High and steep slope rock mass structural surface intelligent identification method based on double clustering

Abstract

The invention relates to the field of rock mass structural surface recognition, in particular to a high and steep slope rock mass structural surface intelligent recognition method based on double clustering. According to the method, high-precision image data of a high and steep slope is obtained through an unmanned aerial vehicle approaching photogrammetry technology, and a three-dimensional digital model of the high and steep slope is constructed by utilizing a three-dimensional reconstruction and calculation technology. And calculating normal vectors of all triangular surfaces based on the triangular surface vertex information of the model, and performing clustering analysis to divide structural surface groups with similar occurrence. And further performing clustering analysis according to the surface center coordinates of the triangular surface, and dividing to obtain the structural surface. And finally, calculating the inclination angle, inclination direction and trend data of the structural surface through an accurate mathematical model. According to the method, the flexibility and the safety of high and steep slope structural plane measurement operation are enhanced, a measurement area of any scale can be completely covered, objective, accurate, rapid and low-cost structural plane intelligent identification and attitude data acquisition are realized, and the method has important significance in engineering application related to high and steep slopes.

Description

Technical Field

[0001] The present invention relates to the field of rock mass structural surface recognition, and in particular to a method for intelligently recognizing rock mass structural surfaces on steep slopes based on double clustering. Background Art

[0002] Rock mass structural surface measurement is a crucial task in geological and geotechnical engineering, crucial for engineering problems such as slope stability analysis. Traditional structural surface measurement requires manual contact measurement using tools such as a tape measure and compass, resulting in high workload, low efficiency, and significant environmental impact. For high and steep slopes, traditional structural surface measurement methods also face challenges such as difficulty reaching the measurement area and difficulty in acquiring comprehensive, large-scale structural surface information.

[0003] Based on rock mass digital elevation models derived from 3D reconstruction techniques such as photogrammetry and 3D laser scanning, intelligent structural surface identification technology has rapidly developed in recent years. These methods primarily include clustering, region growing, neural networks, deep learning, and rule engines. Clustering methods can identify and extract structural surfaces based on characteristics such as their occurrence, grouping, traces, and spacing. However, they are sensitive to the choice of initial cluster centers and may not be effective with complex or noisy datasets. Region growing methods can extract structural surfaces that emerge in planar form, but are sensitive to the selection of seed points, requiring manual intervention to determine appropriate seed points. Neural network methods require a large number of parameters, have limited feature extraction capabilities, and are susceptible to data influence. Deep learning methods require a large amount of labeled data for training, consuming high computational resources, and suffer from low model interpretability. Rule engine methods require specialized knowledge to build and maintain the rule engine, and may require significant manual intervention for complex or irregular structural surfaces.

[0004] Therefore, the field of high and steep slope rock mass structural surface identification urgently needs an intelligent identification method that can be applied to complex data sets, requires less human intervention, and whose identification results are not easily affected by data. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide an intelligent identification method for rock mass structural surfaces on steep slopes.

[0006] The embodiments of the present invention are implemented through the following technical solutions and steps:

[0007] (1) UAV close-up photogrammetry of steep slopes: Select appropriate drones and software tools to collect topographic information and apply for corresponding airspace; plan the drone route to ensure that the route covers the survey area; check the drone status and perform aerial photography to obtain sufficient image data from multiple angles;

[0008] (2) Reconstruction of three-dimensional digital models of steep slopes: import drone photography photos into ContextCapture software, and perform aerial triangulation, specify reconstruction area, and three-dimensional model reconstruction in sequence to obtain a three-dimensional model of the steep slope and save it in obj file format;

[0009] (3) Calculation of normal vectors of triangles in steep slope models: Assume that the three vertices on the triangle are A, B, and C; calculate two edge vectors starting from the same vertex, such as vectors AB and AC; use the cross product operation to calculate the normal vector of the two edge vectors (N = AB × AC); normalize the length of the normal vector N to ensure that its length is 1 (N = N / |N|);

[0010] (4) Clustering of structural surfaces: The normal vectors of the triangles of the steep slope model are finally divided into k categories, and k normal vectors are randomly selected as the initial cluster centers. The normal vectors of a certain triangle are classified according to their distances from the k cluster centers, and the triangle and the cluster center with the closest distance are grouped together. The center of each cluster is recalculated based on the set of normal vectors in each cluster. The above steps are repeated until one of the following conditions is met: all vertices no longer move between the k clusters, the number of iterations exceeds the set upper limit, and the clustering result meets the requirements; the triangles in the same category are set as a group of structural surfaces.

[0011] (5) Clustering of structural faces in the same group: cluster analysis is performed again for the centroid coordinates of the triangles in the same group of structural faces; the distance matrix between the centroids of all triangles is calculated, and the average centroid distance D is calculated; the distance between the centroids of the triangles is used as a judgment indicator, and the triangles with a centroid distance less than the average centroid distance D are classified as the same structural face; the centroid of the structural face is recalculated according to the weighted average of the triangle area; the above process is repeated until all triangles have been classified as structural faces;

[0012] (6) Calculation of the structural surface inclination: The normal vectors of all triangular faces included in the structural surface are averaged to obtain the normal vector of the structural surface. Since the ContextCapture software 3D reconstruction model uses the local northeast celestial coordinate system, and the inclination angle is equal to the angle between the normal vector and the Z axis, the following formula is used to calculate the structural surface inclination angle:

[0013] dip=acos(abs(FN(z)) / norm(FN));

[0014] Where dip is the inclination angle of the structural surface; acos() is the arc cosine function; abs() is the absolute value function; FN is the normal vector of the structural surface; norm(FN) is the length of the normal vector of the structural surface;

[0015] (7) Calculation of structural surface inclination: The inclination is the angle between the projection of the structural surface normal vector on the horizontal plane and the north (Y axis) direction. When the x value of the structural surface normal vector is greater than 0, the structural surface inclination can be calculated according to the following formula:

[0016] dir=acos(FN(y) / (norm([FN(x)FN(y)])));

[0017] Where dip is the structural surface inclination; acos() is the arc cosine function; abs() is the absolute value function; FN is the normal vector of the structural surface;

[0018] When the x value of the structural surface normal vector is less than 0, the structural surface inclination can be calculated according to the following formula:

[0019] dir=360-acos(FN(y) / (norm([FN(x)FN(y)])));

[0020] In addition, when the x value of the structural surface normal vector is equal to 0 and the y value of the normal vector is less than 0, the structural surface inclination is 180°; when the x value of the structural surface normal vector is equal to 0 and the y value of the normal vector is greater than 0, the structural surface inclination is 0°; when the y value of the structural surface normal vector is equal to 0 and the x value of the normal vector is less than 0, the structural surface inclination is 90°; when the y value of the structural surface normal vector is equal to 0 and the x value of the normal vector is greater than 0, the structural surface inclination is 270°;

[0021] (8) Calculation of structural surface strike: The structural surface strike is the vertical direction of the structural surface inclination, and the structural surface strike str = dir + 90°; if the calculated structural surface strike is greater than 360°, 360° needs to be subtracted to ensure that the strike is not greater than 360°.

[0022] In the present invention, the UAV route planning, if the UAV used is a DJI model, can be in the DJI Pilot software, click to create a route, or import a KML file to generate a mapping area, and the boundary points can be moved to change the scope of the survey area. Set the image overlap rate, camera angle, flight altitude and other parameters in the software. Generally, the default settings are sufficient, but can be adjusted according to actual needs. Select a suitable method for route planning from the three methods of waypoint flight, mapping aerial photography, and oblique photography. The waypoint flight method sets a series of waypoints on the map, automatically generates a route, and sets waypoint actions and parameters for each waypoint. The mapping aerial photography method selects the target area to automatically generate a route, and supports adding boundary points by map dotting, KML file importing, aircraft dotting, etc. The oblique photography method selects the target area to automatically plan an orthogonal route and multiple groups of oblique routes in different directions to build a high-precision real-life three-dimensional model.

[0023] In this invention, the overlap ratio of the drone image data must meet certain standards. Generally speaking, the heading overlap ratio should be set between 70% and 80%, and the lateral overlap ratio should be set between 60% and 80% to ensure the integrity and continuity of the image data. The drone should maintain a stable flight attitude during shooting to avoid image distortion caused by attitude changes. The shooting angle of the image should be reasonable to ensure that all details of the target area are captured.

[0024] Description of the invention principle:

[0025] The core principle of the present invention is to obtain high-precision image data of steep slopes through close-up photography by drones, and to construct a three-dimensional digital model of the steep slopes using advanced three-dimensional reconstruction and computing technologies. Based on the vertex information of the triangular faces of the three-dimensional digital model of the steep slopes, the normal vectors of all triangular faces are calculated, and the normal vectors of the triangular faces are clustered and analyzed to divide the structural faces into groups with similar attitudes. For the structural faces in the same group, cluster analysis is further performed based on the coordinates of the centroid of the triangular faces to divide the structural faces. For a single structural face, the normal vectors of all triangular faces contained in the structural face are averaged as the normal vector of the structural face. Based on the normal vector of the structural face, the inclination, dip and strike data of the structural face can be calculated.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] (1) High slope model accuracy: The present invention adopts drone high-steep slope close-up photogrammetry technology, which can obtain high-resolution topographic information and significantly improve the accuracy and details of the model, which is especially important for high-steep slope areas that are difficult to access manually.

[0028] (2) Flexible and safe measurement operations: UAVs can collect data in unsafe or difficult-to-reach areas, thereby improving the flexibility and safety of operations and reducing the risks of manual operations.

[0029] (3) Complete and comprehensive measurement range: By carefully planning the drone route, we can ensure full coverage of the measurement area and obtain data without omissions, which is difficult to achieve with traditional methods.

[0030] (4) Double clustering analysis: The structural surfaces are grouped according to their normal vectors, and then divided according to their centroid positions. This approach takes all factors into consideration and produces objective and accurate clustering results.

[0031] (5) Accurate calculation of dip, inclination and strike: The present invention calculates the dip, inclination and strike of the structural surface through an accurate mathematical model, providing an accurate description of the spatial distribution of the structural surface, which is of great significance for the analysis of geological structures and engineering applications.

[0032] (6) High computational efficiency: The cluster analysis method used in the present invention has high computational efficiency and can quickly produce results compared to traditional methods, which is crucial for the timeliness requirements in engineering applications.

[0033] (7) Low cost: UAV aerial photogrammetry technology reduces the cost of surveying and mapping operations, improves work efficiency, and requires less human resources than traditional aerial photogrammetry. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram for drone route planning; Figure 1 Middle: 1 is the UAV position control point, 2 is the UAV route, and 3 is the slope area to be measured;

[0035] Figure 2 Schematic diagram of drone close-up photogrammetry; Figure 2 Middle: 1 is the drone position at the nth second, 2 is the drone position at the n+1th second, 3 is the image area captured at the nth second, 4 is the image area captured at the n+1th second, and 5 is the overlapping image area captured at the nth and n+1th seconds;

[0036] Figure 3 Schematic diagram for calculating the normal vector of a triangle surface; Figure 3 Middle: 1 is the triangle face, 2 is the edge vector AB, 3 is the edge vector AC, and 4 is the triangle face normal vector;

[0037] Figure 4 This is a schematic diagram of the grouping of triangle normal vectors; Figure 4 Middle: 1 is the normal vector of triangle A, 2 is the normal vector of triangle B, and 3 is the normal vector of triangle C;

[0038] Figure 5 Schematic diagram of structural surface division; Figure 5 Middle: 1 is the centroid of triangle A, 2 is the centroid of triangle B, 3 is the centroid of triangle C, 4 is the centroid of triangle D, and 5 is the centroid of triangle E;

[0039] Figure 6 This is a schematic diagram for calculating the inclination, dip and strike of the structural surface; Figure 6 Middle: 1 is the structural surface, 2 is the structural surface normal vector, 3 is the Z axis, 4 is the Y axis, 5 is the X axis, and 6 is the projection of the structural surface normal vector on the XY plane;

[0040] Figure 7 This is a three-dimensional digital model of a high and steep slope;

[0041] Figure 8 This is the grouping result of a high and steep slope structural surface;

[0042] Figure 9 It is the first set of structural surfaces on a steep slope;

[0043] Figure 10 The second group of structural planes of a high and steep slope;

[0044] Figure 11 The third group of structural planes of a high and steep slope;

[0045] Figure 12 The fourth group of structural planes of a high and steep slope. DETAILED DESCRIPTION

[0046] The present application will be further described below in conjunction with the accompanying drawings. The following specific implementation steps can make the technical personnel in this field more comprehensive understanding of the present application, but not in any form limit the present application.

[0047] First of all, it needs to be pointed out that unmanned aerial photography, three-dimensional model reconstruction, clustering analysis belong to the prior art, its definition or concept has belonged to the contents of the art known or skilled. Therefore, unless there is a specific meaning the present application is made special explanation, other with the prior art consistent with the meaning of the present application is no longer one by one expression.

[0048] Unmanned aerial high and steep slope close to photogrammetry: select with proper flight height, endurance, load capacity, and stability and wind resistance outstanding unmanned aerial vehicle, to adapt to the changing terrain conditions, and ensure that the high-definition camera and high-resolution sensor can provide high-quality image data. Select a series of software tools including GIS software to process topographic information, manage airspace applications, plan routes, and analyze aerial image data. After collecting geographic information and determining the boundaries of the survey area, apply for the required airspace in accordance with local regulations, and ensure flight safety compliance. As shown in Figure 1 , use flight planning software to design routes, cover the entire survey area, and avoid obstacles. Before taking off, check the battery, GPS, camera and other key components of the unmanned aerial vehicle. When performing aerial tasks, adjust the flight height and angle according to the topography, and manually adjust the path if necessary to ensure coverage of key areas. Finally, take pictures from multiple angles to obtain sufficient image data, ensure a certain overlap between adjacent influences Figure 2 ), monitor the status of the unmanned aerial vehicle to ensure the quality and integrity of the data for subsequent accurate topographic analysis.

[0049] Reconstructing a 3D digital model of a steep slope: Copy the drone photos from a storage device to a computer, create a new project in ContextCapture software, and import the photos. Using ContextCapture's aerial triangulation function, set the relevant parameters and start processing to calculate the spatial position and pose of the photos. Define the 3D reconstruction area in the software based on the actual survey area, ensuring coverage of all steep slopes. Start the 3D reconstruction process, generate a 3D mesh model, and optimize the model for detail and accuracy. Use ContextCapture's export function, select the OBJ format, set the export parameters, and execute the export operation to generate the OBJ file. After exporting, open the OBJ file to check the model's completeness and accuracy, and use 3D modeling software for further editing and optimization if necessary.

[0050] Calculation of normal vectors of triangle surface of steep slope model: Figure 3 As shown, assume that the three vertices of a triangle are A, B, and C. Calculate two edge vectors, such as vectors AB and AC, that originate from the same vertex. Use the cross product to calculate the normal vector between these two edge vectors (N = AB × AC). Normalize the length of the normal vector N to ensure that it is 1 (N = N / |N|).

[0051] Structural face clustering: randomly select k different normal vectors from the model as initial cluster centers. Then, calculate the distance between the normal vector of each triangle face and these cluster centers and assign it to the nearest cluster center. Figure 4 As shown in the figure, triangles A and B have similar normal vector directions and belong to the same group of structural faces. The average of the normal vectors in each cluster is calculated and used as the new cluster center. This process is repeated until the normal vectors of all triangles no longer change their clusters during iteration, the preset upper limit of the number of iterations is reached, or the clustering results meet stability and accuracy requirements. Once these conditions are met, the iteration process stops, and the normal vectors of the triangles in the same cluster are considered to belong to the same group of structural faces with similar directions.

[0052] Clustering of the same group of structural faces: Calculate the centroid coordinates of each triangle, that is, take the average of the coordinates of vertices A, B, and C as the coordinates of the centroid G. Then, construct a distance matrix containing the distances between the centroids of all triangles, and calculate the average centroid distance D. Next, classify the triangles with a centroid distance less than D as the same structural face, and recalculate the centroid coordinates of each structural face according to the weighted average of the triangle area. Figure 5As shown, the centroids of triangles A, B, C, and D are close together and belong to the same structural surface. This process is repeated iteratively, including recalculating the distance matrix between centroids, updating the average centroid distance D, re-clustering based on the new distance D, and recalculating the centroids until all triangles are assigned to the corresponding structural surface or the preset upper limit of the number of iterations is reached, ensuring the stability and accuracy of the clustering results.

[0053] Structural surface inclination calculation: Figure 6 As shown in the figure, the normal vectors of all the triangular faces included in the structural surface are averaged to form the normal vector of the structural surface. Since the ContextCapture software 3D reconstruction model uses the local northeast celestial coordinate system, and the inclination angle is equal to the angle between the normal vector and the Z axis, the following formula is needed to calculate the structural surface inclination angle:

[0054] dip=acos(abs(FN(z)) / norm(FN))

[0055] Where dip is the inclination angle of the structural surface; acos() is the arc cosine function; abs() is the absolute value function; FN is the normal vector of the structural surface; norm(FN) is the length of the normal vector of the structural surface.

[0056] Structural surface inclination calculation: The inclination is the angle between the projection of the structural surface normal vector on the horizontal plane and the north (Y axis) direction. Figure 6 As shown in the figure, when the x value of the structural surface normal vector is greater than 0, the structural surface inclination can be calculated according to the following formula:

[0057] dir=acos(FN(y) / (norm([FN(x)FN(y)])))

[0058] Where dip is the inclination of the structural surface; acos() is the arc cosine function; abs() is the absolute value function; and FN is the normal vector of the structural surface.

[0059] When the x value of the structural surface normal vector is less than 0, the structural surface inclination can be calculated according to the following formula:

[0060] dir=360-acos(FN(y) / (norm([FN(x)FN(y)])))

[0061] In addition, when the x value of the structural surface normal vector is equal to 0 and the y value of the normal vector is less than 0, the structural surface inclination is 180°; when the x value of the structural surface normal vector is equal to 0 and the y value of the normal vector is greater than 0, the structural surface inclination is 0°; when the y value of the structural surface normal vector is equal to 0 and the x value of the normal vector is less than 0, the structural surface inclination is 90°; when the y value of the structural surface normal vector is equal to 0 and the x value of the normal vector is greater than 0, the structural surface inclination is 270°.

[0062] Calculation of structural plane strike: The structural plane strike is the perpendicular direction of the structural plane inclination, so strike str = dir + 90°. If the calculated strike is greater than 360°, 360° must be subtracted to ensure that the strike is not greater than 360°.

[0063] Verification of intelligent identification method for rock mass structural surface on steep slopes:

[0064] Figure 7 This is a three-dimensional digital model of a steep slope created using three-dimensional reconstruction of drone photography photos. Figure 8 This is the result of grouping the structural surfaces of a steep slope, which is divided into 4 groups of structural surfaces. Figure 9 is the first set of structural surfaces, Figure 10 is the second set of structural surfaces, Figure 11 For the third set of structural surfaces, Figure 12 The dip, inclination, and strike data of the four groups of structural surfaces are shown in Table 1.

[0065] Table 1. Data of a high and steep slope structure surface

[0066]

Claims

1. A method for intelligent identification of rock mass structural surfaces on steep slopes based on dual clustering, characterized in that: The method comprises the following steps: (1) UAV close-up photogrammetry of steep slopes: Select appropriate drones and software tools to collect topographic information and apply for corresponding airspace; plan the drone route to ensure that the route covers the survey area; check the drone status and perform aerial photography to obtain sufficient image data from multiple angles; (2) Reconstruction of three-dimensional digital models of steep slopes: import drone photography photos into ContextCapture software, and perform aerial triangulation, specify reconstruction area, and three-dimensional model reconstruction in sequence to obtain a three-dimensional model of the steep slope and save it in obj file format; (3) Calculation of normal vectors of triangles in steep slope models: Assume that the three vertices on the triangle are A, B, and C; calculate two edge vectors starting from the same vertex, such as vectors AB and AC; use the cross product operation to calculate the normal vector of the two edge vectors (N = AB × AC); normalize the length of the normal vector N to ensure that its length is 1 (N = N / |N|); (4) Clustering of structural surfaces: The normal vectors of the triangles of the steep slope model are finally divided into k categories, and k normal vectors are randomly selected as the initial cluster centers. The normal vectors of a certain triangle are classified according to their distances from the k cluster centers, and the triangle and the cluster center with the closest distance are grouped together. The center of each cluster is recalculated based on the set of normal vectors in each cluster. The above steps are repeated until one of the following conditions is met: all vertices no longer move between the k clusters, the number of iterations exceeds the set upper limit, and the clustering result meets the requirements; the triangles in the same category are set as a group of structural surfaces. (5) Clustering of structural faces in the same group: cluster analysis is performed again for the centroid coordinates of the triangles in the same group of structural faces; the distance matrix between the centroids of all triangles is calculated, and the average centroid distance D is calculated; the distance between the centroids of the triangles is used as a judgment indicator, and the triangles with a centroid distance less than the average centroid distance D are classified as the same structural face; the centroid of the structural face is recalculated according to the weighted average of the triangle area; the above process is repeated until all triangles have been classified as structural faces; (6) Calculation of the structural surface inclination: The normal vectors of all triangular faces included in the structural surface are averaged to obtain the normal vector of the structural surface. Since the ContextCapture software 3D reconstruction model uses the local northeast celestial coordinate system, and the inclination angle is equal to the angle between the normal vector and the Z axis, the following formula is used to calculate the structural surface inclination angle: dip=acos(abs(FN(z)) / norm(FN)) Where dip is the inclination angle of the structural surface; acos() is the arc cosine function; abs() is the absolute value function; FN is the normal vector of the structural surface; norm(FN) is the length of the normal vector of the structural surface; (7) Calculation of structural surface inclination: The inclination is the angle between the projection of the structural surface normal vector on the horizontal plane and the north (Y axis) direction. When the x value of the structural surface normal vector is greater than 0, the structural surface inclination can be calculated according to the following formula: dir=acos(FN(y) / (norm([FN(x)FN(y)]))) Where dip is the structural surface inclination; acos() is the arc cosine function; abs() is the absolute value function; FN is the normal vector of the structural surface; When the x value of the structural surface normal vector is less than 0, the structural surface inclination can be calculated according to the following formula: dir=360-acos(FN(y) / (norm([FN(x)FN(y)]))) Among them, when the x value of the structural surface normal vector is equal to 0 and the y value of the normal vector is less than 0, the structural surface inclination is 180°; when the x value of the structural surface normal vector is equal to 0 and the y value of the normal vector is greater than 0, the structural surface inclination is 0°; when the y value of the structural surface normal vector is equal to 0 and the x value of the normal vector is less than 0, the structural surface inclination is 90°; when the y value of the structural surface normal vector is equal to 0 and the x value of the normal vector is greater than 0, the structural surface inclination is 270°; (8) Calculation of structural surface strike: The structural surface strike is the vertical direction of the structural surface inclination, and the structural surface strike str = dir + 90°; if the calculated structural surface strike is greater than 360°, 360° needs to be subtracted to ensure that the strike is not greater than 360°.

Images