Slope displacement monitoring method based on residual convolutional network classification of three-dimensional difference image

Abstract

The present invention discloses a slope displacement monitoring method based on a three-dimensional difference map classified by a residual convolutional network, and relates to the technical field of slope safety monitoring. In an embodiment of the present invention, based on the discrete point data of the slope scanned by a laser ranging device in a single-point polling manner, a three-dimensional map of the base slope and a three-dimensional map of the slope during the monitoring period are constructed using a resampling interpolation technique, and the two are subtracted to obtain a three-dimensional height difference map of the slope. The xy coordinates of the difference map are then mapped to the xy positions of the pixels of the image, and the height differences are converted into image pixel values. The problem of mining regional displacement changes in the three-dimensional data is converted into a pattern detection problem of the image, and the mature and powerful image classification network Resnet in the deep learning algorithm is used for training and classification, and the corresponding slope risk category in the difference map is output. Based on a three-dimensional point cloud modeling algorithm and a deep learning algorithm, the present invention establishes a three-dimensional analysis model for displacement monitoring, provides accurate and real-time early warning services for the slope landslide detection system, and reduces the probability of false alarms.

Description

Technical Field

[0001] The present invention relates to the technical field of slope safety monitoring, and in particular to a slope displacement monitoring method based on residual convolutional network classification three-dimensional difference map. Background Art

[0002] my country has a large number of mines, both in operation and in progress, scattered across the country. Mine slope monitoring is crucial for ensuring safe production and protecting the environment. Mine slope monitoring is primarily used to detect safety incidents and potential hazards, such as landslides, large rockfalls, and heavy rains that cause the removal of large amounts of soil from the slope surface, ensuring timely warnings.

[0003] Currently, slope landslides are often monitored using manual inspections supplemented by monitoring equipment. Common monitoring methods include slope surface measurement (rangefinders, total stations, levels, theodolites, etc.) and multi-point displacement meters. Deep horizontal displacement inclinometers, GPS measurements, synthetic aperture radar interferometry, microseismic monitoring, distributed fiber optic displacement measurement, and time domain reflectometry are also available. The disadvantages of these monitoring methods are: 1. Pre-installed sensor equipment can only monitor the situation at a single location. Using a single point for surface monitoring can easily lead to data distortion, and pre-installing multiple points can bring high costs. 2. The degree of automation is low, installation is inconvenient, and real-time data collection is difficult. Traditional measurement work is time-consuming and labor-intensive, and more importantly, the personal safety of on-site surveyors cannot be guaranteed.

[0004] At present, advanced laser rangefinders are widely used in mine slope monitoring. Three-dimensional laser ranging monitoring breaks through the limitations of traditional point-to-point monitoring methods and has the advantages of large acquisition range and long distance, high data density, and high continuity. In recent years, it has gradually been applied to the monitoring of major hazardous sources such as high and steep slopes. However, due to comprehensive considerations of cost and benefit, most of the laser ranging equipment currently used is single-point, which performs single-point polling or random scanning on the slope surface. Figure 1 As shown. This continuous change of measurement points requires mechanical rotation of the equipment or optical adjustment of the spatial angle, making it difficult to measure the distance to the same point each time through polling. At the same time, due to the complex conditions of the slope rock, soil and vegetation, there are errors in the distance data itself each time. The combination of the above two factors means that a large number of false alarms will be generated when the data results of two scans of the same point are directly based on the same point. In response to this, some methods use multi-point voting to improve the stability of the alarm. However, this method can resist individual random measurement noise, but it is ineffective when the laser equipment has continuous point positioning deviations during polling, and false alarms will still be generated. It is also insensitive to slope thinning caused by rain erosion.

[0005] Therefore, other methods use multi-point three-dimensional slope reconstruction for processing, and then use traditional machine learning methods to perform slope analysis. This can monitor changes in the entire plane or local plane, greatly improving the stability of monitoring and the slope risk patterns that can be detected. However, the main problems with this method are: 1. When monitoring a large number of risk scenarios, traditional manually designed feature extractors and classifiers are subjective, blind, and random, making them difficult to adapt to multiple scenarios; 2. The collected data is noisy, and the three-dimensional slope constructed based on this data also contains noise, making noise filters difficult to design; 3. When there are many monitoring modes, using preset thresholds or using traditional feature extractors and classifiers to judge risk patterns will lead to subjectivity in various parameter settings, making them difficult to adjust. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a slope displacement monitoring method based on a three-dimensional difference map classified by a residual convolutional network. Based on a three-dimensional point cloud modeling algorithm and a deep learning algorithm, a three-dimensional analysis model for displacement monitoring is established to provide accurate and real-time early warning services for the slope landslide detection system, thereby reducing the probability of false alarms.

[0007] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0008] A slope displacement monitoring method based on residual convolutional network classification of three-dimensional difference maps, including: a reference slope map construction process, a monitoring slope map construction process, a model training process, and a model classification process;

[0009] The process of constructing the reference slope map includes: scanning the discrete point data of the initial slope surface using a laser ranging device in a single-point polling manner, removing abnormal value points and outliers, performing three-dimensional interpolation using a resampling interpolation algorithm, and constructing a reference slope map based on the three-dimensional data points, wherein the pixel value of each point in the reference slope map is the height value of the three-dimensional data point at the corresponding position;

[0010] The monitoring slope map construction process includes: after removing outliers and outliers from the discrete point data of the slope after single-point polling scanning by a laser ranging device, performing three-dimensional interpolation using a resampling interpolation algorithm, and constructing a monitoring slope map based on the three-dimensional data points, wherein the pixel value of each point in the monitoring slope map is the height value of the three-dimensional data point at the corresponding position;

[0011] The model training process includes: subtracting the height values ​​of the monitoring slope map from the reference slope map to obtain a three-dimensional slope height difference map; collecting a large number of difference map samples, then labeling the difference map samples with displacement types, and then inputting them into a deep learning classification network for training. After the training is completed, a slope displacement monitoring classification model is obtained;

[0012] The model classification process includes: based on the discrete point data of the target slope surface scanned by a laser ranging device using single-point polling, after removing abnormal value points and outliers, using a resampling interpolation algorithm to construct a target reference slope surface map and a target monitoring slope surface map, and then subtracting them to obtain a three-dimensional height difference map of the target slope; and inputting the three-dimensional height difference map of the target slope into the slope displacement monitoring classification model to obtain a classification result.

[0013] Furthermore, the process of constructing the reference slope map specifically includes:

[0014] Step A1: After the single-point laser ranging device is installed, determine the slope scanning area, then perform grid division to determine the grid points, poll and scan the three-dimensional spatial positions (x, y, z) of all grid points, and then remove the scanned outlier points;

[0015] Step A2: Remove outliers from the remaining grid points on the X-axis, Y-axis, and Z-axis, respectively, and then record the region range (Xmin, Xmax) of the X-axis, the region range (Ymin, Ymax) of the Y-axis, and the region range (Zmin, Zmax) of the Z-axis to obtain the reference region;

[0016] Step A3: Using a resampling interpolation algorithm, using all grid points after removing abnormal and outlier points, set interpolation point networks Tnet on the X-axis (Xmin, Xmax) and on the Y-axis (Ymin, Ymax), respectively, and use a cubic linear interpolation function to obtain the height value z of the corresponding interpolation point to construct a three-dimensional data point (m, n, z);

[0017] Step A4: Map the constructed three-dimensional data points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, thereby obtaining a reference slope map.

[0018] Furthermore, the monitoring slope map construction process specifically includes:

[0019] Step B1: poll the three-dimensional spatial position (x, y, z) of the scanned grid points again, and then remove the scanned outlier points;

[0020] Step B2: remove points outside the reference area on the X-axis, Y-axis, and Z-axis, and construct three-dimensional data points (m, n, z) using a resampling interpolation algorithm;

[0021] Step B3: Map the constructed three-dimensional data points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, thus obtaining a monitoring slope map.

[0022] Furthermore, in step B1, when polling again, scan according to 1 / N of the reference number of grid points, where N is a natural number greater than 1.

[0023] Furthermore, in step A2, outliers are removed from the remaining grid points on the X-axis, Y-axis, and Z-axis, respectively, specifically including:

[0024] Step C1: input the position data (x, y, z) of all scanning points on the slope;

[0025] Step C2: On the X-axis, a grid is formed with a distance s times the average spacing of the scanning points on the X-axis, and the histogram projection distribution of the scanning points on the X-axis is counted, where s is a natural number;

[0026] Step C3: Extend the histogram from the highest point to both sides, ending the interval when the first grid number reaches 0 on the left and right sides, and determine the X-axis interval [Xmin, Xmax]. Using the same method for the X-axis, determine the Y-axis interval [Ymin, Ymax] and the Z-axis interval [Zmin, Zmax].

[0027] Step C4: Remove outliers outside the X-axis interval [Xmin, Xmax], remove outliers outside the Y-axis interval [Ymin, Ymax], and remove outliers outside the Z-axis interval [Zmin, Zmax].

[0028] Furthermore, the step A3 specifically includes:

[0029] Step D1: On the XY plane, construct an interpolation point network Tnet in the X-axis interval [Xmin, Xmax] and the Y-axis interval [Ymin, Ymax] every t meters in the XY direction, where t is a set value and t>0;

[0030] Step D2: Input the slope scan point data after removing abnormal points and outliers, project the data points onto the XY plane, and construct a Delaunay triangulation network using the Delaunay triangulation algorithm;

[0031] Step D3: Calculate which triangle each Tnet point falls within in the Delaunay triangulation. Then, use the x, y, and z coordinates of the three points in the triangle to calculate the plane equation z = f(x, y). Substitute the x, y values ​​of the interpolated Tnet point into z = f(x, y) to calculate the corresponding Z value. Calculate the Z values ​​of all points in Tnet through interpolation.

[0032] Step D4: Replace the xy values ​​of Tnet with the index coordinates to obtain a new data point set (m, n, z) of the interpolation points.

[0033] The present invention has at least the following technical effects or advantages:

[0034] 1. By constructing a 3D height difference map, the problem of mining regional displacement changes in 3D data is transformed into an image pattern detection problem. Various deep learning algorithms for image classification and detection based on convolution operations can be directly adopted, and training can be driven by large amounts of data samples to improve classification and detection capabilities. Deep learning algorithms have the ability to automatically extract effective features based on samples and ignore interfering noise, avoiding the blindness of manually designed features and the complexity of noise processing.

[0035] 2. Automatic feature extraction replaces manually designed features. Data-driven, automated feature extraction is more accurate and reliable. A significant advantage of deep learning is its data-driven approach, automatically discovering the most effective features for classification and detection. This data-driven automatic feature extraction method significantly avoids the blindness and randomness inherent in traditional manually designed feature extraction methods. As the number of slope risk types increases, manual feature extraction becomes increasingly inadequate. A data-driven approach, on the other hand, can be effective simply by increasing the number of training samples.

[0036] 3. Avoiding complex noise processing and empirically adjusting various parameters like thresholds improves algorithm adaptability. When using traditional machine learning methods for slope risk classification or regional location, noise interference can severely impact final performance. This often requires the design of complex noise filters before feature extraction. These designs are often tied to empirical knowledge and pre-defined scenarios, and can easily fail when these assumptions don't hold true in real-world applications. Furthermore, various parameters must be designed in both the feature extractor and the final classifier. To simplify the algorithm, various experimental or even empirical parameter values ​​are often used as pre-sets. The scientific validity, effectiveness, and robustness of these parameters remain to be tested. Data-driven deep learning methods, on the other hand, automatically focus on the most critical features for classification, automatically avoiding the influence of noise. They also lack numerous pre-defined parameter values, significantly improving algorithm effectiveness. For new scenarios, the model can be iteratively updated simply by adding training samples for that specific scenario, significantly improving its scalability.

[0037] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] Figure 1 A schematic diagram of scanning points of a single-point laser ranging device in the prior art used for slope monitoring;

[0040] Figure 2 Flowchart of the algorithm of the embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of scanning points according to an embodiment of the present invention;

[0042] Figure 4 This is a flow chart of a histogram projection algorithm for removing outliers according to an embodiment of the present invention;

[0043] Figure 5 Flowchart of reconstructing a three-dimensional slope image using an interpolation algorithm according to an embodiment of the present invention

[0044] Figure 6 Schematic diagram of the process of the embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention provides a slope displacement monitoring method based on a residual convolutional network classification three-dimensional difference map. Based on a three-dimensional point cloud modeling algorithm and a deep learning algorithm, a three-dimensional analysis model for displacement monitoring is established to provide accurate and real-time early warning services for the slope landslide detection system, thereby reducing the probability of false alarms.

[0046] The overall idea of ​​the technical solution of the present invention is as follows:

[0047] The present invention is based on a three-dimensional point cloud modeling algorithm and a deep learning algorithm to establish a three-dimensional analysis model for displacement monitoring, providing accurate and real-time early warning services for the slope landslide detection system, reducing the probability of false alarms, and improving the output of the monitored risk pattern. The core idea of ​​the present invention is to use the resampling interpolation technology to construct a three-dimensional map of the baseline slope and a three-dimensional map of the slope during the monitoring period based on the discrete point data of the slope scanned by the laser ranging equipment in a single-point polling manner, and subtract the two to obtain a three-dimensional height difference map of the slope. The xy coordinates of the difference map are then mapped to the xy positions of the pixels in the image, and the height difference is converted into the pixel value of the image. The problem of mining regional displacement changes in the three-dimensional data is converted into a pattern detection problem in the image. The mature and powerful image classification network Resnet in the deep learning algorithm is used for training and classification, and the corresponding slope risk category in the difference map is output.

[0048] The algorithm has two main processes: training process and classification process. First, collect labeled training samples, train the model, and then provide it to the classification process. Figure 2 shown.

[0049] The main modules in the process are divided into four stages: baseline slope map construction, monitoring slope map construction, deep learning model training, and model deployment and use. The specific process is as follows:

[0050] 1. Baseline slope map construction stage:

[0051] 1) After installing the single-point laser equipment, determine the slope scanning area, divide the grid, determine the scanning points, poll the spatial three-dimensional position of the scanning grid points, and obtain the scanning points (such as Figure 3 );

[0052] 2) Convert the point positions scanned by the laser ranging into three-dimensional spatial positions (x, y, z) and remove the scanned outlier points (data points with x, y, z values ​​of 0 or empty).

[0053] 3) Use the histogram projection algorithm on the X-axis, Y-axis and Z-axis to remove outliers; and record the area range of the X-axis (Xmin, Xmax), the area range of the Y-axis (Ymin, Ymax) and the area range of the Z-axis (Zmin, Zmax) (the algorithm is as follows Figure 4 ).

[0054] 4) Using the interpolation algorithm, construct a 3D data plane using all points after removing abnormal and outlier points. The method is to set an interpolation point network Tnet every t meters (for example, t is 0.1) between the X axis (Xmin, Xmax) and the Y axis (Ymin, Ymax), as shown in Table 1, and use the cubic linear interpolation function to obtain the z value of the corresponding interpolation point (the specific algorithm is as follows Figure 5 ).

[0055] 5) Map the constructed 3D points (m, n, z) into a 2D image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, to obtain the reference slope map.

[0056] Table 1 Interpolation point coordinate matrix of XY plane interpolation point network Tnet

[0057]

[0058] Second, the monitoring slope map construction stage:

[0059] 1) Start a new round of polling to scan the three-dimensional positions of the monitoring points in the area. To speed up the monitoring, you can perform sampling scans, such as scanning 1 / N of the benchmark number. After scanning once, go to step 2;

[0060] 2) Convert the scanned point positions into three-dimensional spatial positions (x, y, z) and remove the scanned outlier points (data points with values ​​of 0 or empty in x, y, z).

[0061] 3) Remove points outside the reference area on the x-axis, y-axis, and z-axis. The area range is the x-axis (Xmin, Xmax), the y-axis (Ymin, Ymax), and the z-axis [Zmin, Zmax].

[0062] 4) Using the interpolation algorithm, all points after removing abnormal and outlier points are constructed into a three-dimensional data plane (the specific algorithm is the same as step 4 in the construction of the benchmark slope map, such as Figure 5 ), and obtain a new data point set (m,n,z), where m and n correspond to the interpolation point index numbers, and z is the Z-axis interpolation of the interpolation point.

[0063] 5) Map the constructed three-dimensional points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, to obtain the monitored slope map.

[0064] 3: Model training stage

[0065] 1) Subtract the baseline slope map from the monitoring slope map to create a slope difference map. Collect a large number of slope difference maps and then annotate locations based on the slope patterns you need to warn of, such as local landslides, deformations, and overall scour. This transforms the slope detection problem into an image classification problem, enabling intelligent processing using deep learning algorithms.

[0066] 2) Use the deep learning residual network Resnet50 model (model parameters are shown in Table 2 in the accompanying figure) to train the training samples to obtain a trained model.

[0067] Table 2 Specification parameters of the residual convolutional network Resnet

[0068]

[0069] Note: This invention uses the structure of resnet 50-layer

[0070] 4: Model classification stage

[0071] 1) Deploy the trained model. When using it, first construct the baseline slope map according to stage one. Then, during each scanning round, use the scanned data to construct the monitoring slope map according to stage two. Subtract the two to obtain the slope difference map.

[0072] 2) The slope difference map is input into the trained model and the risk classification of the slope is output (local landslide, deformation, overall scour, etc.).

[0073] like Figure 6 As shown, an embodiment of the present invention is as follows:

[0074] A slope displacement monitoring method based on residual convolutional network classification of three-dimensional difference maps, including: a reference slope map construction process, a monitoring slope map construction process, a model training process, and a model classification process;

[0075] The process of constructing the reference slope map includes: scanning the discrete point data of the initial slope surface using a laser ranging device in a single-point polling manner, removing abnormal value points and outliers, performing three-dimensional interpolation using a resampling interpolation algorithm, and constructing a reference slope map based on the three-dimensional data points, wherein the pixel value of each point in the reference slope map is the height value of the three-dimensional data point at the corresponding position;

[0076] The monitoring slope map construction process includes: after removing outliers and outliers from the discrete point data of the slope after single-point polling scanning by a laser ranging device, performing three-dimensional interpolation using a resampling interpolation algorithm, and constructing a monitoring slope map based on the three-dimensional data points, wherein the pixel value of each point in the monitoring slope map is the height value of the three-dimensional data point at the corresponding position;

[0077] The model training process includes: subtracting the height values ​​of the monitoring slope map from the reference slope map to obtain a three-dimensional slope height difference map; collecting a large number of difference map samples, then labeling the difference map samples with displacement types, and then inputting them into a deep learning classification network for training. After the training is completed, a slope displacement monitoring classification model is obtained;

[0078] The model classification process includes: based on the discrete point data of the target slope surface scanned by a laser ranging device using single-point polling, after removing abnormal value points and outliers, using a resampling interpolation algorithm to construct a target reference slope surface map and a target monitoring slope surface map, and then subtracting them to obtain a three-dimensional height difference map of the target slope; and inputting the three-dimensional height difference map of the target slope into the slope displacement monitoring classification model to obtain a classification result.

[0079] In a possible implementation, the reference slope map construction process specifically includes:

[0080] Step A1: After the single-point laser ranging device is installed, determine the slope scanning area, then perform grid division to determine the grid points, poll and scan the three-dimensional spatial positions (x, y, z) of all grid points, and then remove the scanned outlier points;

[0081] Step A2: Remove outliers from the remaining grid points on the X-axis, Y-axis, and Z-axis, respectively, and then record the region range (Xmin, Xmax) of the X-axis, the region range (Ymin, Ymax) of the Y-axis, and the region range (Zmin, Zmax) of the Z-axis to obtain the reference region;

[0082] Step A3: Using a resampling interpolation algorithm, using all grid points after removing abnormal and outlier points, set interpolation point networks Tnet on the X-axis (Xmin, Xmax) and on the Y-axis (Ymin, Ymax), respectively, and use a cubic linear interpolation function to obtain the height value z of the corresponding interpolation point to construct a three-dimensional data point (m, n, z);

[0083] Step A4: Map the constructed three-dimensional data points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, thereby obtaining a reference slope map.

[0084] The monitoring slope map construction process specifically includes:

[0085] Step B1: Re-polling the three-dimensional spatial positions (x, y, z) of the scanned grid points, and then removing the scanned outlier points; to speed up the monitoring, re-polling can be sampled and scanned according to 1 / N of the baseline number of grid points, where N is a natural number greater than 1; because interpolation will be performed using the resampling interpolation algorithm in the subsequent step B3, the number of points after interpolation is the same as the baseline slope map.

[0086] Step B2: Remove points outside the reference area on the X-axis, Y-axis, and Z-axis, and construct three-dimensional data points (m, n, z) through a resampling interpolation algorithm.

[0087] Step B3: Map the constructed three-dimensional data points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, thus obtaining a monitoring slope map.

[0088] In step A2, outliers are removed from the remaining grid points on the X-axis, Y-axis, and Z-axis, respectively, specifically including:

[0089] Step C1: input the position data (x, y, z) of all scanning points on the slope;

[0090] Step C2: On the X-axis, a grid is formed with a distance s times the average spacing of the scanning points on the X-axis, and the histogram projection distribution of the scanning points on the X-axis is counted, where s is a natural number, which can usually be 5;

[0091] Step C3: Extend the histogram from the highest point to both sides, ending the interval when the first grid number reaches 0 on the left and right sides, and determine the X-axis interval [Xmin, Xmax]. Using the same method for the X-axis, determine the Y-axis interval [Ymin, Ymax] and the Z-axis interval [Zmin, Zmax].

[0092] Step C4: Remove outliers outside the X-axis interval [Xmin, Xmax], remove outliers outside the Y-axis interval [Ymin, Ymax], and remove outliers outside the Z-axis interval [Zmin, Zmax].

[0093] The step A3 specifically includes:

[0094] Step D1: On the XY plane, construct an interpolation point network Tnet in the X-axis interval [Xmin, Xmax] and the Y-axis interval [Ymin, Ymax] every t meters in the XY direction, where t is a set value and t>0;

[0095] Step D2: Input the slope scan point data after removing abnormal points and outliers, project the data points onto the XY plane, and construct a Delaunay triangulation network using the Delaunay triangulation algorithm;

[0096] Step D3: Calculate which triangle each Tnet point falls within in the Delaunay triangulation. Then, use the x, y, and z coordinates of the three points in the triangle to calculate the plane equation z = f(x, y). Substitute the x, y values ​​of the interpolated Tnet point into z = f(x, y) to calculate the corresponding Z value. Calculate the Z values ​​of all points in Tnet through interpolation.

[0097] Step D4: Replace the xy values ​​of Tnet with the index coordinates to obtain a new data point set (m, n, z) of the interpolation points.

[0098] By constructing a three-dimensional height difference map, the present invention transforms the problem of mining regional displacement changes in three-dimensional data into a pattern detection problem in images. This allows for direct application of various deep learning algorithms for image classification and detection, centered around convolution operations, and utilizes a large number of data samples to drive training to improve classification and detection capabilities. Deep learning algorithms automatically extract effective features based on samples and ignore interfering noise, thus avoiding the blindness and complexity of manually designed features and noise processing. The present method replaces manually designed features with automatic feature extraction, automatically extracting optimal features based on data, resulting in greater accuracy and reliability. A significant advantage of deep learning is its data-driven approach, which automatically discovers the most effective features for classification and detection. This data-driven automatic feature extraction method significantly avoids the blindness and randomness of traditional manually designed feature extraction methods. As the number of slope risk types increases, manual features become increasingly inadequate. A data-driven approach, on the other hand, requires only an increase in training samples to achieve effective results. The present method avoids complex noise processing and empirically adjusted parameters such as thresholds, resulting in greater algorithm adaptability. This data-driven deep learning approach automatically focuses on the most important features for classification, avoids the influence of noise, and does not require many preset parameters, greatly improving the algorithm's effectiveness. For new scenarios, the model can be iteratively updated simply by adding training samples for that scenario, greatly improving its scalability.

[0099] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A slope displacement monitoring method based on residual convolutional network classification three-dimensional difference map, characterized in that: include: The process of building the baseline slope map, the process of building the monitoring slope map, the model training process, and the model classification process; The process of constructing the reference slope map includes: scanning the discrete point data of the initial slope surface using a laser ranging device in a single-point polling manner, removing abnormal value points and outliers, performing three-dimensional interpolation using a resampling interpolation algorithm, and constructing a reference slope map based on the three-dimensional data points, wherein the pixel value of each point in the reference slope map is the height value of the three-dimensional data point at the corresponding position; The monitoring slope map construction process includes: after removing outliers and outliers from the discrete point data of the slope after single-point polling scanning by a laser ranging device, performing three-dimensional interpolation using a resampling interpolation algorithm, and constructing a monitoring slope map based on the three-dimensional data points, wherein the pixel value of each point in the monitoring slope map is the height value of the three-dimensional data point at the corresponding position; The model training process includes: subtracting the height values ​​of the monitoring slope map from the reference slope map to obtain a three-dimensional slope height difference map; collecting a large number of difference map samples, then labeling the difference map samples with displacement types, and then inputting them into a deep learning classification network for training. After the training is completed, a slope displacement monitoring classification model is obtained; The model classification process includes: based on the discrete point data of the target slope surface scanned by a laser ranging device using single-point polling, after removing abnormal value points and outliers, using a resampling interpolation algorithm to construct a target reference slope surface map and a target monitoring slope surface map, and then subtracting them to obtain a three-dimensional height difference map of the target slope; and inputting the three-dimensional height difference map of the target slope into the slope displacement monitoring classification model to obtain a classification result.

2. The method according to claim 1, wherein: The process of constructing the reference slope map specifically includes: Step A1: After the single-point laser ranging device is installed, determine the slope scanning area, then perform grid division to determine the grid points, poll and scan the three-dimensional spatial positions (x, y, z) of all grid points, and then remove the scanned outlier points; Step A2: Remove outliers from the remaining grid points on the X-axis, Y-axis, and Z-axis, respectively, and then record the region range (Xmin, Xmax) of the X-axis, the region range (Ymin, Ymax) of the Y-axis, and the region range (Zmin, Zmax) of the Z-axis to obtain the reference region; Step A3: Using a resampling interpolation algorithm, using all grid points after removing abnormal and outlier points, set interpolation point networks Tnet on the X-axis (Xmin, Xmax) and on the Y-axis (Ymin, Ymax), respectively, and use a cubic linear interpolation function to obtain the height value z of the corresponding interpolation point to construct a three-dimensional data point (m, n, z); Step A4: Map the constructed three-dimensional data points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, thereby obtaining a reference slope map.

3. The method according to claim 2, wherein: The monitoring slope map construction process specifically includes: Step B1: poll the three-dimensional spatial position (x, y, z) of the scanned grid points again, and then remove the scanned outlier points; Step B2: remove points outside the reference area on the X-axis, Y-axis, and Z-axis, and construct three-dimensional data points (m, n, z) using a resampling interpolation algorithm; Step B3: Map the constructed three-dimensional data points (m, n, z) into a two-dimensional image, where m and n correspond to the index points of the width and height of the image, and z corresponds to the pixel value of the image, thus obtaining a monitoring slope map.

4. The method according to claim 3, wherein: In the step B1, when polling again, the number of grid points is scanned as 1 / N of the reference number, where N is a natural number greater than 1.

5. The method according to claim 2, wherein: In step A2, outliers are removed from the remaining grid points on the X-axis, Y-axis, and Z-axis, respectively, specifically including: Step C1: input the position data (x, y, z) of all scanning points on the slope; Step C2: On the X-axis, a grid is formed with a distance s times the average spacing of the scanning points on the X-axis, and the histogram projection distribution of the scanning points on the X-axis is counted, where s is a natural number; Step C3: Extend the histogram from the highest point to both sides, ending the interval when the first grid number reaches 0 on the left and right sides, and determine the X-axis interval [Xmin, Xmax]. Using the same method for the X-axis, determine the Y-axis interval [Ymin, Ymax] and the Z-axis interval [Zmin, Zmax]. Step C4: Remove outliers outside the X-axis interval [Xmin, Xmax], remove outliers outside the Y-axis interval [Ymin, Ymax], and remove outliers outside the Z-axis interval [Zmin, Zmax].

6. The method according to claim 2, wherein: The step A3 specifically includes: Step D1: On the XY plane, construct an interpolation point network Tnet in the X-axis interval [Xmin, Xmax] and the Y-axis interval [Ymin, Ymax] every t meters in the XY direction, where t is a set value and t>0; Step D2: Input the slope scan point data after removing abnormal points and outliers, project the data points onto the XY plane, and construct a Delaunay triangulation network using the Delaunay triangulation algorithm; Step D3: Calculate which triangle each Tnet point falls within in the Delaunay triangulation. Then, use the x, y, and z coordinates of the three points in the triangle to calculate the plane equation z = f(x, y). Substitute the x, y values ​​of the interpolated Tnet point into z = f(x, y) to calculate the corresponding Z value. Calculate the Z values ​​of all points in Tnet through interpolation. Step D4: Replace the xy values ​​of Tnet with the index coordinates to obtain a new data point set (m, n, z) of the interpolation points.

Images