PostgreSQL database, with spatial indexing using the R-tree algorithm, improving query efficiency by 50% and compressing the data volume to 70% of the original, approximately 350MB / square kilometer.

[0030] Furthermore, in step S102, the weighted fusion processing of the multi-source contour data includes: using principal component analysis (PCA) algorithm to reduce the dimensionality of the lidar point cloud data, retaining 95% of the variance, and extracting the main contour direction vector, for example, the main contour direction vector is represented as [0.85, 0.45, 0.23]; normalizing the intensity of the ground-penetrating radar reflected waves to the [0,1] interval; extracting contour lines from the image data through edge detection (this application uses the Canny algorithm, with thresholds set to 50 and 150); constructing a data fusion model, using weighted least squares method to match the multi-source data, and the weights of the multi-source contours are determined by data accuracy and reliability. In this application, the point cloud error of the lidar is 0.01 meters, the depth error of the ground penetrating radar is 0.05 meters, and the image positioning error is 0.1 meters. Therefore, the weight allocation is lidar 0.5, ground penetrating radar 0.3, and image 0.2. After fusion, a unified contour dataset is generated, and the error is reduced to 0.02 meters.

[0031] Further, in step S102, the contour feature similarity between each segment is calculated, and the break repair for areas with insufficient contour feature similarity includes: extracting the main contour direction vector of each segment based on the multi-source contour data after weighted fusion processing; calculating the contour similarity between segments cos(θ) using the cosine similarity algorithm: cos(θ)=A•B / (||A||•||B||), where A is the main contour direction vector of segment 1 and B is the main contour direction vector of segment 2; if the contour similarity between segments is less than a preset threshold, then the spline interpolation algorithm is used for break repair.

[0032] Furthermore, an association matrix for each section of the slope is constructed by using the similarity of the contour features between each section, and then saved to the database.

[0033] In this application, the slope is divided into 10 segments, each 10 meters long. The contour similarity between segments is calculated. For example, the vectors for segment 1 and segment 2 are [1, 0.5, 0.2] and [0.9, 0.4, 0.3], respectively, resulting in a contour similarity of 0.98. A 10×10 symmetrical segment association matrix is ​​generated. In this application, a preset contour similarity threshold of 0.9 is used; values ​​below this threshold are marked as breakpoints. For example, the similarity between segment 3 and segment 4 is 0.85, thus indicating a break. In this application, cubic spline interpolation is used to repair the broken contour, with control points spaced 0.5 meters apart. After interpolation, the contour smoothness is improved to 98%, and the coordinate error of the repaired points is controlled within 0.015 meters.

[0034] Furthermore, in step S103, constructing the topological data model of the slope includes: gridding the data point cloud of the overall geometric features of the slope to form the topological data model of the slope. In this application, the Delaunay triangulation algorithm is used to grid the data point cloud of the overall geometric features of the slope (containing 10,000 points with coordinate format (x, y, z), such as point A (10.5, 20.3, 5.2)).

[0035] In step S103, calculating the continuity between segments in the topological data model includes: dividing the slope into N segments, constructing an N×N adjacency matrix M, where the corresponding matrix element in adjacency matrix M is 1 when two segments share a boundary, and 0 otherwise; calculating the number of adjacent segments for each segment based on adjacency matrix M; extracting the shared boundary length for each pair of adjacent segments; and calculating the continuity between segments by the ratio of the shared boundary length to the perimeter of the segment. For example, the shared side length between segments i and j is 5.2 meters, and the total length of all boundaries is calculated to be 300.7 meters, reflecting the overall topological connectivity. If the perimeter of a segment is 15.8 meters and the shared side length is 4.5 meters, the continuity index is 4.5 / 15.8 = 0.285, which is below the threshold of 0.5 and is considered insufficiently continuous.

[0036] Furthermore, in step S104, for example, let the coordinates of adjacent data points be... , Then the Euclidean distance d between data points A and B is: ; A preset threshold for Euclidean distance is set. If the Euclidean distance is less than the preset threshold, it is determined that the Euclidean distance is insufficient.

[0037] Furthermore, in step S104, the interpolation reconstruction of data points with insufficient Euclidean distance includes: selecting two boundary points within the fracture region, calculating the weighted average coordinates of the two boundary points as new insertion points; correcting the coordinates of the new insertion points using a Gaussian filtering method; generating supplementary data points using a non-uniform rational B-spline algorithm; and connecting the new insertion points, supplementary data points, and original data points.

[0038] For example, the Euclidean distance between points A (1.0, 2.0, 3.0) and B (1.5, 2.2, 3.1) is calculated as sqrt((1.5-1.0)^2+(2.2-2.0)^2+(3.1-3.0)^2)=0.538. If the preset threshold is 0.6, this Euclidean distance is lower than the threshold, indicating insufficient continuity, and interpolation reconstruction is required.

[0039] For example, points C (1.2, 2.1, 3.05) and D (1.7, 2.3, 3.15) are used to calculate their weighted average coordinates as the new insertion point P, using the formula P = (C + D) / 2, resulting in P (1.45, 2.2, 3.1). To ensure smooth interpolation, a Gaussian smoothing filter is applied with a standard deviation σ = 0.3. The weighted distance between the new point and its neighbors is calculated, and the coordinates of P are adjusted to (1.43, 2.19, 3.09). Supplementary data points are generated using the NURBS (Non-Uniform Rational B-Spline) algorithm. Control points are set as P and its neighbors, with a weight of 1.0 and an order of 3. A continuous curve is generated, and 10 supplementary points are uniformly distributed on the curve with a spacing of approximately 0.05. By connecting the supplementary points to the original points using a triangulation algorithm (such as Delaunay triangulation), a repaired spatial continuous contour is generated. The continuity of the contour is verified by checking that the distance between all adjacent points is less than 0.6, ensuring that the repaired contour meets the geometric consistency requirements.

[0040] Further, in step S105, calculating the geometric features of each section in the three-dimensional slope model includes: based on the three-dimensional slope model, constructing a local surface centered on the target data point for each section using the moving least squares method; constructing the covariance matrix of the target data point and neighboring data points within the local surface, extracting the eigenvector corresponding to the smallest eigenvalue of the covariance matrix as the normal vector of the local surface, calculating the angle between the normal vector of the local surface and the vertical direction to obtain the slope of the local surface; solving for the curvature of the local surface using the second-order partial derivatives of the surface; extracting all data points within the local surface, fitting a local reference plane using the least squares method, calculating the distance from each data point within the local surface to the local reference plane and calculating the standard deviation to obtain the surface roughness of the local surface.

[0041] For example, a section has an average slope of 30°, a radius of curvature of 2.5 meters, and a roughness index of 0.02, indicating that the section is relatively smooth.

[0042] Furthermore, to ensure a consistent overall morphology, a global geometric consistency analysis method was employed to calculate the statistical distribution of geometric features in each section. The standard deviation was controlled within 5%, indicating a high degree of consistency in the overall morphology. Simultaneously, principal component analysis (PCA) was used to extract the principal direction vector of the slope's three-dimensional model, and the calculated principal direction deviation angle was 3°, verifying the continuity of the overall trend of the slope's three-dimensional model.

[0043] Furthermore, in step S105, if the geometric features of a certain region change abruptly, the region is smoothed using a local weighted regression algorithm, including: if the slope, curvature, or surface roughness of a certain region changes abruptly, the abnormal region is marked and the data point set of the abnormal region is obtained; a Gaussian kernel function is used to generate the spatial weight of each data point within the abnormal region; a quadratic regression polynomial for the abnormal region is established, and a weighted least squares solution is performed based on the spatial weight of each data point and the quadratic regression polynomial of the abnormal region to output the smoothed data point set of the abnormal region.

[0044] For example, if the slope of a certain section suddenly changes from 30° to 45°, the abnormal area of ​​the section is smoothed by a local weighted regression algorithm. The slope change rate after processing is controlled within 10° / meter to ensure geometric consistency.

[0045] Furthermore, in step S106, determining the deformation region based on the change amplitude of the deformation feature vector includes: if the change amplitude of the deformation feature vector exceeds a preset threshold, then the principal component analysis algorithm is used to extract the main deformation direction to obtain the deformation trend distribution; based on the deformation trend distribution, the deformation value is processed into a grid by a spatial interpolation algorithm to generate a deformation distribution heat map; based on the deformation distribution heat map, a clustering analysis algorithm is used to divide the deformation region.

[0046] In this application, the Iterative Closest Point (ICP) algorithm is used to align two consecutive scans of data. The deformation distribution heatmap has a resolution of 1 meter × 1 meter, displaying the gradient change of deformation values ​​from 0.01 meters to 0.08 meters, revealing the deformation areas. Combined with time series analysis, a linear regression algorithm is used to fit the deformation trend and calculate the deformation rate. Assuming that the deformation rate of a certain section reaches 0.002 meters / day and shows an accelerating trend, it is determined that there is a potential landslide risk in this area. If the deformation rate exceeds a preset threshold of 0.003 meters / day, an early warning system can be automatically triggered.

[0047] Furthermore, in step S107, the time series prediction method is used to calculate the deformation trend of the high-risk section of the slope in the future, including: integrating the historical deformation feature vector of the high-risk section; establishing a neural network deep learning model based on the historical deformation feature vector of the high-risk section, and outputting the deformation feature vector in the future.

[0048] Furthermore, in step S101, environmental variable data of multiple environmental factors are collected simultaneously; in step S107, a neural network deep learning model is established based on the environmental variable data of multiple environmental factors and the historical deformation feature vector of the high-risk section; the deviation between the predicted value and the measured value of the deformation feature vector is analyzed using confidence intervals; if the deviation between the predicted value and the measured value of the deformation feature vector exceeds the preset deviation, a data correction mechanism is triggered, and interpolation correction is performed by combining the historical deformation feature vector and the predicted value of the neural network deep learning model; the contribution of each factor is analyzed using the entropy method, the interference weight of environmental factors on the deformation feature vector is calculated, the environmental variable data is fused using a weighted fusion algorithm, and the neural network deep learning model is updated using the fused environmental variable data.

[0049] For example, in a bridge monitoring project, environmental variable data such as temperature, humidity, and wind speed were collected. The temperature range was -5 to 35 degrees Celsius, humidity range was 30% to 90%, and wind speed range was 0 to 15 meters per second. Simultaneously, the deformation of key bridge points was recorded, with deformation values ​​ranging from -0.5 to 0.3 millimeters. Next, using a dynamic feature vector extraction method, principal component analysis (PCA) was applied to reduce the dimensionality of the multidimensional environmental data, extracting the main feature vectors. Assuming that 85% of the variance was explained after dimensionality reduction, temperature and wind speed were identified as the main influencing factors, with weights of 0.6 and 0.3, respectively. Subsequently, to adapt to the changing environmental influences, an adaptive filtering algorithm (such as Kalman filtering) was used to smooth the data and remove noise. It was assumed that the standard deviation of the deformation data decreased from 0.1 millimeters to 0.05 millimeters after filtering, improving data stability. Then, a correlation model between the deformation feature vector and environmental variables was constructed. Using multiple regression analysis, it was calculated that for every 1 degree Celsius increase in temperature, the deformation increases by 0.02 mm, and for every 1 meter per second increase in wind speed, the deformation increases by 0.01 mm. The R-squared value of the regression model was 0.82, indicating strong explanatory power. Next, by calculating the interference weights of environmental factors on the deformation trend, the contribution of each environmental factor was analyzed using the entropy method. The interference weights for temperature, wind speed, and humidity were found to be 0.55, 0.35, and 0.1, respectively. Finally, the reliability correction requirements of the monitoring data were assessed. Using confidence interval analysis, if the deviation between the predicted and measured deformation values ​​exceeded the 95% confidence interval (e.g., a deviation greater than 0.08 mm), a data correction mechanism was triggered. Interpolation correction was performed by combining historical deformation feature vector data and model predictions. The corrected data deviation was reduced to within 0.03 mm, ensuring monitoring accuracy.

[0050] This application also provides a dynamic monitoring system for minor deformation of slopes based on multi-source remote sensing data, including: a data acquisition module, a data processing module, a three-dimensional analysis module, and a prediction and early warning module.

[0051] The data acquisition module is used to collect multi-source profile data of different sections of the slope.

[0052] The data processing module includes a weighted fusion unit, a contour repair unit, and a topology modeling unit.

[0053] The weighted fusion unit performs weighted fusion processing on multi-source contour data. The contour repair unit calculates the contour feature similarity between segments, repairs fractures in areas with insufficient contour feature similarity, and obtains fracture-repaired slope contour data. The topology modeling unit constructs a topology model of the slope based on the fracture-repaired slope contour data, calculates the continuity between segments in the topology model, identifies fracture regions based on segments with insufficient continuity, calculates the Euclidean distance between adjacent data points in the fracture region, and performs interpolation reconstruction on data points with insufficient Euclidean distance to obtain reconstructed slope contour data.

[0054] The 3D analysis module includes a geometric modeling unit, a deformation monitoring unit, and a risk classification unit.

[0055] The geometric modeling unit constructs a 3D slope model based on the reconstructed slope profile data, calculates the geometric features of each segment in the 3D model, and if the geometric features of a certain area change abruptly, a local weighted regression algorithm is used to smooth that area, obtaining smoothed slope profile data. The deformation monitoring unit aligns the smoothed slope profile data from different time points and calculates deformation feature vectors based on the geometric features of each segment. The risk classification unit determines deformation areas based on the magnitude of change in the deformation feature vectors, calculates the deformation amount and deformation rate of the deformation areas, and uses a support vector machine algorithm to classify the deformation areas based on the deformation amount and deformation rate, identifying high-risk segments.

[0056] The prediction and early warning module is used to calculate the deformation trend of high-risk sections of slopes over a future period using time series prediction methods.

[0057] It should be noted that the terms "first," "second," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, "a" or "one," and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. "A plurality" or "several" indicates at least two. Unless otherwise stated, terms such as "front," "back," "left," "right," "lower," and / or "upper" are for illustrative purposes only and are not limited to a location or spatial orientation. Terms such as "comprising" or "including" indicate that the elements or objects preceding "comprising" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0058] The singular forms “a,” “the,” and “the” used in this application specification and appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0059] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for dynamic monitoring of minute deformations of slopes based on multi-source remote sensing data, characterized in that, The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data includes the following steps: S101, collect multi-source contour data of different sections of the slope; S102. After weighted fusion processing of multi-source contour data, the contour feature similarity between each section is calculated. For areas with insufficient contour feature similarity, fracture repair is performed to obtain the slope contour data after fracture repair. S103, construct a topological data model of the slope based on the slope profile data after fracture repair, and calculate the continuity between each segment in the topological data model; S104. Determine the fracture area based on the section with insufficient continuity, calculate the Euclidean distance between adjacent data points in the fracture area, and perform interpolation reconstruction on the data points with insufficient Euclidean distance to obtain the reconstructed slope profile data. S105. Based on the reconstructed slope profile data, construct a three-dimensional slope model, calculate the geometric features of each section in the three-dimensional slope model, and if the geometric features of a certain area change abruptly, smooth the area through a local weighted regression algorithm to obtain the smoothed slope profile data. S106: Align the smoothed slope profile data at different time points, calculate the deformation feature vector through the geometric features of each section, determine the deformation area based on the change range of the deformation feature vector, and calculate the deformation amount and deformation rate of the deformation area. S107 uses the support vector machine algorithm to classify the deformation area based on the deformation amount and deformation rate of the deformation area, identifies high-risk sections, and uses time series prediction method to calculate the deformation trend of the high-risk sections of the slope in the future.

2. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 1, characterized in that, In step S102, calculating the contour feature similarity between each segment and repairing the breakage in areas with insufficient contour feature similarity includes: extracting the main contour direction vector of each segment based on the multi-source contour data after weighted fusion processing; calculating the contour similarity between segments using a cosine similarity algorithm; and repairing the breakage by using a spline interpolation algorithm if the contour similarity between segments is less than a preset threshold.

3. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 1, characterized in that, In step S103, constructing the topological data model of the slope includes: gridding the data point cloud of the overall geometric features of the slope to form the topological data model of the slope. In step S103, the calculation of the continuity between segments in the topological data model includes: dividing the slope into N segments, constructing an N×N adjacency matrix M, where the corresponding matrix element in the adjacency matrix M is 1 when two segments share a boundary, otherwise the matrix element is 0; calculating the number of adjacent segments for each segment based on the adjacency matrix M; extracting the length of the shared boundary for each pair of adjacent segments; and calculating the continuity between segments by calculating the ratio of the length of the shared boundary to the perimeter of the segment.

4. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 1, characterized in that, In step S104, the interpolation reconstruction of data points with insufficient Euclidean distance includes: selecting two boundary points in the fracture region, calculating the weighted average coordinates of the two boundary points as new insertion points; correcting the coordinates of the new insertion points using a Gaussian filtering method; generating supplementary data points using a non-uniform rational B-spline algorithm; and connecting the new insertion points, supplementary data points, and original data points.

5. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 1, characterized in that, In step S105, calculating the geometric features of each section in the three-dimensional slope model includes: based on the three-dimensional slope model, constructing a local surface centered on the target data point for each section using the moving least squares method; constructing the covariance matrix of the target data point and neighboring data points within the local surface, extracting the eigenvector corresponding to the smallest eigenvalue of the covariance matrix as the normal vector of the local surface, calculating the angle between the normal vector of the local surface and the vertical direction to obtain the slope of the local surface; solving for the curvature of the local surface using the second-order partial derivatives of the surface; extracting all data points within the local surface, fitting a local reference plane using the least squares method, calculating the distance from each data point within the local surface to the local reference plane and calculating the standard deviation to obtain the surface roughness of the local surface.

6. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 5, characterized in that, In step S105, the step of smoothing the region by means of a local weighted regression algorithm if the geometric features of a certain region change abruptly includes: if the slope, curvature or surface roughness of a certain region changes abruptly, the abnormal region is marked and the data point set of the abnormal region is obtained; the spatial weight of each data point in the abnormal region is generated by using a Gaussian kernel function; a quadratic regression polynomial of the abnormal region is established, and a weighted least squares solution is performed based on the spatial weight of each data point and the quadratic regression polynomial of the abnormal region to output the data point set of the smoothed abnormal region.

7. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 1, characterized in that, In step S106, determining the deformation region based on the change amplitude of the deformation feature vector includes: if the change amplitude of the deformation feature vector exceeds a preset threshold, then using principal component analysis algorithm to extract the main deformation direction and obtain the deformation trend distribution; based on the deformation trend distribution, using spatial interpolation algorithm to perform gridding processing on the deformation values ​​to generate a deformation distribution heat map; and based on the deformation distribution heat map, using clustering analysis algorithm to divide the deformation region.

8. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 1, characterized in that, In step S107, the step of using time series prediction to calculate the deformation trend of the high-risk section of the slope in the future includes: integrating the historical deformation feature vector of the high-risk section; establishing a neural network deep learning model based on the historical deformation feature vector of the high-risk section, and outputting the deformation feature vector in the future.

9. The method for dynamic monitoring of minor slope deformation based on multi-source remote sensing data according to claim 8, characterized in that, In step S101, environmental variable data of multiple environmental factors are collected simultaneously. In step S107, the neural network deep learning model is established based on environmental variable data of multiple environmental factors and historical deformation feature vectors of high-risk areas; the deviation between the predicted and measured values ​​of the deformation feature vectors is analyzed using confidence intervals; if the deviation exceeds a preset deviation, a data correction mechanism is triggered, and interpolation correction is performed by combining historical deformation feature vectors and the predicted values ​​of the neural network deep learning model; the contribution of each factor is analyzed using the entropy method, the interference weight of environmental factors on the deformation feature vectors is calculated, the environmental variable data is fused using a weighted fusion algorithm, and the neural network deep learning model is updated using the fused environmental variable data.

10. A dynamic monitoring system for minor deformations of slopes based on multi-source remote sensing data, characterized in that, The slope micro-deformation dynamic monitoring system based on multi-source remote sensing data includes: The data acquisition module is used to collect multi-source contour data of different sections of the slope; The data processing module includes a weighted fusion unit, a contour repair unit, and a topology modeling unit. The weighted fusion unit performs weighted fusion processing on multi-source contour data. The contour repair unit calculates the contour feature similarity between segments, repairs fractures in areas with insufficient contour feature similarity, and obtains fracture-repaired slope contour data. The topology modeling unit constructs a topology model of the slope based on the fracture-repaired slope contour data, calculates the continuity between segments in the topology model, identifies fracture areas based on segments with insufficient continuity, calculates the Euclidean distance between adjacent data points in the fracture area, and performs interpolation reconstruction on data points with insufficient Euclidean distance to obtain reconstructed slope contour data. The three-dimensional analysis module includes a geometric modeling unit, a deformation monitoring unit, and a risk classification unit. The geometric modeling unit constructs a three-dimensional slope model based on the reconstructed slope outline data, calculates the geometric features of each segment in the model, and if the geometric features of a certain area change abruptly, smooths the area using a local weighted regression algorithm to obtain smoothed slope outline data. The deformation monitoring unit aligns the smoothed slope outline data from different time points and calculates deformation feature vectors based on the geometric features of each segment. The risk classification unit determines the deformation area based on the magnitude of change in the deformation feature vector, calculates the deformation amount and deformation rate of the deformation area, and uses a support vector machine algorithm to classify the deformation area based on the deformation amount and deformation rate, identifying high-risk segments. The prediction and early warning module is used to calculate the deformation trend of high-risk sections of slopes over a future period using time series prediction methods.

Images