Hot melt landslide disaster early warning system based on time sequence remote sensing and ground fusion monitoring

Abstract

The invention relates to the technical field of disaster early warning, and discloses a hot melt landslide disaster early warning system based on time sequence remote sensing and ground fusion monitoring. The system comprises a remote sensing time sequence data processing module, a field data acquisition module, a data transmission module, a data processing and fusion module, an analysis and early warning module, a database and management module and a data result visualization module. According to the hot melt landslide disaster early warning system based on time sequence remote sensing and ground fusion monitoring, the monitoring efficiency is improved, the monitoring cost is reduced, the frequency of manual field patrol can be reduced to a certain extent through remote sensing monitoring, the risk of working of workers in dangerous areas is reduced, and the working efficiency is improved. Meanwhile, timely and accurate early warning can also guarantee life and property safety of surrounding residents, dual-engine modeling of'air and space remote sensing-ground equipment 'collaborative perception and'data driving-mechanism constraint' breaks through limitation of single remote sensing or ground monitoring, and a new method is provided for early warning of hot melt slump disasters in permafrost regions.

Description

Technical Field

[0001] The present invention relates to the technical field of disaster warning, and particularly to a thermokarst landslide disaster warning system based on time-series remote sensing and ground integrated monitoring. Background Technique

[0002] Thermokarst slumping refers to a geological phenomenon in which, after the thermal balance of a slope containing thick-layered ground ice is damaged by natural forces or human activities, the soil mass slides along the thaw-freeze interface under the action of gravity. It is mainly distributed in the slope areas of high-latitude or high-altitude permafrost regions, and is directly related to the melting of ground ice. The slumped materials are often diluted by meltwater to form mudflows, and have periodic activity patterns. Usually, it starts in spring, reaches its peak in summer, and stops in autumn. Initial stage: The exposed ice layer melts, and the overlying soil layer that loses support collapses. Expansion stage: The collapsed materials slide down the slope under the action of gravity, and the new ice layer continues to melt, and the slumping range extends towards the slope top. Repeated cycle: After this process is repeated many times, a slumped body in the shape of a crescent, armchair, etc. is finally formed.

[0003] Under the existing technology, by installing real-time monitoring equipment on the slope and monitoring the minute changes of the slope body through technologies such as GPS and inclinometers for early warning, these devices can timely detect the unstable signs of the slope body, so as to give early warning. However, monitoring the slope body through technologies such as GPS and inclinometers will result in a poor scanning effect due to the influence of the equipment itself by obstacles, and the monitored area is reduced, and then there will be a phenomenon of incomplete monitoring.

[0004] Based on this, the present invention provides a thermokarst landslide disaster warning system based on time-series remote sensing and ground integrated monitoring. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a thermokarst landslide disaster warning system based on time-series remote sensing and ground integrated monitoring, which has the advantages of "cooperative perception of space-air remote sensing - ground equipment" and "data-driven - mechanism constraint" dual-engine modeling, and breaks through the limitations of single remote sensing or ground monitoring, and solves the problems raised in the background technique.

[0006] The present invention provides the following technical solutions: A thermokarst landslide disaster warning system based on time-series remote sensing and ground integrated monitoring, including a remote sensing time-series data processing module, a field data acquisition module, a data transmission module, a data processing and fusion module, an analysis and warning module, a database and management module, and a data result visualization module:

[0007] Remote sensing time-series data processing module:

[0008] (1) SAR data processing of time-series remote sensing satellites: A SAR time-series processing method SBASs-InSAR that combines small baseline subsets and optimized homogeneous pixels is proposed for permafrost regions. The SAR data of time-series remote sensing satellites are registered, cropped, and orbit-refined. A small baseline subset is constructed, and differential interferograms are generated. Through homogeneous pixel constraint, phase optimization, and multi-scatterer fusion, eigenvalue decomposition is performed on the coherence matrix, and the first principal component phase is taken as the optimized phase to suppress the influence of backscattering power imbalance. Only interferogram pairs with low noise and clear fringes are selected to calculate the temporal coherence, avoiding contamination by low-quality phases. Preferred scatterers are used to replace the original phase, and the deformation rate time series is solved;

[0009] Using N + 1 SAR image maps covering the same study area, they are sorted by time as follows:

[0010] t = [t0, t1, t2, …, t M-1 T

[0011] The N single interferogram pairs are generated from the M multi-looked differential interferogram pairs obtained by the maximum spatio-temporal baseline threshold, and M satisfies:

[0012]

[0013] Independent estimation of orbit errors is performed on each interferogram using the least squares method. For the interferogram composed of two SAR acquisitions, the minimum norm l2 of i and j and ||d ij (x, y) - R ij (x, y)||2 is used, where d ij (x, y) is the pixel at azimuth x and range y, and

[0014] R ij (x, y) = e ij ·xy + a ij ·x + b ij ·y + c ij

[0015] In the formula: a ij Corresponds to the linear error coefficient in the azimuth direction (x direction), representing the linear change trend of the orbit error in the azimuth direction; b ij Corresponds to the linear error coefficient in the range direction (y direction), reflecting the linear change characteristics of the orbit error in the range direction; c ij Is the constant term, representing the orbit error offset evenly distributed over the entire interferogram;

[0016] After the interferogram is corrected for atmospheric effects and orbit errors, the temporal surface deformation of the study area is calculated using the basic formula of SBAS-InSAR:

[0017] ​

[0018] where Δφ ij is the interferometric phase difference between two SAR images i and j, λ is the radar wavelength, and Δr ij is the distance change along the line of sight between the two images, and L ij is the δ between the two images of the vertical baseline ij is the topographic phase change caused by the satellite elevation angle and surface terrain change;

[0019] Homogeneous pixel identification and coherence matrix construction:

[0020] Divide the homogeneous region based on the average amplitude interval estimation method, and only pixels of the same type of ground objects participate in the coherence matrix calculation:

[0021] T = E{zz H}, z = [z1, z 2, …z n T

[0022] In the above formula, z i is the normalized complex data to avoid interference from heterogeneous pixels;

[0023] Phase optimization and eigenvalue decomposition, perform eigenvalue decomposition on the coherence matrix, and take the first principal component phase as the optimized phase to suppress the influence of the backscattering power imbalance:

[0024]

[0025] In the above formula, λ1 is the maximum eigenvalue and u1 is the corresponding eigenvector;

[0026] Temporal coherence screening, only select interferometric pairs with low noise and clear fringes to calculate the temporal coherence, and avoid contamination of the low-quality satellite phase:

[0027]

[0028] In the above formula, m is the number of high-quality interferometric pairs;

[0029] Scatterer classification: PS points: Screen high-coherence points through the amplitude deviation threshold (such as σ < 0.2) and the average amplitude threshold; DS points: Extract medium-coherence points based on the homogeneous pixel threshold (such as amplitude difference < 15%) and the temporal coherence threshold (x > 0.3).

[0030] Spatio-temporal fusion modeling: Use the local Delaunay triangulation to connect PS and DS points, solve the deformation rate by combining the weighted least squares method, and suppress noise using spatio-temporal filtering at the same time.

[0031] ​(2) Multi-temporal remote sensing satellite multi-spectral data: Radiometric correction, geometric correction, image enhancement, and feature extraction are performed on multi-temporal remote sensing multi-spectral images to extract boundary, area, topographic and geomorphic feature information of thermokarst slumps.

[0032] On-site data acquisition module: Displacement sensors, inclination sensors, rain gauges, and soil moisture sensors are deployed at the thermokarst slump site;

[0033] Data transmission module: With the help of 4G / 5G wireless communication technology, the data collected by displacement sensors, inclination sensors, rain gauges, and soil moisture sensors are transmitted to the data processing center;

[0034] Data processing and fusion module: Includes a remote sensing data and ground monitoring data fusion processing unit;

[0035] Analysis and early warning module: Timely send early warning information to relevant departments and personnel through multiple methods such as text messages, APP push, and broadcasts;

[0036] Database and management module: Use the database to establish a thermokarst slump disaster monitoring database, store various data such as remote sensing images, ground monitoring data, processing results, and analysis reports, facilitate data query, retrieval, and sharing, and be responsible for managing and maintaining the user permissions, device status, and data quality of the system;

[0037] Data result visualization module: Through 3D map visualization technology, visually display the monitoring data and analysis results in the form of maps, charts, and 3D models, and intuitively present the distribution and development trend of thermokarst slumps.

[0038] Preferably, the analysis and early warning module adopts a machine learning method of the K-NN algorithm. When dealing with the problem of landslide displacement prediction, the training data set T = {(X T 1, y T 1), (X T 2, y T 2), …; (X T N , y T N )} is set, where X T i is the lag term of all inducing factors with positive correlation with the landslide displacement increment in the training set, and y T i is the landslide displacement increment data in the training set; there is also a test set V = {(X V 1, y V 1), (X V 2, y V 2), …, (X V M , y VM )}, select the k points closest to the test set X from the training set according to the weighted Euclidean distance V j (only compare the X in the two geometries), and the value of y V j is the mean of the landslide displacement increments among the corresponding k points in the test set. The calculation formula of the Euclidean weighted distance is:

[0039]

[0040] In the above formula: H is the number of lag terms of the selected inducing factors

[0041] The calculation steps for establishing the landslide deformation prediction model based on the K-NN algorithm are described as follows:

[0042] The first step: Calculate the Euclidean distance between the test data X V j and each training data to form a sequence

[0043] d(X T , X V j );

[0044] The second step: Sort in ascending order;

[0045] The third step: Select the k points with the smallest distance;

[0046] The fourth step: Extract the displacement increment values of the first k points;

[0047] The fifth step: Return the mean of the k displacement increments as the output of the displacement increment.

[0048] In practice, the value of k is often determined through ten-fold cross-validation. The initial k value in this paper is [3, 4, 5, 6, 7, 8]. Through cross-validation, compare the prediction errors, obtain the optimal k value, and then know the landslide displacement increment value through the k value. According to the real-time data, predict the possibility and danger level of landslide occurrence. When the monitoring data reaches the warning threshold or the model predicts a landslide risk, send warning information to relevant departments and personnel in a timely manner through multiple methods such as text messages, APP push, and broadcasts.

[0049] Preferably, the remote sensing data and ground monitoring data fusion processing unit uses GNSS time series to constrain the SBASs-InSAR deformation rate, eliminate orbit errors and long-term baseline drifts. Establish a state space model, use the Kalman filter to fuse the remote sensing data and ground monitoring data, and continuously optimize the fusion result through prediction and update steps; use mean square error, correlation coefficient, and information entropy indicators to evaluate the accuracy, reliability, and information content of the fusion result, and construct a joint adjustment model:

[0050] X = (A T W I A + B T W G B) -1 (A T W I d I + B T W G d G )

[0051] In the above formula: A is the InSAR design matrix, B is the GNSS design matrix, and W is the weight matrix.

[0052] Preferably, the deployed displacement sensors, inclination sensors, rain gauges, and soil moisture sensors are used to obtain real-time data on the displacement, inclination angle, rainfall, and soil moisture of the thermokarst landslide, and extract corresponding characteristic information based on the monitoring content, such as extracting the deformation rate (displacement amount per unit time, reflecting the activity intensity), displacement trend (long-term cumulative direction and magnitude, judging the evolution trend), and mutation characteristics (short-term abnormal fluctuations, warning key signals); the inclination angle data includes the angle change rate (angle change amount per unit time, characterizing the internal deformation rate), critical inclination threshold (extreme angle at instability, triggering an alarm), and periodic fluctuations (seasonal changes affected by freeze-thaw, analyzing the driving mechanism); the rainfall data involves the cumulative rainfall (total amount in a time period, reflecting the cumulative impact), rainfall intensity (rainfall amount per unit time, affecting the pore water pressure), and rainfall period correlation (lag relationship with displacement, constructing a prediction model); the soil moisture data covers the moisture change trend (reflecting the influence of soil water content on stability), moisture extreme values and thresholds (critical values causing soil softening), and multi-factor coupling characteristics (synergistic relationship with displacement, characterizing the water - force coupling process).

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The thermokarst landslide disaster early warning system based on time-series remote sensing and ground integration monitoring can conduct periodic observations on large areas from a macroscopic perspective through time-series remote sensing, and timely detect potential landslide areas. Ground monitoring can conduct close-range and high-precision monitoring on key areas or identified potential hazard points to obtain specific deformation information and relevant physical parameters of the landslide body. The combination of the two can achieve comprehensive monitoring from macro to micro. Remote sensing data can provide information on topography, geomorphology, vegetation, etc., and ground monitoring can obtain data such as displacement, stress, and groundwater. After integration, the characteristics and development trends of landslide disasters can be described from multiple dimensions. The two types of monitoring data complement and verify each other, effectively reducing the errors and uncertainties of single data, making the analysis and judgment of landslide disasters more accurate. Through the analysis of time-series remote sensing data, the long-term evolution trend of the landslide body can be discovered, and ground monitoring can capture the short-term rapid changes of the landslide body in real time. After integration, the occurrence probability and danger level of the landslide can be determined more precisely. Ground monitoring equipment can transmit data in real time and issue an alarm quickly once an anomaly is detected. Remote sensing satellites observe the target area at a certain period, and can timely discover new landslide hazards or the development and changes of existing landslide bodies, providing the latest information for early warning. The combination of the two can discover landslide signs and issue an alarm in the first time. Time-series remote sensing can conduct large-area surveys to determine key monitoring areas, and then use ground monitoring for precise monitoring, which can avoid unnecessary construction of ground monitoring stations and waste of resources, improve monitoring efficiency, and reduce monitoring costs. Remote sensing monitoring can reduce the frequency of on-site manual inspections to a certain extent, reduce the risk of staff working in dangerous areas, and at the same time, timely and accurate early warning can also ensure the life and property safety of surrounding residents. The "space-air remote sensing - ground equipment" collaborative perception and the "data-driven - mechanism constraint" dual-engine modeling break through the limitations of single remote sensing or ground monitoring, providing a new method for the early warning of thermokarst landslides in permafrost regions. Description of the Drawings

[0055] Figure 1 It is a schematic diagram of the system flow of the present invention. Detailed Embodiments

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0057] Please refer to Figure 1, A thermokarst landslide disaster warning system based on temporal remote sensing and ground fusion monitoring, including a remote sensing temporal data processing module, a field data acquisition module, a data transmission module, a data processing and fusion module, an analysis and warning module, a database and management module, and a data result visualization module:

[0058] Remote sensing temporal data processing module:

[0059] (1) Temporal remote sensing satellite SAR data processing: A SAR temporal processing method SBASs-InSAR that combines small baseline subset and optimized homogeneous pixels is proposed for permafrost regions. The temporal remote sensing satellite SAR data is registered, cropped, and orbit refined, a small baseline subset is constructed, differential interferograms are generated, and through homogeneous pixel constraint, phase optimization, and multi-scatterer fusion, eigenvalue decomposition is performed on the coherence matrix, and the first principal component phase is taken as the optimized phase to suppress the influence of backscattering power imbalance. Only interferogram pairs with low noise and clear fringes are selected to calculate the temporal coherence, avoiding contamination by low-quality phases, and preferred scatterers are used to replace the original phase to solve the deformation rate time series;

[0060] If N + 1 SAR image maps covering the same study area are used, they are sorted by time as follows:

[0061] t = [t0, t1, t2,..., t M-1 T

[0062] The N single interferogram pairs are generated from M multi-looked differential interferogram pairs obtained by the maximum spatio-temporal baseline threshold, and M satisfies:

[0063]

[0064] The least squares method is used to independently estimate the orbit error on each interferogram. For the interferogram composed of two SAR acquisitions, the minimum norm l2 of i and j and ||d ij (x, y) - R ij (x, y)||2 is used, where d ij (x, y) is the pixel at distance x and azimuth y, and

[0065] R ij (x, y) = e ij ·xy + a ij ·x + b ij ·y + c ij

[0066] In the formula: a ij Corresponds to the linear error coefficient in the azimuth direction (x direction), characterizing the linear change trend of the orbit error in the azimuth direction; b ij ​The linear error coefficient corresponding to the range direction (y - direction) reflects the linear variation characteristics of the orbital error in the range direction; c ij is the constant term, representing the orbital error offset evenly distributed over the entire interferogram;

[0067] After the atmospheric correction and orbital error correction of the interferogram, the time - series surface deformation of the study area is calculated using the basic formula of SBAS - InSAR:

[0068]

[0069] where Δφ ij is the interferometric phase difference between two SAR images i and j, λ is the radar wavelength, Δr ij is the distance change along the line - of - sight direction between the two images, L ij is the vertical baseline between the two images, and δ ij is the topographic phase change caused by the satellite elevation angle and surface topography change;

[0070] Homogeneous pixel identification and coherence matrix construction:

[0071] Based on the average amplitude interval estimation method, homogeneous regions are divided, and only pixels of the same type of ground objects participate in the coherence matrix calculation:

[0072] T = E{zz H}, z = [z1, z 2, …z n T

[0073] In the above formula, z i is the normalized complex - number data to avoid the interference of heterogeneous pixels;

[0074] Phase optimization and eigenvalue decomposition: The coherence matrix is decomposed by eigenvalues, and the first principal - component phase is taken as the optimized phase to suppress the influence of the backscattering power imbalance:

[0075]

[0076] In the above formula, λ1 is the maximum eigenvalue, and u1 is the corresponding eigenvector;

[0077] Temporal coherence screening: Only select interferogram pairs with low noise and clear fringes to calculate the temporal coherence, avoiding the contamination of low - quality satellite phases:

[0078]

[0079] In the above formula, m is the number of high - quality interferogram pairs;

[0080] ​Scatterer classification: PS points: High-coherence points are screened through the amplitude deviation threshold (such as σ < 0.2) and the average amplitude threshold; DS points: Medium-coherence points are extracted based on the homogeneous pixel threshold (such as amplitude difference < 15%) and the temporal coherence threshold (x > 0.3).

[0081] Space-time fusion modeling: The local Delaunay triangulation is used to connect PS and DS points, and the weighted least squares method is combined to solve the deformation rate. At the same time, space-time filtering is used to suppress noise.

[0082] (2) Multi-temporal remote sensing satellite multispectral data: Radiometric correction, geometric correction, image enhancement, and feature extraction are performed on multi-temporal remote sensing multispectral images to extract the boundary, area, topographic and geomorphic feature information of thermokarst slumps.

[0083] Field data acquisition module: Displacement sensors, inclinometers, rain gauges, and soil moisture sensors are deployed at the thermokarst slump site;

[0084] Data transmission module: With the help of 4G / 5G wireless communication technology, the data collected by displacement sensors, inclinometers, rain gauges, and soil moisture sensors are transmitted to the data processing center;

[0085] Data processing and fusion module: Includes a remote sensing data and ground monitoring data fusion processing unit;

[0086] Analysis and early warning module: Early warning information is timely released to relevant departments and personnel through multiple methods such as text messages, APP push, and broadcasts;

[0087] Database and management module: A thermokarst slump disaster monitoring database is established using the database to store various data such as remote sensing images, ground monitoring data, processing results, and analysis reports, facilitating data query, retrieval, and sharing. It is responsible for managing and maintaining the user permissions, device status, and data quality of the system;

[0088] Data result visualization module: Through 3D map visualization technology, the monitoring data and analysis results are visually displayed in the form of maps, charts, and 3D models, intuitively presenting the distribution and development trend of thermokarst slumps.

[0089] Among them, the analysis and early warning module adopts the machine learning method of the K-NN algorithm. When dealing with the problem of landslide displacement prediction, the training data set T = {(X T 1,y T 1),(X T 2,y T 2),…;(X T N ,y T N )} is set, where X T iFor the lag terms of all inducing factors in the training set that have a positive correlation with the landslide displacement increment, y T i is the landslide displacement increment data in the training set; there is also a test set V = {(X V 1, y V 1), (X V 2, y V 2), …, (X V M , y V M )}. According to the weighted Euclidean distance, select the k points in the training set that are closest to the test set X V j (only compare the X in the two geometries), and the value of y V j is the mean value of the landslide displacement increment among the corresponding k points in the test set. The calculation formula of the Euclidean weighted distance:

[0090]

[0091] In the above formula: H is the number of lag terms of the selected inducing factors

[0092] The calculation steps for establishing a landslide deformation prediction model based on the K-NN algorithm are described as follows:

[0093] The first step: Calculate the Euclidean distance between the test data X V j and each training data to form a sequence

[0094] d(X T , X V j );

[0095] The second step: Sort in ascending order;

[0096] The third step: Select the k points with the smallest distance;

[0097] The fourth step: Extract the displacement increment values of the first k points;

[0098] The fifth step: Return the mean value of the k displacements increments as the output of the displacement increment.

[0099] In practice, the value of k is often determined through ten-fold cross-validation. The initial k value in this paper is [3, 4, 5, 6, 7, 8]. Through cross-validation, compare the prediction errors, obtain the optimal k value, and then know the landslide displacement increment value through the k value. According to the real-time data, predict the possibility and danger level of landslide occurrence. When the monitoring data reaches the warning threshold or the model predicts a landslide risk, send warning information to relevant departments and personnel in a timely manner through multiple methods such as text messages, APP push, and broadcasts.

[0100] Among them, the remote sensing data and ground monitoring data fusion processing unit uses GNSS time series to constrain the deformation rate of SBASs-InSAR, eliminating orbital errors and long-term baseline drift. A state space model is established, and the Kalman filter is used to fuse the remote sensing data and ground monitoring data. Through prediction and update steps, the fusion result is continuously optimized; the mean square error, correlation coefficient, and information entropy indexes are used to evaluate the accuracy, reliability, and information content of the fusion result, and a joint adjustment model is constructed:

[0101] X = (A T W I A + B T W G B) -1 (A T W I d I + B T W G d G )

[0102] In the above formula: A is the InSAR design matrix, B is the GNSS design matrix, and W is the weight matrix.

[0103] Among them, displacement sensors, tilt sensors, rain gauges, and soil moisture sensors are deployed to obtain real-time data on the displacement, tilt angle, rainfall, and soil moisture of the thermokarst landslide, and corresponding characteristic information is extracted according to the monitoring content. For example, the deformation rate (displacement per unit time, reflecting the activity intensity), displacement trend (long-term cumulative direction and magnitude, judging the evolution trend), and mutation characteristics (short-term abnormal fluctuations, warning key signals) are extracted; the tilt angle data includes the angle change rate (angle change per unit time, characterizing the internal deformation rate), critical tilt threshold (extreme angle at instability, triggering an alarm), and periodic fluctuations (seasonal changes affected by freeze-thaw, analyzing the driving mechanism); the rainfall data involves the cumulative rainfall (total in a time period, reflecting the cumulative impact), rainfall intensity (rainfall per unit time, affecting the pore water pressure), and rainfall period correlation (lag relationship with displacement, constructing a prediction model); the soil moisture data covers the humidity change trend (reflecting the influence of soil water content on stability), humidity extreme values and thresholds (critical values causing soil softening), and multi-factor coupling characteristics (synergistic relationship with displacement, characterizing the water-hydro-mechanical coupling process).

[0104] Through time-series remote sensing, large areas can be observed periodically from a macroscopic perspective to promptly detect potential landslide areas. Ground monitoring can conduct close-range and high-precision monitoring of key areas or identified potential hazard points to obtain specific deformation information and relevant physical parameters of the landslide body. The combination of the two can achieve comprehensive monitoring from macro to micro. Remote sensing data can provide information on topography, geomorphology, vegetation, etc., and ground monitoring can obtain data such as displacement, stress, and groundwater. After fusion, the characteristics and development trends of landslide disasters can be described from multiple dimensions. The two types of monitoring data complement and verify each other, effectively reducing the errors and uncertainties of single data and making the analysis and judgment of landslide disasters more accurate. Through the analysis of time-series remote sensing data, the long-term evolution trend of the landslide body can be discovered, and ground monitoring can capture the short-term rapid changes of the landslide body in real time. After fusion, the occurrence probability and risk level of the landslide can be determined more precisely. Ground monitoring equipment can transmit data in real time and promptly issue early warnings once anomalies are detected. Remote sensing satellites observe the target area at a certain cycle, and can promptly discover new landslide hazards or the development and changes of existing landslide bodies, providing the latest information for early warnings. The combination of the two can detect landslide signs and issue early warnings in the first place. Time-series remote sensing can conduct large-scale surveys to determine key monitoring areas, and then use ground monitoring for precise monitoring, which can avoid unnecessary construction of ground monitoring stations and waste of resources, improve monitoring efficiency, and reduce monitoring costs. Remote sensing monitoring can, to a certain extent, reduce the frequency of on-site manual inspections and reduce the risks of staff operating in dangerous areas. At the same time, timely and accurate early warnings can also ensure the life and property safety of surrounding residents. The collaborative perception of "space-air remote sensing - ground equipment" and the dual-engine modeling of "data-driven - mechanism constraint" break through the limitations of single remote sensing or ground monitoring, providing a new method for the early warning of thermokarst disasters in permafrost regions.

[0105] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0106] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The thermal melt landslide disaster early warning system based on time series remote sensing and ground fusion monitoring is characterized by: It includes remote sensing time series data processing module, field data acquisition module, data transmission module, data processing and fusion module, analysis and early warning module, database and management module and data result visualization module: Remote sensing time series data processing module: (1) Time series remote sensing satellite SAR data processing: A SAR time series processing method SBASs-InSAR combining small baseline sets and optimized homogeneous pixels is proposed for permafrost areas. Time series remote sensing satellite SAR data are registered, cropped, and orbit refined to construct a small baseline set and generate differential interferograms. The coherence matrix is ​​decomposed by eigenvalue through homogeneous pixel constraints, phase optimization, and multi-scatterer fusion. The first principal component phase is taken as the optimized phase to suppress the influence of backscattering power imbalance. Only low-noise, clear-fringe interference pairs are selected to calculate the temporal coherence to avoid low-quality phase pollution. The original phase is replaced by the optimal scatterer to solve the deformation rate time series. Using N+1 SAR images covering the same study area, the time sequence is as follows: t=[t0,t1,t2,...,t M-1 ] T N single-interference pairs are generated by M pairs of multi-view differential interferometry pairs obtained by the maximum spatiotemporal baseline threshold, and M satisfies: The orbit error is estimated independently on each interferogram using least squares. For the interferogram composed of two SAR acquisitions, the minimum norm l2 and || d ij (x, y)-R ij (x, y)||2, where d ij (x, y) is the pixel at distance x and angle y, and R ij (x,y)=e ij ·xy+a ij ·x+b ij ·y+c ij Where: a ij The linear error coefficient corresponding to the azimuth direction (x direction) represents the linear variation trend of the orbit error in the azimuth direction; b ij The linear error coefficient corresponding to the range direction (y direction) reflects the linear variation characteristics of the track error in the range direction; c ij is a constant term, representing the track error offset uniformly distributed over the entire interferogram; After completing the atmospheric correction and orbit error correction of the interferogram, the time series surface deformation of the study area is calculated using the SBAS-InSAR basic formula: where Δφ ij is the interferometric phase difference between two SAR images i and j, λ is the radar wavelength, Δr ij is the distance change between the two images along the line of sight, L ij is the vertical baseline delta between the two images ij It is the terrain phase change caused by the satellite elevation angle and the surface topography change; Homogeneous pixel identification and coherence matrix construction: Based on the average amplitude interval estimation method, the homogeneous area is divided, and only the pixels of the same type of objects are involved in the coherence matrix calculation: T=E{zz H },z=[z1,z 2, …With n ] T In the above formula, z i To normalize complex data and avoid interference from heterogeneous pixels; Phase optimization and eigenvalue decomposition, perform eigenvalue decomposition on the coherence matrix, take the first principal component phase as the optimized phase, and suppress the influence of backscatter power imbalance: In the above formula, λ1 is the maximum eigenvalue and u1 is the corresponding eigenvector; Temporal coherence screening, only selecting interference pairs with low noise and clear fringes to calculate temporal coherence, avoiding phase contamination of low-quality stars: In the above formula, m is the number of high-quality interference pairs; Scatterer classification: PS points: high coherence points are screened by amplitude deviation threshold (such as σ<0.2) and average amplitude threshold; DS points: medium coherence points are extracted based on homogeneous pixel threshold (such as amplitude difference <15%) and temporal coherence threshold (x>0.3). Spatiotemporal fusion modeling: The local Delaunay triangulation is used to connect the PS and DS points, and the weighted least squares method is combined to solve the deformation rate, while spatiotemporal filtering is used to suppress noise. (2) Multi-temporal remote sensing satellite multispectral data: Radiation correction, geometric correction, image enhancement, and feature extraction are performed on multi-temporal remote sensing multispectral images to extract information on the boundaries, area, and topographic features of thermal melt landslides. On-site data acquisition module: Deploy displacement sensors, inclination sensors, rain gauges, and soil moisture sensors at the thermal melt landslide site; Data transmission module: With the help of 4G / 5G wireless communication technology, the data collected by the displacement sensor, inclination sensor, rain gauge, and soil moisture sensor are transmitted to the data processing center; Data processing and fusion module: including remote sensing data and ground monitoring data fusion processing unit; Analysis and early warning module: timely release early warning information to relevant departments and personnel through SMS, APP push, and broadcast; Database and management module: Use the database to establish a thermal melt landslide disaster monitoring database to store remote sensing images, ground monitoring data, processing results, analysis reports and other types of data to facilitate data query, retrieval and sharing. Use it to manage and maintain the system's user permissions, equipment status, and data quality; Data result visualization module: Through three-dimensional map visualization technology, the monitoring data and analysis results are visualized in the form of maps, charts, and three-dimensional models, intuitively presenting the distribution and development trend of thermal melt landslides.

2. The thermal melt landslide disaster early warning system based on time series remote sensing and ground fusion monitoring according to claim 1 is characterized by: The analysis and warning module adopts the machine learning method of K-NN algorithm. When dealing with the landslide displacement prediction problem, the training data set T is set to {(X T 1,y T 1),(X T 2,y T 2),…;(X T N ,y T N )}, where X T i is the lag term of all the inducing factors that have a positive correlation with the displacement increment of the collapse in the training set, y T i is the incremental displacement data of the collapse in the training set; there is also a test set V = {(X V 1,y V 1),(X V 2,y V 2),…,(X V M ,y V M )}, select from the training set the same set as the test set X according to the weighted Euclidean distance V j The nearest k points (only compare the X in the two geometries), y V j The value of is the mean of the displacement increments of the corresponding k points in the test set. The calculation formula of the Euclidean weighted distance is: In the above formula: H is the number of lag terms of the selected inducing factors The calculation steps of establishing the landslide deformation prediction model based on the K-NN algorithm are described as follows: Step 1: Calculate the test data X V j The Euclidean distance between each training data forms a sequence d(X T ,X V j ); Step 2: Sort from small to large; Step 3: Select k points with the smallest distance; Step 4: Extract the displacement increment values ​​of the first k points; Step 5: Return the mean of the k displacement increments as the output of the displacement increment. In practice, the k value is often determined through ten-fold cross validation. The initial k value in this paper is [3, 4, 5, 6, 7, 8]. Through cross-validation and comparison of prediction errors, the optimal k value is obtained, and then the landslide displacement increment value is known through the k value. The possibility and degree of danger of landslide are predicted based on real-time data. When the monitoring data reaches the warning threshold or the model predicts that there is a risk of landslide, warning information is promptly issued to relevant departments and personnel through SMS, APP push, and broadcast.

3. The thermal melt landslide disaster early warning system based on time series remote sensing and ground fusion monitoring according to claim 1 is characterized by: The remote sensing data and ground monitoring data fusion processing unit uses the GNSS time series to constrain the SBASs-InSAR deformation rate and eliminate orbit errors and long-term baseline drift. A state space model is established, and the remote sensing data and ground monitoring data are fused using a Kalman filter. The fusion results are continuously optimized through prediction and update steps; the mean square error, correlation coefficient and information entropy indicators are used to evaluate the accuracy, reliability and information content of the fusion results, and a joint adjustment model is constructed: X=(A T W I A+B T W G B) -1 (A T W I d I +B T W G d G ) In the above formula: A is the InSAR design matrix, B is the GNSS design matrix, and W is the weight matrix.

4. The thermal melt landslide disaster early warning system based on time series remote sensing and ground fusion monitoring according to claim 1 is characterized by: The deployed displacement sensors, inclination sensors, rain gauges, and soil moisture sensors are used to obtain real-time data on the displacement, inclination angle, rainfall, and soil moisture of the thermal melt landslide body, and extract corresponding characteristic information according to the monitoring content, such as deformation rate (displacement per unit time, reflecting the intensity of activity), displacement trend (long-term cumulative direction and magnitude, judging the evolution trend), and mutation characteristics (abnormal fluctuations in short time, early warning key signals); The tilt angle data include the angle change rate (angle change per unit time, characterizing the internal deformation rate), critical tilt threshold (angle extreme value when unstable, triggering early warning), and periodic fluctuations (seasonal changes affected by freeze-thaw, analyzing driving mechanisms); Precipitation data involve cumulative rainfall (sum of time periods, reflecting cumulative impact), rainfall intensity (rainfall per unit time, affecting pore water pressure), and rainfall period correlation (lagged relationship with displacement, used to build a prediction model); soil moisture data cover moisture change trends (reflecting the impact of soil moisture on stability), moisture extremes and thresholds (critical values ​​cause soil softening), and multi-factor coupling characteristics (synergistic relationship with displacement, characterizing the water-mechanical coupling process).

Images