A method and system for monitoring local deformation of a high slope in a complex mountainous area

Abstract

The application provides a complex mountain high slope local deformation monitoring method and system, and relates to the technical field of slope deformation monitoring. A deformation rate map is generated through satellite data, crack features are extracted in combination with unmanned aerial vehicle LiDAR point clouds, and ground sensor displacement data is combined to construct a multi-scale monitoring system from macro trends to micro details. Based on a mixed model of LSTM and a random forest, environmental variables and crack expansion rates are fused to realize landslide probability prediction and dynamic early warning. Through GIS layer superposition and Otsu algorithm automatic identification of high-risk areas, the technical scheme can realize early warning of landslide risks, and provides high-precision, high-efficiency and strong adaptability technical support for mountain geological disaster prevention.

Description

Technical Field

[0001] This invention relates to the field of slope deformation monitoring technology, and in particular to a method and system for monitoring local deformation of high slopes in complex mountainous areas. Background Technology

[0002] Currently, monitoring of high slopes in complex mountainous areas faces three major technical bottlenecks: First, insufficient accuracy of multi-source data fusion. Traditional methods rely on single satellite remote sensing (such as InSAR deformation monitoring with an accuracy of only 5-10 mm) or ground sensors, making it difficult to capture millimeter-level micro-deformations, and sensors are prone to failure under heavy rain / freeze-thaw conditions; Second, delayed dynamic early warning response. Existing technologies cannot correlate crack propagation with environmental variables (such as rainfall intensity and soil saturation) in real time, and early warning models are mostly based on static thresholds, ignoring the dynamic impact of minute-level meteorological changes on slope stability; Third, blind spots in local risk identification. Conventional GIS risk layers have low spatial resolution (≥1m), failing to accurately locate surface cracks (width ≤0.5mm) in unstable rock masses, resulting in a missed detection rate of over 30% in high-risk areas. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring local deformation of high slopes in complex mountainous areas, which can realize early warning of landslide risks and provide high-precision, high-efficiency and highly adaptable technical support for the prevention and control of geological disasters in mountainous areas.

[0004] This application proposes a method for monitoring local deformation of high slopes in complex mountainous areas, the method comprising:

[0005] S1: Determine the first deformation trend map based on the first satellite data, and perform the first risk analysis on the first deformation trend map to obtain a list of high-risk area coordinates;

[0006] S2: Based on the list of high-risk area coordinates, the first UAV point cloud data, and the ground sensor data, obtain the first crack propagation prediction result;

[0007] S3: Collect the first meteorological and environmental data and determine the environmental variable matrix;

[0008] S4: Based on the first crack propagation prediction result and the environmental variable matrix, a dynamic early warning map of landslide risk is determined.

[0009] Preferably, S1 includes:

[0010] S11: Collect first satellite data and perform first analysis and processing operations to obtain the first deformation rate time series;

[0011] S12: Train the deformation trend map based on the first deformation rate time series to determine the model, and output the first deformation trend map;

[0012] S13: Add a first risk layer to the first GIS map based on the first deformation trend map to obtain a deformation risk GIS map, and output a list of high-risk area coordinates based on the deformation risk GIS map.

[0013] Preferably, the first risk analysis and processing refers to overlaying the first deformation trend map onto the first satellite data to obtain a deformation risk GIS map that can output the coordinate list of the high-risk areas.

[0014] Preferably, S13 includes:

[0015] The first deformation trend map output by the deformation trend map determination model is imported into the first GIS map, and a first risk layer is added to the first GIS map to obtain a deformation risk GIS map.

[0016] For the aforementioned deformation risk GIS map, the Otsu algorithm is used to automatically segment high-risk areas to obtain a list of coordinates for the high-risk areas.

[0017] Preferably, S2 includes:

[0018] S21: Collect the first UAV point cloud data according to the high-risk area coordinate list, and perform first deformation rate registration processing on the first UAV point cloud data to obtain the first crack feature map.

[0019] S22: Perform the first fusion processing on the ground sensor data to obtain the ground sensor displacement feature map;

[0020] S23: Input the first crack feature map and the ground sensor displacement feature map into the crack fusion prediction model to output the first crack propagation prediction result.

[0021] Preferably, S21 includes:

[0022] S211: Using the list of high-risk area coordinates as reference points, collect point cloud data of the first UAV;

[0023] S212: Extract at least one first crack information from the first UAV point cloud data;

[0024] S213: Perform first deformation rate registration processing on at least one of the first crack information and the deformation risk GIS map one by one to obtain a first crack feature map.

[0025] Preferably, the first crack feature map includes the crack location, width, and propagation direction.

[0026] Preferably, the ground sensor data includes first GNSS data and first crack meter data.

[0027] Preferably, S4 includes:

[0028] S41: Input the crack propagation prediction result and the environmental variable matrix into the landslide probability prediction model to obtain at least one first landslide probability value;

[0029] S42: Generate a dynamic early warning map of landslide risk based on at least one of the landslide probability values ​​and the first crack feature map.

[0030] This application also proposes a local deformation monitoring system for complex mountain high slopes, which is used to implement the aforementioned method for monitoring local deformation of complex mountain high slopes.

[0031] This application proposes a method and system for monitoring local deformation of high slopes in complex mountainous areas, relating to the field of intelligent port monitoring technology. This invention achieves closed-loop management of the entire process from data acquisition to risk warning by constructing a high-precision digital twin model and using multi-source data fusion technology. First, the integration of virtual-real data from BIM models, real-time sensor data, and GIS environmental data can accurately simulate the impact of external factors such as tides and wind speed on the structure, improving prediction accuracy. Second, the AI-driven dynamic prediction model, combined with expert knowledge base verification, significantly improves the accuracy of risk warning. This method provides a complete intelligent and digital solution for port safety management, with broad application value and promising prospects. Attached Figure Description

[0032] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0033] Figure 1 This is an execution flowchart of a method for monitoring local deformation of high slopes in complex mountainous areas according to the present invention.

[0034] Figure 2 This is a schematic diagram illustrating the integration of various data types in the technical solution of this application. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0037] The following is a detailed description of a method and system for monitoring local deformation of high slopes in complex mountainous areas, based on the present invention.

[0038] This embodiment proposes a method for monitoring local deformation of high slopes in complex mountainous areas. The specific process is as follows: Figure 1 As shown.

[0039] S1: Determine the first deformation trend map based on the first satellite data, and perform the first risk analysis on the first deformation trend map to obtain a list of high-risk area coordinates.

[0040] In this step, it is necessary to obtain large-scale surface deformation trends through satellite remote sensing data to provide a benchmark for subsequent meso- / micro-level data, and generate high-precision deformation rate maps to support the initial screening of landslide risks.

[0041] The first risk analysis and processing refers to overlaying the first deformation trend map onto the first satellite data to obtain a deformation risk GIS map that can output the coordinate list of the high-risk areas.

[0042] S1 includes the following sub-steps:

[0043] S11: Collect first satellite data and perform first analysis and processing operations to obtain the first deformation rate time series.

[0044] In this step, InSAR data needs to be collected and preprocessed to obtain the deformation rate within a specified time interval, in order to prepare for subsequent model training.

[0045] The source of the first satellite data in this step is Sentinel-1 SAR satellite data. Exemplary data sampling parameters could be: a revisit period of 6 days, a resolution of 5 meters, and polarization: dual polarization HH+HV.

[0046] The processing flow of the first analysis and processing operation is as follows:

[0047] 1. For the first satellite data, the SBAS-InSAR algorithm (Small Baseline Subset) is used to generate a regional deformation rate map. The baseline threshold is set to ΔB<100m and the time baseline threshold is set to ΔT<120 days to ensure the stability of the time series.

[0048] 2. For the aforementioned regional deformation rate map, denoising is performed using Discrete Wavelet Transform (DWT). An exemplary parameter for the Discrete Wavelet Transform is: selecting the db4 wavelet basis function, a decomposition scale of 5 levels, and an improvement in signal-to-noise ratio of ≥30dB.

[0049] 3. Use wavelet packet decomposition (WPD) to extract multi-scale deformation features, and retain the low-frequency components of scales 3-5 (corresponding to the principal components of deformation).

[0050] The low-frequency component obtained from the processing is used as the output, namely the first deformation rate time series.

[0051] S12: Based on the first deformation rate time series, train to obtain the deformation trend map to determine the model, and output the first deformation trend map.

[0052] In this step, the first deformation rate time series is used to train a deformation trend map determination model, thereby outputting the first deformation trend map for use in subsequent steps.

[0053] The deformation trend diagram determination model is obtained through training in the following manner:

[0054] The model design details are as follows:

[0055] Input: The first deformation rate time series (24 samples in total, with each time step lasting 6 days over the past 12 months).

[0056] Model structure: LSTM network (3 layers, 128 units per layer, Dropout=0.2, activation function tanh), outputting the predicted deformation rate for the next 30 days (0-10 mm / year).

[0057] Training parameters: Adam optimizer (learning rate 0.001), batch size 32, training epochs 100 (early stopping mechanism).

[0058] Training data: Historical landslide event data (sample size ≥ 100,000, labeled as whether a landslide occurred).

[0059] Output: First deformation trend map (marking high-risk areas with deformation rate > 5 mm / year and landslide probability > 70%).

[0060] S13: Add a first risk layer to the first GIS map based on the first deformation trend map to obtain a deformation risk GIS map, and output a list of high-risk area coordinates based on the deformation risk GIS map.

[0061] In step S12, the first deformation trend map has been determined, which records the deformation trend of a specified area within a preset time interval. In this step, the first deformation trend map is overlaid on the GIS map to add a risk layer to the GIS map, and finally, a list of coordinates of high-risk areas is output for subsequent focused monitoring.

[0062] Specific technical means in this step may include:

[0063] 1. GIS Integration: Import the first deformation trend map output by the deformation trend map determination model into the first GIS map (preferably QGIS 3.22), and add a first risk layer to the first GIS map in combination with terrain data (DEM resolution 1m).

[0064] 2. Threshold segmentation: The Otsu algorithm is used to automatically segment high-risk areas (deformation rate > 5 mm / year) and output a coordinate list (format: WGS-84, accuracy ±0.5m).

[0065] 3. Output: A list of high-risk area coordinates, used for UAV data acquisition and positioning in step S2. This list records at least one coordinate point where there is a significant risk of localized slope deformation.

[0066] S2: Based on the list of high-risk area coordinates, the first UAV point cloud data, and the ground sensor data, obtain the first crack propagation prediction result.

[0067] In step S1, a preliminary list of high-risk area coordinates at the macroscopic level has been determined based on satellite data. In this step, it is necessary to combine UAV LiDAR point cloud data and ground sensor data to extract high-precision crack features and displacement details, providing key mesoscale parameters for the multimodal model.

[0068] S2 includes the following sub-steps:

[0069] S21: Collect the first UAV point cloud data according to the high-risk area coordinate list, and perform first deformation rate registration processing on the first UAV point cloud data to obtain the first crack feature map.

[0070] In this step, the coordinates in the high-risk area coordinate list are used as a reference to acquire the first UAV point cloud data using a UAV, and the first UAV point cloud data is registered with the deformation risk GIS map to obtain the first crack feature map used to characterize slope deformation.

[0071] The preferred parameter settings for the drone are as follows:

[0072] Equipment parameters: Riegl VUX-1HA LiDAR (point cloud density 500 points / m², flight altitude 50m, scan frequency 160kHz).

[0073] The processing flow of S21 may specifically include:

[0074] S211: Using the list of high-risk area coordinates as a reference point, collect point cloud data of the first UAV.

[0075] Since the high-risk area coordinate list determined in S1 contains coordinate points with a high probability of slope deformation, the coordinate points in the high-risk area coordinate list need to be used as a reference to complete the point cloud data collection process when using UAVs to collect point cloud data.

[0076] Preferably, the coordinate points in the high-risk area coordinate list can be used as the center, and a specified number of coordinate points can be collected at each coordinate center.

[0077] Preferably, the coordinate points in the high-risk area coordinate list can be clustered to obtain at least one cluster center point, and a specified number of coordinate points can be collected around each of the cluster center points.

[0078] S212: Extract at least one first crack information from the first UAV point cloud data.

[0079] In this step, at least one first crack information is extracted from the first UAV point cloud data using image processing algorithms such as edge detection.

[0080] The first crack information can be in the form of an image or a text description.

[0081] In the specific implementation process, CloudCompare 2.12 can be used to extract the crack region in the first UAV point cloud data (crack width recognition error ≤ 0.5mm, using Canny edge detection + morphological opening operation).

[0082] S213: Perform first deformation rate registration processing on at least one of the first crack information and the deformation risk GIS map one by one to obtain a first crack feature map.

[0083] The deformation risk GIS map records slope deformation information obtained through satellite data analysis. In this step, in order to make the crack detection results more accurate, at least one of the first crack information needs to be registered with the deformation risk GIS map one by one, so as to stitch together the first crack information with similarity meeting the preset requirements into the first crack feature map.

[0084] In the specific implementation process, it is necessary to perform ICP registration (preferably using the iterative nearest point algorithm) on at least one of the first crack information obtained from the LiDAR point cloud analysis and the deformation risk GIS map output by S1. The first crack information that meets the preset matching degree is determined as the target crack information.

[0085] The first crack feature map is obtained by stitching together the information of at least one of the identified target cracks.

[0086] Preferably, the first crack feature map is a LiDAR crack feature map, including crack location, width, and propagation direction, with a spatial resolution of 0.1m.

[0087] S22: Perform the first fusion processing on the ground sensor data to obtain the ground sensor displacement feature map.

[0088] Satellite data and UAV data have already been fused into the first crack feature map. In order to make the slope deformation risk more accurate, ground sensor data needs to be further determined in this step for subsequent risk prediction.

[0089] The specific operation procedure of S22 is as follows:

[0090] The sources of ground sensor data are as follows:

[0091] First GNSS data (millimeter-level displacement, sampling frequency 1Hz, device: Trimble SPS986).

[0092] First crack gauge data (0.01mm accuracy, sampling frequency 0.5Hz, device: Vaisala WMT510).

[0093] Processing flow:

[0094] The first GNSS data and the first crack gauge data are fused using Kalman filtering to output a sub-millimeter displacement vector field (displacement in the X / Y / Z directions, time resolution 1Hz). Specific fusion methods can refer to methods known in the prior art and are not specifically limited here.

[0095] Output: Ground sensor displacement feature map (preferred, displacement rate, orientation angle, time series span of 7 days).

[0096] S23: Input the first crack feature map and the ground sensor displacement feature map into the crack fusion prediction model to output the first crack propagation prediction result.

[0097] In previous steps, we have obtained multi-layered analytical data based on satellite data, UAV point cloud data, and ground sensor data. In this step, we need to combine the above data to comprehensively predict the first crack fusion prediction result. The specific relationships between the data types can be found in [link to relevant documentation]. Figure 2 .

[0098] The specific structure of the crack fusion prediction model is as follows:

[0099] Input: The first crack feature map, specifically a LiDAR crack feature map (3 channels, size 256×256) + ground sensor displacement feature map (3 channels, size 256×256).

[0100] Model structure: CNN-LSTM hybrid network (CNN part: 3 convolutional layers, 64 3×3 filters per layer, ReLU activation; LSTM part: 2 layers, 64 units per layer, Dropout=0.3).

[0101] Output: Predicted crack propagation rate, crack region coordinates.

[0102] Training data: Historical crack propagation data (sample size ≥ 50,000, labeled as crack propagation rate).

[0103] Output: Crack propagation prediction results (crack region coordinates, propagation rate, accuracy ±0.3mm / day).

[0104] The crack propagation prediction results incorporate satellite data, UAV point cloud data, and ground sensor data, making the predicted crack propagation results more accurate.

[0105] S3: Collect the first meteorological and environmental data and determine the environmental variable matrix.

[0106] In steps S1 and S2, satellite data, UAV data, and ground sensor data have been integrated to predict the crack propagation. However, since slope deformation is also affected by meteorological data and environmental factors, this step requires further integration of meteorological data and environmental factors to adaptively adjust the crack propagation prediction results.

[0107] The first meteorological environmental data specifically includes:

[0108] Rainfall (minutes, from weather station, equipment: HOBO RG3-M);

[0109] Temperature and humidity (from IoT sensor, device: Sensirion SHT35).

[0110] The collected meteorological and environmental data also needs to be preprocessed. The specific process is as follows:

[0111] Rainfall intensity (cumulative amount, unit: mm / h) is calculated using a sliding window algorithm (window length 24 hours, step size 1 hour).

[0112] Soil saturation is calculated based on a temperature-humidity model. The specific calculation method can refer to the known calculation methods in the existing technology, and no specific limitation is made here.

[0113] Output: Environmental variable matrix (including rainfall intensity, soil saturation, temperature fluctuation, time resolution 1 hour).

[0114] S4: Based on the crack propagation prediction results and the environmental variable matrix, a dynamic early warning map of landslide risk is determined.

[0115] In this step, the crack propagation prediction results determined in the previous steps and the environmental variable matrix need to be combined to determine the dynamic early warning map of landslide risk, so as to realize timely early warning of local deformation of the slope.

[0116] S4 specifically includes:

[0117] S41: Input the crack propagation prediction result and the environmental variable matrix into the landslide probability prediction model to obtain at least one first landslide probability value.

[0118] In this step, the coordinates of at least one crack region in the crack propagation prediction results, along with the environmental variable matrix, need to be input into the landslide probability prediction model to determine the first landslide probability value corresponding to each crack region coordinate.

[0119] The specific landslide probability prediction model is as follows:

[0120] Input features:

[0121] The crack fusion prediction model outputs the predicted crack propagation rate (0-5 mm / day).

[0122] Environmental variable matrix (rainfall intensity, soil saturation, temperature fluctuation).

[0123] Model Design:

[0124] Random Forest (100 decision trees, maximum depth = 20, feature sampling ratio = 0.8).

[0125] Feature importance analysis: The MDI (Mean Decrease Impurity) method was used to screen key variables (crack propagation rate contribution > 40%).

[0126] Training data: Historical landslide event data (sample size ≥ 100,000, labeled as landslide probability 0-1).

[0127] Output: Landslide probability value (0-1 interval, accuracy ±0.05).

[0128] S42: Generate a dynamic early warning map of landslide risk based on at least one of the landslide probability values ​​and the first crack feature map.

[0129] In this step, a dynamic early warning map of landslide risk needs to be determined based on at least one of the landslide probability values ​​determined in S41. The dynamic early warning map of landslide risk is used to classify and display events of different risk levels using different colors and display methods.

[0130] The specific technical methods are as follows:

[0131] GIS Integration: Import landslide probability values ​​into QGIS 3.22 and generate dynamic early warning maps by combining them with terrain data (DEM resolution 1m).

[0132] Tiered strategy:

[0133] High risk (probability > 0.75): Marked in red, critical sliding surface depth ≤ 10m;

[0134] Medium risk (0.5 < probability ≤ 0.75): marked in yellow, critical sliding surface depth 10-20m.

[0135] Output: Dynamic early warning map of landslide risk (updated every hour).

[0136] This application also proposes a local deformation monitoring system for complex mountain high slopes, used to perform the aforementioned method for monitoring local deformation of complex mountain high slopes.

[0137] This application proposes a method and system for monitoring local deformation of complex mountainous high slopes, belonging to the field of slope deformation monitoring technology. It generates deformation rate maps from satellite data, extracts crack features from UAV LiDAR point clouds, and constructs a multi-scale monitoring system from macroscopic trends to microscopic details by combining data from ground sensors. Based on a hybrid model of LSTM and random forest, it integrates environmental variables and crack propagation rates to achieve landslide probability prediction and dynamic early warning. High-risk areas are automatically identified through GIS layer overlay and the Otsu algorithm. The technical solution of this application can achieve early warning of landslide risks, providing high-precision, high-efficiency, and highly adaptable technical support for the prevention and control of geological disasters in mountainous areas.

[0138] The above description is only a preferred embodiment of the present invention. Therefore, all equivalent changes or modifications made to the structure, features and principles described in the claims of this patent application are included in the scope of this patent application.

Claims

1. A method for monitoring local deformation of a high slope in a complex mountainous area, characterized in that, The method comprises: S1: determining a first deformation trend graph according to first satellite data, performing first risk analysis processing on the first deformation trend graph to obtain a high-risk area coordinate list; S2: obtaining a first crack propagation prediction result according to the high-risk area coordinate list, first unmanned aerial vehicle point cloud data and ground sensor data; S3: collecting first meteorological environment data to determine an environmental variable matrix; S4: determining a landslide risk dynamic warning graph according to the first crack propagation prediction result and the environmental variable matrix; The S1 comprises: S11: collecting first satellite data and performing first analysis processing to obtain a first deformation rate time sequence; S12: training a deformation trend graph determination model according to the first deformation rate time sequence and outputting a first deformation trend graph; S13: adding a first risk layer to the first GIS graph according to the first deformation trend graph to obtain a deformation risk GIS graph, and outputting a high-risk area coordinate list based on the deformation risk GIS graph; The S2 comprises: S21: collecting first unmanned aerial vehicle point cloud data according to the high-risk area coordinate list, and performing first deformation rate registration processing on the first unmanned aerial vehicle point cloud data to obtain a first crack feature graph; S22: performing first fusion processing on ground sensor data to obtain a ground sensor displacement feature graph; S23: inputting the first crack feature graph and the ground sensor displacement feature graph into a crack fusion prediction model to output a first crack propagation prediction result.

2. The method according to claim 1, characterized in that, The first risk analysis processing refers to superimposing the first deformation trend graph on the first satellite data to obtain a deformation risk GIS graph capable of outputting the high-risk area coordinate list.

3. The method according to claim 2, wherein, The S13 comprises: Importing the first deformation trend graph output by the deformation trend graph determination model into a first GIS graph, adding a first risk layer to the first GIS graph to obtain a deformation risk GIS graph; For the deformation risk GIS graph, an Otsu algorithm is used to automatically segment a high-risk area to obtain the high-risk area coordinate list.

4. The method according to claim 1, characterized in that, The S21 comprises: S211: collecting first unmanned aerial vehicle point cloud data with the high-risk area coordinate list as a reference point; S212: extracting at least one first crack information from the first unmanned aerial vehicle point cloud data; S213: performing first deformation rate registration processing on at least one first crack information and the deformation risk GIS graph one by one to obtain a first crack feature graph.

5. The method according to claim 4, wherein, The first crack feature graph comprises a crack position, a width and an expansion direction.

6. The method according to claim 5, wherein, The ground sensor data comprises first GNSS data and first crack meter data.

7. The method according to claim 6, wherein, The S4 comprises: S41: inputting the crack propagation prediction result and the environmental variable matrix into a landslide probability prediction model to obtain at least one first landslide probability value; S42: generating a landslide risk dynamic warning graph according to at least one landslide probability value and the first crack feature graph.

8. A complex mountainous high slope local deformation monitoring system, characterized in that, A complex mountainous high-slope local deformation monitoring method is used to realize any one of the above claims 1-7.

Images