Landslide evolution stage early warning criterion determination method fusing multi-source heterogeneous remote sensing data

Abstract

The invention provides a landslide evolution stage early warning criterion determination method fusing multi-source heterogeneous remote sensing data, and relates to the technical field of landslide disaster monitoring early warning and landslide identification. Comprising the following steps: acquiring an SAR image and preprocessing the SAR image to obtain a preprocessed SAR image; the preprocessed SAR image is processed through the Stacking-InSAR technology and the SBAS-InSAR technology, and then earth surface deformation data are obtained; extracting landslide surface fracture characteristics to obtain surface fracture information; based on the earth surface deformation data and the earth surface fracture information, integrating multi-source data to establish a time sequence sample library; and based on the time sequence sample library, mining a deformation and fracture association relationship by using an Apriori algorithm, extracting features in combination with a Hash coding strategy, and determining identification factors in the landslide evolution process. The landslide early-stage identification efficiency can be effectively improved, and powerful support is provided for landslide disaster early-warning prevention and control.

Description

Technical Field

[0001] The invention provides a method for determining early warning criteria of a landslide evolution stage by fusing multi-source heterogeneous remote sensing data, and relates to the technical field of landslide disaster monitoring and early warning and landslide identification. Background Art

[0002] At present, the main methods for identifying landslides include total station measurement, GPS measurement, unmanned aerial vehicle (UAV) remote sensing monitoring and synthetic aperture radar interferometry (InSAR) measurement.

[0003] Among them, total station measurement can simultaneously measure angles (horizontal angles and vertical angles) and distances. In landslide deformation monitoring, by setting up stations at control points with known coordinates, monitoring points on the landslide body are observed to obtain information such as the horizontal angle, vertical angle, and slope distance of the monitoring points. Then, based on the trigonometric function relationship, the three-dimensional coordinates of the monitoring points are calculated. After multiple measurements and by comparing the coordinate changes of the monitoring points at different times, the displacement of the monitoring points in the horizontal and vertical directions can be obtained. This method can simultaneously measure horizontal and vertical displacements and fully understand the deformation state of the landslide body, but this method requires line of sight between the monitoring point and the control point, which may be limited in areas with complex terrain, dense vegetation, or obstructions from buildings.

[0004] GPS measurement uses GPS satellite signal receiving equipment to measure the three-dimensional coordinates of the monitoring point, and obtains the slope deformation by comparing the elevation data at different times. It has the characteristics of high precision and high efficiency, and can realize real-time dynamic monitoring of large areas and long distances. It is not limited by line of sight conditions and can obtain horizontal and vertical displacement information at the same time. However, GPS signals may be affected by terrain, buildings, etc., and the accuracy will decrease in some complex environments, and the equipment cost is relatively high.

[0005] Remote sensing monitoring of landslides by drones is based on aerial photogrammetry. It can obtain fine surface image data by carrying high-resolution cameras, LiDAR (Light Detection and Recognition) or other sensors to achieve long-distance non-contact detection. By utilizing the low-altitude flight capability of drones, it is possible to quickly reach the landslide area, photograph and collect data on the landslide body from different angles and heights. After processing and analysis of the collected image data and other sensor data, characteristic information of the landslide body, such as topography, surface deformation, crack distribution, etc., can be extracted to monitor the occurrence and development of landslides. When in use, drones can be quickly deployed according to monitoring needs, fly at low altitudes in complex terrain and difficult-to-reach areas, obtain high-resolution images and data, and are not restricted by ground traffic and terrain conditions. Compared with traditional aerial remote sensing and satellite remote sensing, the equipment purchase, operation and maintenance costs of drone remote sensing are relatively low, and there is no need to rent large aircraft or satellite resources, which reduces monitoring costs. However, the flight of drones is restricted by weather conditions (such as strong winds, rainfall, heavy fog, etc.) and flight airspace. At the same time, their endurance is limited, and batteries need to be replaced or refueled frequently, which affects the continuity of monitoring. Although drones can monitor within a certain range, their monitoring range is relatively small compared to satellite remote sensing. For large landslide areas, multiple flights may be required to complete the monitoring task.

[0006] InSAR uses two or more SAR images acquired by synthetic aperture radar (SAR) at different times in the same area to obtain micro-deformation information of the surface through interference processing technology. Its basic principle is based on the interference phenomenon of electromagnetic waves. When the radar wave hits the ground target and reflects back to the radar antenna, the echo signal will have a phase difference due to the different distances between the targets at different positions and the radar. By processing SAR images of different phases in the same area, the phase difference information is extracted and then converted into displacement change information of the surface. Stacking-InSAR, that is, stacked synthetic aperture radar interferometry, is based on multiple SAR image data. By stacking multiple interference pairs in the same area, the signal is enhanced and the noise is suppressed, thereby improving the monitoring accuracy and reliability. This method mainly uses long-term SAR data to fully explore the surface deformation information contained therein. By comprehensively analyzing and processing the phase information of multiple interference pairs, the phase component related to the landslide deformation is separated, and the displacement change of the landslide body is obtained. It can obtain large-scale deformation information of the surface in a short time, but in areas with complex environmental geological conditions, it is necessary to combine other means for more in-depth research. Small Baseline Set Interferometry (SBAS-InSAR) technology reduces the impact of spatiotemporal decorrelation by selecting a set of interferometric pairs of small baselines, thereby enabling monitoring of surface deformation. SBAS-InSAR technology increases the temporal sampling rate and spatial density of the observation data, and is not constrained by the linear deformation model. It can better restore the nonlinear deformation characteristics of the surface and obtain regional time series ground deformation. However, the processing flow of SBAS-InSAR is relatively complex, the amount of calculation is large, and the selection of the baseline is required to be high. If the baseline is not selected properly, errors may be introduced, affecting the accuracy of deformation monitoring. Targeting the complex scattering characteristics of landslide bodies (uneven vegetation coverage) and the difference characteristics of deformation rates (measurement accuracy of different deformation rates in the same survey area).

[0007] The integration of InSAR and UAV monitoring of landslides can give full play to the advantages of both and improve the accuracy and efficiency of landslide monitoring. Stacking-InSAR and SBAS-InSAR technologies can obtain surface deformation information in a large spatial range. They can monitor surface deformation at the millimeter level or even smaller, providing key data for early identification and dynamic monitoring of landslides. UAVs can obtain high-resolution optical images, which can clearly show the topographic details of the landslide area. By processing UAV images, the landslide boundary and surface feature information can be accurately extracted. InSAR technology is not limited by weather and lighting conditions and can be monitored all day and all weather. It can obtain continuous deformation data over a large area and can also effectively monitor the slow deformation of landslides. UAVs can take pictures of landslide areas at close range and accurately determine the boundary range of landslides. Combined with InSAR technology, the deformation area of ​​the landslide can be more accurately located and analyzed. InSAR technology and UAV monitoring technology can verify the accuracy of monitoring results. By comparing the data obtained by the two technologies, potential errors and uncertainties can be discovered, improving the reliability of monitoring results. Therefore, integrating InSAR and UAV to monitor landslide deformation has very important engineering significance.

[0008] In addition, the fused data is very broad, and finding the criteria that can really be used for early warning of landslide evolution stages is also a big challenge. Summary of the invention

[0009] The present invention provides a method for determining early warning criteria for landslide evolution stages by integrating multi-source heterogeneous remote sensing data, effectively obtaining the evolution law of landslide change identification indicators, accurately and efficiently finding the criteria for early warning of landslide evolution stages, and improving the efficiency of early identification of landslides.

[0010] To solve the above problems, the present invention provides the following technical solutions:

[0011] In a first aspect, the present invention provides a method for determining early warning criteria for landslide evolution stages by integrating multi-source heterogeneous remote sensing data, which comprises the following steps:

[0012] Acquire and preprocess the SAR image to obtain a preprocessed SAR image;

[0013] The pre-processed SAR images are processed by Stacking-InSAR technology and SBAS-InSAR technology to obtain surface deformation data;

[0014] Use UAV photogrammetry and LiDAR data to extract the surface rupture characteristics of the landslide, build a three-dimensional virtual reality model, and obtain surface rupture information;

[0015] Based on surface deformation data and surface rupture information, multi-source data are integrated to establish a time series sample library;

[0016] Based on the time series sample library, the Apriori algorithm is used to mine the correlation between deformation and rupture, and the hash coding strategy is combined to refine the features and determine the identification factors in the landslide evolution process.

[0017] Furthermore, the preprocessing method of SAR image includes:

[0018] Image registration uses the main image as the registration reference image, and registers the remaining images to the selected main image imaging space, so that the azimuth and distance offsets of the same object imaging pixels are less than 0.25 pixels.

[0019] Furthermore, the pre-processed SAR images are processed by Stacking-InSAR technology and SBAS-InSAR technology to obtain surface deformation data, including:

[0020] The interferometric stacking method is used to process the SAR image data set to obtain the average surface deformation rate of the monitoring area.

[0021] The SAR image data is processed using the permanent scatterer synthetic aperture radar interferometry method to obtain the surface deformation rate in time series.

[0022] The average surface deformation rate of the monitoring area obtained by Stacking-InSAR and SBAS-InSAR and the surface deformation rate in the time series are calculated using the weighted average method to obtain the fused deformation rate and the surface deformation data are sorted out.

[0023] Furthermore, during the Stacking-InSAR processing, the spatial baseline threshold was set to 100 meters and the temporal baseline threshold was set to 50 days.

[0024] Furthermore, the average surface deformation rate of the monitoring area and the surface deformation rate in the time series obtained by Stacking-InSAR and SBAS-InSAR are calculated using the weighted average method, including:

[0025] Calculate the average deformation rate variance of the average surface deformation rate in the monitoring area ;

[0026] Calculate the average deformation rate variance of the surface deformation rate over the time series ;

[0027] Calculated by weighted average formula:

[0028]

[0029] in and The average deformation rate variance obtained by Stacking-InSAR and SBAS-InSAR is calculated as follows:

[0030]

[0031]

[0032] Furthermore, UAV photogrammetry and LiDAR data were used to extract the surface rupture characteristics of the landslide, and a three-dimensional virtual reality model was constructed to obtain surface rupture information, including:

[0033] The digital orthophotos generated by UAV photogrammetry and the digital earth model (DEM) and digital surface model (DSM) obtained by airborne LiDAR are used to extract the detailed features of the landslide surface ruptures. The distribution pattern, boundary range, surface coordinates and three-dimensional spatial elements of the landslide cracks are extracted by using data recognition technology. A three-dimensional virtual reality model of the landslide is constructed to obtain surface rupture information including terrain morphology, structural characteristics, geological structure, crack type and extension.

[0034] The detailed features of the landslide surface rupture include ground fissures, drumlins, pull-out troughs, dislocations, and steep landslide walls.

[0035] Furthermore, based on the surface deformation data and surface rupture information, multi-source data are integrated to establish a time series sample library, including:

[0036] Arrange the surface deformation data, extract the average surface deformation rate, the surface deformation rate in time series and the surface deformation information in the geographic coordinate system obtained by processing with Stacking-InSAR and SBAS-InSAR technology;

[0037] Sorting out the surface rupture information, extracting the detailed features of the landslide surface rupture, crack distribution patterns, boundary ranges, surface coordinates and three-dimensional spatial elements;

[0038] The two types of extracted data are matched in chronological order, and the surface deformation data and surface rupture information at the same time point or in a similar time period are integrated together to obtain a time series sample library covering all surface deformation data and surface rupture information.

[0039] Furthermore, based on the time series sample library, the Apriori algorithm is used to mine the correlation between deformation and rupture, and the hash coding strategy is combined to extract features and determine the discriminative factors in the landslide evolution process, including:

[0040] Based on the time series sample library, the spatial association rule mining-Apriori algorithm is used to process and analyze the monitored deformation data to explore the correlation between the spatiotemporal evolution mechanism of landslides and the influencing factors;

[0041] The influencing factors include topographic morphology, structural characteristics, geological structure, fracture type, and expansion;

[0042] Treat each landslide monitoring point or area as a transaction. Each transaction contains the deformation rate data of the monitoring point or area and the corresponding terrain morphology, structural characteristics, geological structure, crack type, and expansion influencing factors;

[0043] Through preliminary statistical analysis of the data, we can understand the frequency of each item set in the data set, select a value and start trying to set the minimum support threshold. In all data samples, the frequency of an item set must at least reach the minimum support threshold to be considered a meaningful item set.

[0044] According to the subsequent calculation results and analysis requirements, the support threshold is gradually adjusted. According to the set minimum confidence threshold, the association rules with confidence greater than or equal to the minimum confidence are screened out.

[0045] The feature extraction method is used to extract features from the ground deformation data obtained by integrating InSAR and the surface rupture information constructed based on UAV and LiDAR. The features extracted from the ground deformation data include the changing trend of the deformation rate and the spatial distribution characteristics of the deformation area; the features extracted from the surface rupture information include the geometric characteristics of the surface cracks, the terrain slope and the slope direction;

[0046] The extracted features are hash-encoded, the high-dimensional feature vectors are mapped to the low-dimensional hash space, and the hash values ​​obtained after hash encoding of different types of data are combined to construct a joint hash code.

[0047] On the basis of joint hash coding, symbiosis analysis is carried out to explore the intrinsic connection and synergistic relationship between different data, and the similarity matrix is ​​constructed by calculating the similarity between joint hash coding vectors. Based on the similarity matrix, the data is clustered using a clustering algorithm, and samples with similar characteristics are clustered into one category to obtain the symbiosis analysis results;

[0048] Combined with the results of symbiosis analysis, collaborative representation learning is performed to obtain low-dimensional representation;

[0049] From the low-dimensional representation obtained by collaborative representation learning, features that have an important impact on landslide evolution are screened out as discriminative factors in the landslide evolution process.

[0050] Furthermore, the features that have important influence on landslide evolution are screened out, including:

[0051] The feature selection algorithm is used to screen and sort the features, and the features with importance scores exceeding the preset threshold are selected as the discriminant factors.

[0052] Furthermore, based on the time series sample library, the Apriori algorithm is used to mine the correlation between deformation and rupture, and the hash coding strategy is combined to extract features and determine the identification factors in the landslide evolution process, which also includes:

[0053] The screened discriminant factors were verified and evaluated using the cross-validation method.

[0054] Furthermore, the method further comprises:

[0055] The feature clustering algorithm and correlation analysis method are used to quantify the weight distribution of the discriminant factors.

[0056] Furthermore, feature clustering algorithm and correlation analysis method are used to quantify the weight distribution of the identification factors, including:

[0057] The data of historical landslide cases and corresponding identification factors in the study area are standardized to make the data of different indicators comparable;

[0058] Use the K-Means clustering algorithm to cluster the features of the discriminant factors, and determine the appropriate number of clusters K based on experience or the elbow method. Then, randomly initialize K cluster centers, calculate the distance from each data point to these cluster centers, and assign the data points to the category to which the closest cluster center belongs; continuously iterate and update the cluster centers until the cluster centers no longer change significantly or the preset number of iterations is reached. Cluster similar indicators into one category;

[0059] Calculate the correlation coefficient between each discriminant factor. For each pair of indicators, calculate its correlation coefficient to obtain the correlation result, identify the discriminant factor group with a correlation coefficient higher than the preset value, and comprehensively process each group of discriminant factors when allocating weights to avoid repeated calculation of weights.

[0060] The analytic hierarchy process is used to allocate weights and construct a judgment matrix. The relative importance of different identification factors in landslide disasters is determined through data analysis to form a judgment matrix. Then, the maximum eigenvalue and corresponding eigenvector of the judgment matrix are calculated, and the eigenvector is normalized to obtain the weight of each identification factor.

[0061] In the process of determining the weights, a consistency check is performed to ensure the rationality of the judgment matrix.

[0062] In a second aspect, the present invention provides a system for determining early warning criteria for landslide evolution stages by integrating multi-source heterogeneous remote sensing data, including a processor and a storage medium;

[0063] The storage medium is used to store instructions;

[0064] The processor is used to operate according to the instructions to execute the steps of the method according to the first aspect.

[0065] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0066] (1) The present invention integrates InSAR and UAV technologies, giving full play to their respective advantages, and can comprehensively and accurately monitor landslide deformation, thereby improving the accuracy and reliability of monitoring.

[0067] (2) Through the spatial association rule algorithm and data mining algorithm, the intrinsic relationship between landslide deformation and evolution law and evolution characteristics is excavated, providing a scientific basis for the early warning and prevention of landslide disasters.

[0068] (3) The present invention can predict the dynamic change law of the identification index during the landslide evolution process, which helps to timely discover the potential risks of landslide disasters and provides time guarantee for taking effective prevention and control measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 is a flow chart of the method of the present invention;

[0070] Figure 2 It is the three-stage evolution curve of landslide deformation;

[0071] Figure 3 Schematic diagram of the evolution stages of landslide disaster process. DETAILED DESCRIPTION

[0072] Embodiment 1:

[0073] This embodiment provides a method for determining early warning criteria for landslide evolution stages by integrating multi-source heterogeneous remote sensing data. Figure 1 As shown, it includes the following steps:

[0074] (1) Acquire multiple SAR image data from the Sentinel-1A satellite within the monitoring area. During image registration, select one image as the main image, and use the image registration algorithm based on feature matching to register the remaining images to the main image imaging space to ensure that the azimuth and range offsets of the same ground object imaging pixel are less than 0.25 pixels. After registration, use the image cropping tool to crop the image according to the boundary vector file of the monitoring area. Subsequently, perform orbit correction on the image using the orbit parameter file, perform radiation correction using the radiation calibration coefficient, and use the Goldstein filter to remove noise from the image;

[0075] (2) Set the spatial baseline threshold to 100 meters and the temporal baseline threshold to 50 days. Use InSAR data processing software to perform stacking interferometry on SAR images that meet the baseline threshold conditions to generate landslide differential interferometry phase maps;

[0076] (3) Perform weighted averaging on the differential interferometry phase map to obtain the average surface deformation rate in the monitoring area , and calculate the average deformation rate variance Based on the orbit information of satellite imaging and the geometric information of the reference ellipsoid, a lookup table is generated in the data processing software, and the coordinates of each pixel are calculated and converted through the lookup table to obtain the surface deformation information in the geographic coordinate system;

[0077] and It can be calculated by the following formula:

[0078]

[0079] in, is the average deformation rate; n is the number of interference patterns involved in the weighted average calculation; and are the time baseline of the i-th interference pattern and its corresponding phase value respectively;

[0080]

[0081] in, is the average deformation rate variance, n is the number of interference patterns involved in the weighted average calculation, is the velocity estimate corresponding to each interference pattern;

[0082] (4) Using SBAS-InSAR technology, data processing software is used to generate a connection map and determine the image pairs involved in the interference. The selected image pairs are subjected to interference processing to generate an interference map. Track refinement and re-leveling operations are performed to remove interference factors such as track errors and terrain phase. The surface deformation rate in the time series is obtained through the inversion algorithm. , draw the time series curve, and calculate the average deformation rate variance Based on the method in step (3), convert it into surface deformation information in the geographic coordinate system;

[0083] The calculation formula is:

[0084]

[0085] in, is the deformation rate, is the total deformation, is the total time interval;

[0086] Mean deformation rate variance The calculation method refers to the calculation formula given in step (3);

[0087] (5) The deformation rates obtained by Stacking-InSAR and SBAS-InSAR are fused by weighted average method. According to the variance of the average deformation rate obtained by the two, weighted calculation is performed according to the formula to obtain the fused deformation rate ;

[0088] The weighted average calculation formula is:

[0089]

[0090] in and The deformation rate variances obtained by Stacking-InSAR and SBAS-InSAR are calculated respectively, and the calculation formula is:

[0091]

[0092]

[0093] (6) Use unmanned aerial vehicles (UAVs) to fly and photograph the landslide area. During the flight, ensure that the entire landslide area is covered and that there is sufficient forward and lateral overlap. By setting an appropriate flight altitude, high-resolution image data can be obtained (the specific flight altitude should be dynamically adjusted according to the size of the landslide area and the complexity of the terrain). Use the LiDAR device on the aircraft to scan the landslide area and obtain high-precision three-dimensional point cloud data (during the LiDAR scanning process, it is necessary to consider parameters such as flight altitude, scanning angle, and point cloud density to ensure that the acquired data can accurately reflect the terrain characteristics of the landslide area). Then, perform geometric correction on the collected UAV images to eliminate image deformation caused by factors such as camera lens distortion and unstable flight posture. Use image matching algorithms to stitch adjacent images together to generate a complete image map of the landslide area. Use digital photogrammetry software to process the images and generate digital orthophotos (DOM). Then, filter the LiDAR point cloud data to remove noise points and non-ground points to obtain pure ground point cloud data. The digital earth model (DEM) is generated by using the filtered ground point cloud data, and the digital surface model (DSM) is generated by interpolating the LiDAR point cloud data;

[0094] According to the detailed features of the landslide surface rupture, such as ground fissures, drumlins, troughs, misalignments, and steep landslide walls, manual identification is performed on digital orthophotos and digital surface models or automatic identification is performed using image recognition algorithms. For ground fissures, they can be identified by analyzing the texture features and color changes of the image; for drumlins and misalignments, they can be identified by analyzing the elevation changes of the terrain; and for steep landslide walls, they can be identified by analyzing the edge features of the image. Once the detailed features of the landslide surface rupture are identified, data extraction technology is used to extract relevant information about these features, such as location, shape, size, elevation, etc. For ground fissures, information such as the length, width, and depth of the cracks can be extracted; for drumlins, information such as the height and diameter of the drumlins can be extracted; for misalignments, information such as the height difference of the misalignments can be extracted; for steep landslide walls, information such as the height and slope of the steep walls can be extracted.

[0095] Analyze the extracted landslide crack data and count the number, length, width and other parameters of the cracks. Use digital orthophotos and digital surface models, combined with the detailed features of the landslide surface ruptures, to determine the boundary range of the landslide (the boundary of the landslide can be determined by analyzing factors such as terrain changes and vegetation distribution). Use digital orthophotos and digital surface models to extract the coordinate information of the landslide surface, including longitude, latitude, elevation, etc. Integrate the extracted data such as the detailed features of the landslide surface ruptures, crack distribution patterns, boundary range, surface coordinates and three-dimensional spatial elements. Use three-dimensional modeling software to construct a three-dimensional virtual reality model of the landslide based on the integrated data.

[0096] (7) Integrate the surface deformation data obtained by InSAR and the surface rupture information extracted by drones to establish a time series sample library of landslide evolution characteristics, and classify and annotate the data in the sample library.

[0097] The integration methods include:

[0098] First, the surface deformation data obtained by InSAR are sorted out. These data include the average surface deformation rate obtained by Stacking-InSAR and SBAS-InSAR technology, the surface deformation rate in time series, and the surface deformation information in the geographic coordinate system. At the same time, the surface rupture information extracted by the drone is sorted out, covering the detailed characteristics of the landslide surface rupture (such as ground fissures, drumlins, etc.), crack distribution patterns, boundary ranges, surface coordinates, and three-dimensional spatial elements. Then, the above two types of data are matched in chronological order, and the surface deformation data and surface rupture information at the same time point or in a similar time period are integrated together. For example, the surface deformation rate data at a certain moment is associated with the landslide crack length, width and other data corresponding to that moment to form a record containing multiple feature information. Repeat the above process, integrate the relevant data at different times in turn, and construct a data set that can reflect the evolution characteristics of the landslide over time.

[0099] From the data structure point of view, the time series sample library is similar to a two-dimensional table, in which each row represents a record of the landslide status at a time point. Each row contains multiple columns, which store different types of characteristic data, including various types of surface deformation data obtained by InSAR (such as average deformation rate, deformation rate variance and other values), and surface rupture information extracted by drones (such as the number of cracks, length, boundary range coordinates, etc.). From the perspective of data type, there are both numerical data (such as various rates, lengths, and heights) and text or categorical data (such as crack types, etc.).

[0100] (8) Apply fusion InSAR technology to obtain the ground deformation information of the landslide, solve the deformation-time monitoring data of the study area, and determine the evolution stage of the landslide ground deformation from a time perspective;

[0101] Combine the optical image with the 3D virtual reality model in step 6 to interpret the spatial development and distribution of landslide deformation and identify the evolution characteristics of ground deformation from a spatial perspective:

[0102] By analyzing the deformation-time monitoring data, the evolution stage of the landslide ground deformation can be determined. Generally speaking, the evolution process of a landslide can be divided into the initial deformation stage, the constant-speed deformation stage, and the accelerated deformation stage. In the initial deformation stage, the deformation of the landslide body is small and the deformation rate is relatively low. As time goes by, it enters the constant-speed deformation stage, and the deformation rate remains relatively stable. When the landslide body enters the accelerated deformation stage, the deformation rate increases rapidly, indicating that the landslide may be about to fail, such as Figure 3 As shown in the figure, the three-stage evolution curve of the landslide. By combining optical images with 3D virtual reality models, the spatial development and distribution of landslide deformation can be analyzed, and the sliding direction, location of the sliding surface, distribution of deposits and other characteristics of the landslide can be identified.

[0103] from Figure 2 It can be seen that the several stages of landslide evolution usually do not have absolutely fixed thresholds, but they can be judged from the displacement rate-time curve. In the initial deformation stage, the displacement rate basically remains stable and at a low level, which is shown as a relatively gentle straight line on the displacement rate-time curve, and the slope is almost zero or very small; in the constant-speed deformation stage, it appears as a straight line with a fixed slope on the displacement rate-time curve; in the accelerated deformation stage, the displacement rate shows a rapid upward trend, which is shown as a rapid increase in the slope of the curve on the displacement rate-time curve, and the curve is obviously bent upward.

[0104] like Figure 2 The deformation values ​​corresponding to different landslides vary greatly, and there is no unified threshold value. The judgment standard is the "cumulative displacement-time curve".

[0105] (9) Use spatial association rule mining - Apriori algorithm to process and analyze the monitored deformation data, and explore the correlation between the spatiotemporal evolution of landslides and influencing factors such as terrain morphology, structural characteristics, geological structure, crack type, and expansion;

[0106] Treat each landslide monitoring point or area as a transaction. Each transaction contains the deformation rate data of the monitoring point or area and the corresponding influencing factors such as terrain morphology, structural characteristics, geological structure, crack type, expansion, etc.

[0107] The Apriori algorithm usually processes discrete data, while some features of ground deformation and surface rupture data may be continuous, such as deformation rate, crack width, etc.

[0108] According to the deformation rate obtained by fusion InSAR, it is divided into four deformation evolution stages according to the deformation rate. These four deformation stages are marked as "initial deformation stage", "constant speed deformation stage", "accelerated deformation stage" and "sliding deformation stage". Figure 3 As shown. For continuous surface rupture data such as crack width obtained by the three-dimensional virtual reality model, the equal frequency division method is adopted. First, the frequency distribution of the crack width data is counted, and the crack width is divided into three intervals: narrow, medium, and wide, so that the number of cracks in each interval is roughly the same. For categorical variables, such as crack types, the unique hot encoding method is adopted. If there are three types of cracks: tensile cracks, shear cracks, and mixed cracks, then tensile cracks can be encoded as [1, 0, 0], shear cracks are encoded as [0, 1, 0], and mixed cracks are encoded as [0, 0, 1]. Encoding is performed for the discretized variables. The four intervals of the above deformation rate can be encoded as 1, 2, 3, and 4, respectively, to facilitate subsequent algorithm calculations.

[0109] First, through preliminary statistical analysis of the data, we can understand the approximate frequency of each item set in the data set. Then, we try to set the minimum support threshold from a relatively moderate value, such as 0.1. This means that in all data samples, the frequency of an item set (such as "the deformation rate is medium-speed positive deformation and the crack width is wide") must reach at least 10% to be considered a meaningful item set.

[0110] According to the subsequent calculation results and analysis requirements, the support threshold is gradually adjusted. If the number of frequent item sets generated is too small, it may be that the support threshold is set too high and needs to be appropriately lowered; if the number of frequent item sets is too large and contains a large amount of irrelevant information, the support threshold may need to be increased. According to the set minimum confidence threshold, the association rules with confidence greater than or equal to the minimum confidence are screened out. These strong association rules are the relationship between the evolution law and evolution characteristics of the landslide and the ground deformation and surface rupture that we hope to find. The confidence threshold will change with the change of data, and specific analysis is required in specific projects.

[0111] (10) By using joint hash coding and symbiosis-privacy driven strategy, the common features and collaborative representations between data are extracted to obtain the discriminative factors in the landslide evolution process. The specific process is as follows:

[0112] Corresponding feature extraction methods are used for the ground deformation data obtained by integrating InSAR and the 3D virtual reality model data constructed based on UAV and LiDAR. For ground deformation data, the changing trend of deformation rate and the spatial distribution characteristics of deformation area can be extracted; for 3D virtual reality model data, the geometric characteristics of surface cracks (such as length, width, curvature, etc.), terrain slope and slope direction, etc. can be extracted;

[0113] The extracted features are hashed (local sensitive hashing) to map high-dimensional feature vectors to low-dimensional hash space. By constructing a hash function, feature vectors with similar distances in the original high-dimensional space are also mapped to the same or similar hash values ​​in the hash space with a high probability. The hash values ​​obtained after hash coding of different types of data are combined to construct a joint hash code. Using a simple splicing method, the hash values ​​of different data are sequentially spliced ​​into a new hash vector. Joint hash coding can integrate the feature information of multi-source data and lay the foundation for the subsequent extraction of common features and collaborative representation.

[0114] Based on the joint hash coding, symbiosis analysis is performed. Symbiosis analysis aims to explore the intrinsic connections and synergistic relationships between different data. By calculating the similarity between the joint hash coding vectors (such as cosine similarity, Euclidean distance, etc.), a similarity matrix is ​​constructed. Based on the similarity matrix, a clustering algorithm is used to cluster the data, and samples with similar characteristics are clustered into one category. In the clustering process, the data features of different categories are analyzed to discover the common features and synergistic patterns between the data.

[0115] In the process of data processing and analysis, the privacy protection of data is considered. A differential privacy-driven strategy is adopted to perturb the data by adding a certain amount of noise to the data, so that attackers cannot accurately infer the original sensitive information from the processed data. When applying differential privacy, it is necessary to reasonably set the privacy budget parameters according to the sensitivity and privacy requirements of the data, and balance the relationship between privacy protection and data availability.

[0116] Combined with the results of symbiosis analysis and privacy protection, collaborative representation learning is performed. By building a collaborative representation model, the collaborative representation method between different data is learned, so that the model can better capture the common characteristics and collaborative relationships between data. The collaborative representation model is built based on a deep learning framework, using model structures such as autoencoders and generative adversarial networks. The low-dimensional representation of data is learned through training models. In this low-dimensional representation, the collaborative relationship between different data is effectively reflected.

[0117] From the low-dimensional representation obtained by collaborative representation learning, the features that have an important impact on the landslide evolution are screened out, and these features are the discriminant factors in the landslide evolution process. The feature selection algorithm can be used to screen and sort the features, and the features with higher importance scores are selected as the discriminant factors.

[0118] The features that have an important impact on landslide evolution are selected from the low-dimensional representation obtained by collaborative representation learning, generally covering deformation rate, cumulative displacement, acceleration, feature identification index, etc. These features play a key role in the landslide evolution process and can provide an important basis for landslide monitoring, early warning and prevention.

[0119] Through analysis, it is found that the changing rate of deformation velocity, density and length of surface cracks are of high importance in the landslide evolution process and can be used as identification factors.

[0120] The selected discriminant factors were verified to ensure that they had a practical correlation with the landslide evolution process. The cross-validation method was used to evaluate the discriminant factors. The data was divided into a training set and a test set, the model was trained on the training set, and the performance of the model was verified on the test set to observe the impact of the discriminant factors on the accuracy and reliability of the model in predicting the landslide evolution state.

[0121] (11) Using feature clustering algorithm and correlation analysis method, quantify the weight distribution and sensitivity of disaster identification indicators:

[0122] Based on the historical landslide cases in the study area and the corresponding detailed data such as geological environment indicators, topographic and geomorphological indicators, lithology indicators, rock development indicators, rock weathering degree indicators, structural surface characteristics, type indicators, slope morphology indicators, etc., standardized processing is carried out to make the data of different indicators comparable.

[0123] Use the K-Means clustering algorithm to cluster the features. First, determine the appropriate number of clusters K based on experience or the elbow method. Then, randomly initialize K cluster centers, calculate the distance (such as Euclidean distance) from each data point to these cluster centers, and assign the data points to the category to which the nearest cluster center belongs. Continuously iterate and update the cluster center until the cluster center no longer changes significantly or reaches the preset number of iterations. Through cluster analysis, similar indicators are clustered into one category to facilitate subsequent analysis of the impact of different types of indicators on disaster identification. Calculate the correlation coefficient (Pearson correlation coefficient) between each indicator. For each pair of indicators, the correlation coefficient is calculated by the formula. The closer the absolute value of the coefficient is to 1, the stronger the correlation between the two indicators. Analyze the correlation results and identify highly correlated indicator groups. When assigning weights, consider comprehensive processing of these related indicators to avoid repeated calculation of weights.

[0124] The analytic hierarchy process (AHP) was used for weight allocation. First, a judgment matrix was constructed. The relative importance of different indicators in landslide disasters was determined through data analysis to form a judgment matrix. Then, the maximum eigenvalue and corresponding eigenvector of the judgment matrix were calculated, and the eigenvector was normalized to obtain the weight of each indicator. In the process of determining the weight, a consistency test is required to ensure the rationality of the judgment matrix. For the assessment of sensitivity, it can be determined by analyzing the size of the indicator weight and the degree of influence of the indicator change on the disaster identification results in different landslide cases. Indicators with large weights and significant changes in the results can be regarded as indicators with high sensitivity.

[0125] The multi-index, multi-weighted multi-layer nested ISODATA clustering method is used to reveal the changing rules of the discriminant indicators during the landslide disaster process: by determining the key parameters of the ISODATA clustering algorithm (the number of initial cluster centers, the expected range of cluster numbers, and the thresholds for splitting and merging). The setting of these parameters needs to be adjusted in combination with the characteristics of the study area and the data distribution. In areas with complex terrain and diverse landslide types, it may be necessary to set a larger number of initial cluster centers and a wider range of expected cluster numbers to ensure that the changing rules of different types of landslide discriminant indicators can be captured. Then, the data is layered according to the nature and hierarchical relationship of the indicators. First, the geological environment indicators are taken as one layer, and the topographic and geomorphic indicators are taken as another layer. The ISODATA clustering algorithm is applied to each layer for preliminary clustering to obtain the clustering results of each layer. Then, the clustering results of different levels are combined and nested to form a multi-layer nested clustering structure. In the nesting process, it is necessary to consider the association and interaction between the clustering results of different levels. By defining appropriate connection rules and fusion methods, the information of each layer is integrated to more comprehensively reveal the changing rules of the discriminant indicators during the landslide disaster process.

[0126] (12) Taking advantage of the inherent correlation of multi-source data, the surface deformation rate, crack distribution characteristics, terrain slope changes, etc. are taken as key dimensions. By establishing a mathematical model or mapping function, the multi-source data are mapped into a unified multi-dimensional space to form a multi-dimensional common coupling space shared at the bottom. During the construction process, principal component analysis (PCA) can be used for dimensionality reduction and feature extraction to reduce the redundancy and complexity of the data while retaining information that is important for identifying landslide disasters.

[0127] The core goal of this method is not simply to obtain features related to landslide disaster identification, but to predict the evolution law of landslide disaster identification indicators, and then serve the early identification of landslides and disaster warning and prevention. However, in the process of achieving this goal, obtaining relevant features is an important intermediate link.

[0128] Design a multi-task learning model architecture and define multiple tasks related to landslide disaster identification, such as landslide possibility prediction, disaster time prediction, landslide scale prediction, etc. In the model training process, the correlation and shared information between different tasks are used to learn the model parameters of multiple tasks at the same time, so as to realize the migration and sharing of data between different tasks and fields. Adopt the structure of sharing the underlying neural network layer, set specific output layers and loss functions on the branches of different tasks, and optimize the loss functions of multiple tasks at the same time through the back propagation algorithm, so that the model can learn the commonalities and differences of multi-source data in different landslide disaster identification tasks, and improve the generalization ability and prediction accuracy of the model.

[0129] Designing a multi-task learning model architecture and defining multiple tasks related to landslide disaster identification does not result in multiple models, but in a unified model in which different tasks share part of the network structure, especially the underlying neural network layer.

[0130] The purpose is to optimize the loss functions of multiple tasks simultaneously through the back propagation algorithm during model training. This joint training method enables the models to promote each other when learning different tasks, and utilizes the correlation and shared information between tasks to improve the generalization ability and prediction accuracy of the model.

[0131] For example, the information learned about topography and deformation characteristics when predicting the possibility of landslides can also help the model better predict the time of disaster and the scale of landslides.

[0132] On the basis of multi-task learning, a contrastive learning mechanism is introduced. By constructing positive sample pairs and negative sample pairs, the model can learn the differences between different landslide states and data features. Data from the same landslide area at different times but with similar disaster trends are used as positive sample pairs, and data from different landslide areas or with significantly different disaster characteristics are used as negative sample pairs. During the model training process, by optimizing the contrastive learning loss function, the model can better distinguish different landslide situations, further optimize the model parameters, and improve the model's data mining ability for landslide disaster identification indicators and the accuracy of predicting the dynamic change law of identification indicators during the landslide evolution process. During the model training process, it is necessary to reasonably set hyperparameters such as learning rate, batch size, and number of training rounds, and use the validation set to monitor and adjust the model performance to prevent overfitting. Through continuous experiments and optimization, the characteristic identification indicators that affect the instability and destruction of landslides in the disaster identification indicators are separated, the deformation evolution stage of the landslide is divided, and the thresholds of each evolution stage are extracted as early warning criteria for dividing the evolution stage in the landslide disaster process.

[0133] Embodiment 2:

[0134] This embodiment provides a system for determining early warning criteria for landslide evolution stages by integrating multi-source heterogeneous remote sensing data, including a processor and a storage medium;

[0135] The storage medium is used to store instructions;

[0136] The processor is used to operate according to the instructions to execute the steps of the method according to embodiment 1.

[0137] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for determining early warning criteria for landslide evolution stages by integrating multi-source heterogeneous remote sensing data, characterized in that: It includes the following steps: Acquire and preprocess the SAR image to obtain a preprocessed SAR image; The pre-processed SAR images are processed by Stacking-InSAR technology and SBAS-InSAR technology to obtain surface deformation data; Use UAV photogrammetry and LiDAR data to extract the surface rupture characteristics of the landslide, build a three-dimensional virtual reality model, and obtain surface rupture information; Based on surface deformation data and surface rupture information, multi-source data are integrated to establish a time series sample library; Based on the time series sample library, the Apriori algorithm is used to mine the correlation between deformation and rupture, and the hash coding strategy is combined to refine the features and determine the identification factors in the landslide evolution process.

2. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 1 is characterized in that: The preprocessing methods of SAR images include: Image registration uses the main image as the registration reference image, and registers the remaining images to the selected main image imaging space, so that the azimuth and distance offsets of the same object imaging pixels are less than 0.25 pixels.

3. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 1 is characterized in that: The pre-processed SAR images are processed by Stacking-InSAR technology and SBAS-InSAR technology to obtain surface deformation data, including: The Stacking-InSAR technology is used to process the SAR image data set to obtain the average surface deformation rate of the monitoring area; The SAR image data was processed using SBAS-InSAR technology to obtain the surface deformation rate in time series. The average surface deformation rate of the monitored area and the surface deformation rate in the time series are calculated using the weighted average method to obtain the fused deformation rate and the surface deformation data are sorted out.

4. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 3 is characterized in that: The average surface deformation rate of the monitoring area and the surface deformation rate in the time series are calculated using the weighted average method, including: Calculate the average deformation rate variance of the average surface deformation rate in the monitoring area ; Calculate the average deformation rate variance of the surface deformation rate over the time series ; Calculated by weighted average formula: ； in, is the deformation rate after fusion, To monitor the average deformation rate of the regional surface, is the surface deformation rate in the time series, and are the average deformation rate variances obtained by Stacking-InSAR and SBAS-InSAR, respectively, and the calculation formula is: ； 。 5. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 1 is characterized in that: UAV photogrammetry and LiDAR data are used to extract the surface rupture characteristics of the landslide, build a three-dimensional virtual reality model, and obtain surface rupture information, including: The digital orthophotos generated by UAV photogrammetry and the digital ground model and digital surface model obtained by airborne LiDAR are used to extract the detailed features of the landslide surface ruptures. The distribution pattern, boundary range, surface coordinates and three-dimensional spatial elements of the landslide cracks are extracted by using data recognition technology. A three-dimensional virtual reality model of the landslide is constructed to obtain surface rupture information including terrain morphology, structural characteristics, geological structure, crack type, and extension status. The detailed features of the landslide surface rupture include ground fissures, drumlins, pull-out troughs, dislocations, and steep landslide walls.

6. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 1 is characterized in that: Based on surface deformation data and surface rupture information, a time series sample library is established by integrating multi-source data, including: Arrange the surface deformation data to extract the average surface deformation rate, the surface deformation rate in time series and the surface deformation information in the geographic coordinate system; Sorting out the surface rupture information, extracting the detailed features of the landslide surface rupture, crack distribution patterns, boundary ranges, surface coordinates and three-dimensional spatial elements; The two types of extracted data are matched in chronological order, and the surface deformation data and surface rupture information at the same time point or in a similar time period are integrated together to obtain a time series sample library covering all surface deformation data and surface rupture information.

7. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 1 is characterized in that: Based on the time series sample library, the Apriori algorithm is used to mine the correlation between deformation and rupture, and the hash coding strategy is combined to extract features and determine the identification factors in the landslide evolution process, including: Based on the time series sample library, the spatial association rule mining-Apriori algorithm is used to process and analyze the monitored deformation data to explore the correlation between the spatiotemporal evolution mechanism of landslides and the influencing factors; The influencing factors include topographic morphology, structural characteristics, geological structure, fracture type, and expansion; Each landslide monitoring point or area is regarded as a transaction; each transaction contains the deformation rate data of the monitoring point or area and the corresponding terrain morphology, structural characteristics, geological structure, crack type, and data on factors affecting the expansion; Through preliminary statistical analysis of the data, we can understand the frequency of each item set in the data set, and select a value to set the minimum support threshold. In all data samples, the frequency of an item set must at least reach the minimum support threshold to be considered as a meaningful item set; Gradually adjust the support threshold; according to the set minimum confidence threshold, filter out the association rules whose confidence is greater than or equal to the minimum confidence, and obtain the processed item set; For the surface deformation data and surface rupture information in the processed item set, feature extraction methods are used to extract features. The features extracted from the ground deformation data include the changing trend of the deformation rate and the spatial distribution characteristics of the deformation area; the features extracted from the surface rupture information include the geometric characteristics of the surface cracks, the terrain slope and the slope direction; Hash code the extracted features, map the high-dimensional feature vectors to the low-dimensional hash space, combine the hash values ​​obtained after hash coding of different types of data, and construct a joint hash code; On the basis of joint hash coding, symbiosis analysis is carried out to explore the intrinsic connection and synergistic relationship between different data. By calculating the similarity between joint hash coding vectors, a similarity matrix is ​​constructed. Based on the similarity matrix, a clustering algorithm is used to cluster the data, and samples with similar characteristics are clustered into one category to obtain the results of symbiosis analysis. Combined with the results of symbiosis analysis, collaborative representation learning is performed to obtain low-dimensional representation; From the low-dimensional representation obtained by collaborative representation learning, features that have an important impact on landslide evolution are screened out as discriminative factors in the landslide evolution process.

8. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 7 is characterized in that: Screen out the features that have important influence on landslide evolution, including: The feature selection algorithm is used to screen and sort the features, and the features with importance scores exceeding the preset threshold are selected as the discriminant factors.

9. The method for determining landslide evolution stage early warning criteria by integrating multi-source heterogeneous remote sensing data according to claim 1, characterized in that: The method further comprises: The feature clustering algorithm and correlation analysis method are used to quantify the weight distribution of the discriminant factors.

10. A landslide evolution stage early warning criterion determination system integrating multi-source heterogeneous remote sensing data, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-9.

Images