Method for quickly identifying engineering geologic features of altered rock mass

Abstract

The invention discloses an altered rock mass engineering geologic feature rapid identification method, and belongs to the technical field of engineering rock mass identification. Comprising the steps of obtaining topographic data of a rock mass feature region to be recognized, obtaining an earth surface plane model by fitting the topographic data, and generating a three-dimensional route based on the obtained earth surface plane model; controlling the unmanned aerial vehicle to fly close to the surface of the rock mass through the three-dimensional route, and shooting an image according to a preset position; performing data preprocessing on the shot image, performing image segmentation on the preprocessed image, and constructing a structural plane occurrence data set; and performing data processing based on the structural plane occurrence data set, and obtaining a rock mass feature recognition result through an FCM-WOA algorithm of structural plane occurrence automatic clustering and grouping. According to the method, the normal vector robust estimation method is optimized to improve the region growth efficiency and accuracy of the rock mass structural plane, the region growth algorithm is improved to optimize the initial structural plane recognition result, and the rock mass structural plane is quickly and accurately recognized from mass point cloud data of the rock mass outcrop.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of engineering rock mass identification, and particularly relates to a method for quickly identifying engineering geological characteristics of altered rock mass. BACKGROUND

[0002] Altered rock is commonly found in water conservancy projects, mining engineering and transportation engineering, and has a great impact on engineering stability. Patent CN108764235A proposes a method for extracting altered mineral information based on hyperspectral remote sensing data, which extracts the altered characteristics of the target area through band ratio analysis and principal component analysis (PCA). The advantage of this method is that it can efficiently cover a large area and identify multiple altered minerals. However, it is very sensitive to vegetation-covered areas and terrain shadows, which can lead to unstable data quality, and its separation ability for mixed minerals is limited, and its accuracy is highly dependent on high-quality data.

[0003] Patent CN112901183A proposes a method for determining geological characteristics during shield construction based on machine learning, which processes and transforms real-time parameters of the shield machine, and after labeling by the K-means++ algorithm, inputs it into the stacked classification algorithm to obtain the geological characteristic determination method after optimization. The advantage of this method is that it is simple and easy to operate, low in cost, and can significantly improve the construction efficiency of the shield and ensure the safety of shield tunneling, but the parameters in the machine learning model need to be determined by humans, and the selected parameters greatly affect the accuracy of the model.

[0004] Limitations of existing methods: (1) For example, although the method based on remote sensing images is efficient, it may not provide sufficient accuracy in the face of complex geological conditions; (2) While the laboratory analysis method is accurate, it is not suitable for rapid on-site evaluation.

[0005] (3) Although machine learning methods have shown strong data analysis capabilities, they still face challenges in data acquisition difficulty and interpretability. SUMMARY

[0006] The present application aims to overcome these limitations by integrating various technical means (close photography, artificial intelligence algorithm optimization) to achieve rapid and accurate identification of the engineering geological characteristics of altered rock mass, and to provide dynamic monitoring functions to adapt to different engineering application scenarios. Specifically, the present application provides a method for quickly identifying the engineering geological characteristics of altered rock mass, comprising the following steps: S1, obtaining topographic data of the rock mass feature area to be identified, and obtaining a surface plane model by fitting the topographic data, and generating a three-dimensional flight path based on the obtained surface plane model; S2, controlling the unmanned aerial vehicle to fly close to the rock mass surface through the three-dimensional flight path, and taking images according to the preset position; S3, data pre-processing is performed on the photographed images, image segmentation is performed on the pre-processed images, and a structural surface occurrence data set is constructed; S4, data processing is performed based on the structural surface occurrence data set, and a rock mass feature recognition result is obtained through the FCM-WOA algorithm of structural surface occurrence automatic clustering grouping.

[0007] As preferred, topographic data of the rock mass feature to be recognized region is obtained by using the unmanned aerial vehicle, and a ground surface plane model is obtained by fitting the topographic data; a three-dimensional flight path is generated based on the obtained ground surface plane model; the unmanned aerial vehicle is a rotary wing unmanned aerial vehicle, the camera system is arranged below the rotary wing unmanned aerial vehicle, and a flight path planning software is installed in the unmanned aerial vehicle system.

[0008] As preferred, the topographic trend of the rock mass feature recognition region is obtained by the camera system when the rotary wing unmanned aerial vehicle flies for the first time, and a topographic trend plane 3-15 m away from the ground surface plane is fitted by means of the flight path planning software; when the shooting target is determined, the camera system is perpendicular to the fitted plane for vertical shooting, and is deflected by a preset angle left and right for shooting.

[0009] As preferred, in S3, point cloud data is obtained by processing the photographed images, and is pre-processed; a point cloud neighborhood set is constructed by using KNN nearest neighbor point search; a K-d tree structure of the point cloud data is constructed and a neighborhood index is established; a normal vector and a curvature of each point are calculated; a region growing algorithm is constructed by taking the angle between the point cloud normal vector and the curvature as the growing rule and setting a threshold value; a point with the smallest curvature value is selected as a seed point for plane growing; since the threshold value of the normal vector angle is small, the point cloud segmentation will be undergrown, and a large number of small-scale point cloud plane subsets will be generated. After the region growing algorithm extracts the structural surface, the structural surfaces are merged by using smoothness constraint and coplanar constraint.

[0010] As preferred, the K-d tree structure of the point cloud data and the establishment of the neighborhood index specifically include: Sa, judging whether the point cloud neighborhood set is empty, if yes, creating an empty K-d tree structure; if no, executing Sb; Sb, calculating the variance of the X, Y and Z coordinate axes directions, and selecting an axis as a dimension direction; Sc, based on the value of the dimension direction, sorting the points in the dimension direction according to the coordinate size, determining the split dimension corresponding data in the point cloud neighborhood set, and taking the split dimension corresponding data as the center point; and the points greater than are included in the left subspace, and the points less than are included in the right subspace.

[0011] Sd, repeat steps Sa to Sc on the point cloud data, traverse the entire dataset until each space contains only one data point.

[0012] As preferred, the variance of the data points in the X, Y and Z coordinate axis directions is calculated by the following formula: ; In the formula, , , respectively, n represents the number of data points.

[0013] As preferred, the normal vector is calculated by the following formula: ; ; In the formula, k is the number of points in the neighborhood; is the three-dimensional centroid of the neighborhood; the eigenvalue of the covariance matrix C; is the eigenvalue corresponding eigenvector; The curvature is calculated by the following formula: ; In the formula, , , is the eigenvalue.

[0014] As preferred, in the S4, the FCM-WOA algorithm includes an FCM algorithm for optimizing initial clustering centers and global optimization clustering based on WOA algorithm, and the FCM algorithm for optimizing initial clustering centers specifically includes the following steps: Sa, initialize an empty data set to store k initial clustering centers; Sb, when k = 1, calculate the average similarity of each structure surface sample in the structure surface occurrence data set, and take the corresponding structure surface initial clustering center with the highest average similarity; Sc, when k = 2, calculate the distance from any structure surface sample in set S to the remaining N-1 structure surface samples, and record the maximum value of maxd; the two structure surfaces that make the maxd value maximum are moved to the initial clustering set as initial clustering centers; Sd, when k = 3, first determine the first two initial clustering centers according to the method of Sb, then calculate the distance from the remaining N-2 samples to the initial clustering centers, then the structure surface corresponding to the maximum value of the minimum value is the third initial clustering center, and it is moved to the initial clustering set; Se, if k > 3, repeat the above steps until k initial clustering centers are selected, forming an initial clustering center set.

[0015] As preferred, the global optimization clustering based on the WOA algorithm includes the following steps: Sa, initialize algorithm parameters, including whale individual population size M, clustering group number k, maximum iteration number ; Sb, randomly generate M whale individuals representing the j clustering center according to the characteristics of the rock mass structure surface occurrence unit vector and the clustering group number; randomly replace one of the whale individuals with the clustering center unit vector obtained by the FCM algorithm; Sc, calculate the distance from the sample data set to the jth clustering center of the whale individual, and for each whale individual, N x j distances will be obtained; Sd, sequentially divide each sample into the cluster corresponding to the clustering center of each whale individual according to the minimum distance principle, until all samples are divided into the sub-cluster corresponding to the clustering center of the whale individual; Se, calculate the fitness function of the whale population individual, wherein the minimum value of the fitness function corresponds to the current optimal whale individual; Sf, enter the main loop of the WOA algorithm, update the whale individual position through the whale predation mechanism, and the whale population completes one update, iteration t = t + 1; Sg, evaluate the fitness function of the entire whale population individual again, find the global optimal whale individual and its position; if the termination condition of the algorithm is met or the maximum iteration number is met, execute Sh; if not, execute Sd and continue iteration; Sh, output the global optimal whale individual, take the k unit vectors of the whale individual as the optimal clustering center, and output the optimal clustering center and its clustering division.

[0016] As preferred, the FCM-WOA algorithm clustering group includes the following steps: Sa, import the structure surface occurrence data and convert it into a unit vector; Sb, set the noise data elimination threshold, execute the noise data elimination algorithm to clean up the structure surface occurrence data to eliminate noise data, and form a structure surface occurrence unit vector data sample set; Sc, select k initial clustering centers from the sample set composed of structure surface unit vectors using the improved maximum and minimum distance method; Sd, set the iteration number of the FCM algorithm, update the clustering center using the FCM algorithm, and obtain the clustering center; Se, set the size of the whale population, set the number of iterations of the WOA algorithm for global optimization clustering analysis, obtain the global optimization clustering center and clustering grouping result; Sf, assign k=k+1, return to Sc to start a new cycle; Sg, after all k values are calculated according to the foregoing steps, the elbow method is used to determine the optimal number of clusters, and the clustering grouping result is output.

[0017] Compared with the prior art, the beneficial effects of the present application are: 1. The present application automatically obtains the rock mass feature image by close-to-surface photography of the unmanned aerial vehicle, and uses the improved region growing algorithm to identify the rock mass features, thereby improving the accuracy of the altered rock mass engineering geological feature identification.

[0018] 2. In view of the problems of low search efficiency, inaccurate normal vector calculation, and strong parameter sensitivity of the traditional region growing algorithm method in processing outcrop point cloud, the point cloud data efficient organization method is researched to improve the calculation efficiency, the normal vector robust estimation method is optimized to improve the rock mass structure surface region growing efficiency and accuracy, and the region growing algorithm is improved to optimize the initial structure surface identification result, so as to realize the fast and accurate identification of the rock mass structure surface from the massive point cloud data of the rock mass outcrop.

[0019] 3. The weight distribution is combined with the multi-expert model, the weight of the fault prediction model is adjusted according to the real data and the prediction result, the prediction performance is optimized, the identification ability of the model to the complex fault mode is improved, and the accuracy of the fault prediction model is greatly improved.

[0020] 4. The present application realizes the automatic flight and automatic image data collection of the unmanned aerial vehicle by means of close-to-surface photography and three-dimensional route planning, truly realizes the fast and intelligent identification of the altered rock engineering geological features, has high automation, and realizes the fast identification of the rock mass features through the close-to-surface photography of the unmanned aerial vehicle, the point cloud data preprocessing process and the feature identification method; liberates labor, and improves the rock mass feature identification efficiency; at the same time, the feature identification is carried out in the area which cannot be reached by the technical personnel, overcomes the limitation of geographical conditions, and reduces the danger of the technical personnel. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The flowchart of the method of the present application; Figure 2 The multi-view unmanned aerial vehicle close-to-surface photogrammetry is used to collect three-dimensional point cloud data; Figure 3 The flowchart of creating K-d tree structure of the present application; Figure 4 The point cloud normal vector and curvature calculation schematic diagram of the present application; Figure 5This is a schematic diagram illustrating the principle of using the angle between normal vectors to distinguish different planes in this invention; Figure 6 These are rock mass structure surfaces identified by the region growth algorithm of this invention; Figure 7 This is a schematic diagram of the structural surface constraint optimization of the present invention; Figure 8 This is the first result of rock mass structure surface extraction in this invention; Figure 9 This is the second result of the rock mass structure surface extraction of the present invention; Figure 10 This is a logical framework diagram of the FCM-WOA algorithm used in this invention to implement structural surface orientation clustering and grouping. Detailed Implementation

[0022] Example 1: As Figures 1-10 As shown, the purpose of this example is to provide a method for rapid identification of engineering geological features of altered rock masses. This method includes using a drone to acquire topographic data of the area to be identified for rock mass features, and obtaining a surface plane model by fitting the topographic data; generating a three-dimensional flight path based on the acquired surface plane model; controlling the drone to automatically approach the rock mass surface via the three-dimensional flight path and taking images at preset positions; preprocessing the captured images and segmenting the preprocessed images; and inputting the segmented images into a pre-trained rock mass feature recognition model to obtain the rock mass feature recognition result.

[0023] S1. Obtain topographic data of the area to be identified for rock mass features, and obtain a surface plane model by fitting the topographic data. Generate a three-dimensional flight path based on the obtained surface plane model.

[0024] A drone is used to acquire terrain data of the area to be identified for rock features, and a surface plane model is obtained by fitting the terrain data. A three-dimensional flight path is generated based on the acquired surface plane model. The drone is a rotary-wing drone, with a camera system configured below it, and flight path planning software installed within the drone system. During the initial flight of the rotary-wing drone, the camera system acquires the terrain trend of the rock feature identification area, and the flight path planning software fits a terrain trend plane 3-15m above the surface. When the shooting target is determined, the camera system performs vertical shooting perpendicular to the fitted plane and rotates left and right at preset angles.

[0025] Specifically, drones are used to acquire terrain data of the area to be identified for rock mass features, and a surface plane model is obtained by fitting the terrain data; a three-dimensional flight path is generated based on the acquired surface plane model.

[0026] The unmanned aerial vehicle can adopt a rotor unmanned aerial vehicle, which is more stable than other unmanned aerial vehicles and can meet the requirement of high-definition photographic image. The camera system is arranged below the rotor unmanned aerial vehicle, the route planning software is installed in the unmanned aerial vehicle system, and the image acquisition device and other sensors on the unmanned aerial vehicle are set to realize the function in step one S2, controlling the unmanned aerial vehicle to fly close to the rock mass surface through the three-dimensional route, and performing image shooting according to the preset position.

[0027] The unmanned aerial vehicle automatically flies close to the rock mass surface through the three-dimensional route, and performs image shooting according to the preset period. As shown in Figure 2 The main process of the close flight photography process is that the photographic system obtains the general terrain of the rock mass feature recognition area when the rotor unmanned aerial vehicle flies for the first time, and a plane 3-15 m away from the ground plane is fitted with the help of the route planning software. When the shooting target is determined, the unmanned aerial vehicle does not need to land, and the camera automatically performs vertical shooting perpendicular to the fitted plane and shooting with a left-right deflection of a preset angle. The camera system is provided with a flash to facilitate shooting in cloudy and rainy days or in shadow areas with poor light to obtain clearer pictures.

[0028] The present scheme realizes automatic flight and automatic image data acquisition of the unmanned aerial vehicle through close photography and three-dimensional route planning, truly realizes rapid and intelligent recognition of the engineering geological features of altered rock, has high automation, and realizes rapid recognition of the features of the rock mass through the unmanned aerial vehicle, the pre-processing process of the point cloud data and the feature recognition method. The present scheme liberates labor and improves the efficiency of rock feature recognition. Meanwhile, the present scheme can recognize features in areas inaccessible to technical personnel, overcomes the limitation of geographical conditions, and reduces the danger of technical personnel.

[0029] S3, performing data pre-processing on the photographed image, performing image segmentation on the pre-processed image, and constructing a structural surface occurrence data set.

[0030] In S3, the photographed image is processed to obtain point cloud data, and the point cloud data is pre-processed. The KNN nearest neighbor search is used to construct a point cloud neighborhood set, a K-d tree structure of the point cloud data is constructed, and a neighborhood index is established. The normal vector and the curvature of each point are calculated. The point cloud normal vector angle and the curvature are used as the growth rule, and a threshold value is set. The point with the smallest curvature value is selected as the seed point for plane growth. Since the threshold value of the normal vector angle is small, the point cloud segmentation will be undergrown, and a large number of small-scale point cloud plane subsets will be generated. After the structural surface is extracted by the region growing algorithm, the structural surfaces are merged by using smoothness constraint and coplanar constraint. The point cloud plane angles close to each other are subjected to smoothness constraint, and the point cloud planes with small distance and drop are subjected to coplanar constraint.

[0031] Specifically, constructing the Kd tree structure for point cloud data includes: (1) For a spatial point cloud Q, first determine whether the spatial point cloud is an empty set. If it is an empty set, create an empty Kd tree structure; if it is not an empty set, continue with the subsequent process.

[0032] (2) Calculate the variance of the X, Y and Z coordinate axes of the data points using the following formula. , , The axis with the larger variance is chosen as the spatial segmentation marker, i.e., the dimensional direction. ( (This refers to the direction axis perpendicular to the dividing hyperplane), and the best dividing effect can be obtained by dividing in this direction.

[0033] ; In the formula, , , , respectively, are the variances of the data points along the X, Y, and Z coordinate axes, and n represents the number of data points.

[0034] (3) Determine the dimensional direction according to step (2). The value of is used to sort the points along the dimension according to their coordinate size, thus determining the data corresponding to the segmentation dimension. Using this as the center point, it will be greater than Points smaller than 1 are included in the left subspace. The point is included in the right subspace.

[0035] (4) Repeat steps (1) to (3) on the point cloud data continuously until each space contains only one data point, thus completing the kd-tree construction process for the point cloud dataset. Having topological relationships between point cloud data can more effectively improve search efficiency.

[0036] Normal vector and curvature calculation For any point in the point cloud data It is necessary to use a Kd tree index query to search for the k neighborhood points of the data point, and then use the least squares distance method to find the local plane L of the point and its neighborhood points.

[0037] Let be the normal vector of plane L and satisfy... Then the fitting plane L can be expressed by the following equation: ; In the formula, d is the distance from plane L to the origin of the coordinate system. When any point The distance from the local k nearest points to the plane L is minimized, which satisfies the requirement of the best fitting plane, i.e., there is ; Since the three-dimensional centroid of the k nearest points is located on the plane L, and the normal vector n is a unit vector. By using the principal component analysis method, the problem of solving the point cloud normal vector is converted into the problem of eigenvalue decomposition of the covariance matrix C constructed by the k nearest points. That is, The normal vector is calculated by the following formula: ; ; In the formula, k is the number of points in the neighborhood; is the three-dimensional centroid of the neighborhood; the eigenvalue of the covariance matrix C; is the eigenvalue corresponding to the eigenvector; since the eigenvalue respectively represents the variation degree of the point in 3 directions, therefore the curvature can be approximately estimated by the following formula.

[0038] The curvature is calculated by the following formula: ; In the formula, , , is the eigenvalue.

[0039] Initial under-segmentation region growing The seed point with the smallest curvature is selected to start growing, and the effect of region growing is controlled by using the local surface normal vector angle threshold and the curvature threshold, as shown in the following figure. If the normal vector angle between the current point and the adjacent point is less than the threshold Sth, the adjacent point is grown into the current region.

[0040] In this case, the number of nearest neighbor search points threshold knn is set to 30 by default, the normal vector angle threshold Sth is set to 3° by default; the maximum cluster point number CSmax is fixedly set to 0.05, and the minimum cluster point number CSmin is fixedly set to 30 by default. The following figure is the structure surface identified by using the region growing algorithm, in which the red part is the seed point set, and the blue part is the plane grown based on the seed point set.

[0041] Multi-rule constraint optimization To address the undergrowth problem in the extraction results, a geometric constraint optimization rule based on visual characteristics is proposed. This rule is based on human visual perception of scenes and reflects the connectivity relationships between different point cloud planar elements. The specific rules are as follows: (1) Smoothness constraint: To merge two similar point cloud planes into one plane, the angles of the two point cloud planes should be similar, and the merged plane should visually form a smooth plane. The above visual constraint rule can be determined by the angle between the normal vectors of the two point cloud planes to be merged, that is, for any structural surface... If structural surfaces exist Satisfy its relationship with structural surfaces The angle between the normal vectors is less than .

[0042] (2) Coplanarity Constraint: The so-called coplanarity constraint means that there should not be a large difference in elevation or distance between two merged point cloud planes, that is, they should not form a stepped connection. The above visual constraint rules can be determined by the distance rules between the point clouds of the two point cloud planes to be merged, that is, the structural plane. any point in the middle To structural plane any point in minimum Euclidean distance Less than and structural surface any point To structural plane Maximum vertical distance Less than .

[0043] The structural surface constraint optimization principle is represented by the following diagram, and its mathematical expression is as follows: ; In the formula, and Structural planes and structural plane The normal vector, For structural planes The fitted plane equation, where . Figure 6 The differences between traditional region growing results and multi-rule constraint optimization results are shown. It can be seen that by optimizing the initial identification results with multi-rule constraints, the structure surface identification results are more reasonable. For structure surfaces that do not meet the merging strategy and have a point cloud count less than Minp, they are classified as unidentified structure surfaces, which can reduce noise data in the structure surface sample data.

[0044] S4, based on the structural plane occurrence data set, the FCM-WOA algorithm of automatic clustering grouping of structural plane occurrence is used to obtain the rock mass feature recognition result, that is, the clustering grouping of rock mass structural plane occurrence.

[0045] In S4, the FCM-WOA algorithm of automatic clustering grouping of structural plane occurrence: ①Before formal analysis of data, remove noise data in the structural plane data set; ②Improve the selection of FCM initial clustering center, so as to use FCM algorithm to avoid local optimal clustering center, and provide initial whale population position (initial clustering center of WOA algorithm) for WOA algorithm to obtain higher precision global optimal solution; ③Automatically determine the optimal clustering grouping number without manually specifying the grouping number.

[0046] The FCM-WOA algorithm includes the FCM algorithm for optimizing the initial clustering center and the global optimization clustering based on WOA algorithm, and the FCM algorithm for optimizing the initial clustering center specifically includes the following steps: Sa, initialize an empty data set , store k initial clustering centers; Sb, when k = 1, calculate the average similarity of each structural plane sample in the structural plane occurrence data set, and take the corresponding structural plane initial clustering center with the highest average similarity; Sc, when k = 2, calculate the distance from any structural plane sample in set S to the remaining N-1 structural plane samples, and record the maximum value as maxd; move the two structural planes with the maximum maxd value to the initial clustering set as the initial clustering centers; Sd, when k = 3, first determine the first two initial clustering centers according to the method of Sb, then calculate the distance from the remaining N-2 samples to the initial clustering centers, then take the structural plane corresponding to the maximum value of the minimum value as the third initial clustering center, and move it to the initial clustering set; Se, if k > 3, repeat the above steps until k initial clustering centers are selected, and form the initial clustering center set.

[0047] Specifically, the FCM algorithm for optimizing the initial clustering center: In order to ensure the clustering accuracy of using clustering center to represent the dominant structural plane occurrence of rock mass, the maximum and minimum distance method is usually used to select the initial clustering center. In order to improve the shortcomings of the maximum and minimum distance method in selecting the first initial clustering center, the concept of density is added when selecting the first initial clustering center, so as to improve the selection of the first initial clustering center. For a data sample set , the initial clustering center determination steps are as follows: (1) Initialize an empty data set , which is used to store k initial clustering centers.

[0048] (2) When k = 1, calculate the average similarity of each structural plane sample in the set , take the average similarity highest corresponding structural plane initial clustering center .

[0049] (3) When k = 2, calculate the distance from any structural plane sample in set S to the remaining N-1 structural plane samples, and record the maximum value as mind; the two structural planes that make the maxd value the largest are moved to the initial clustering set , at this time . That is ; (4) When k = 3, first determine the first two initial clustering centers according to method (2), then calculate the distance from the remaining N-2 samples to the initial clustering centers , record the minimum value as mind , ), then the structural plane corresponding to the maximum mind , ) is the third initial clustering center , and it is moved to the set , at this time ; (5) If k > 3, repeat step (4) until k initial clustering centers are selected, forming the initial clustering center set .

[0050] Global optimization clustering based on WOA algorithm includes the following steps: Sa, initialize algorithm parameters, including whale individual population size M, clustering grouping number k, and maximum iteration number ; Sb, randomly generate M whale individuals representing j clustering centers according to the characteristics of the unit normal vector of rock mass structural plane and the clustering grouping number; replace one of the whale individuals with the clustering center unit normal vector obtained by FCM algorithm; Sc, calculate the distance from the sample data set to the jth clustering center of the whale individual, and for each whale individual, N x j distances will be obtained; ​Sd, each sample is divided into the cluster where the corresponding cluster center of each whale individual is located according to the principle of minimum distance, until all samples are divided into the sub-cluster corresponding to the cluster center of the whale individual; Se, the fitness function of the whale population individual is calculated, wherein the whale individual corresponding to the minimum value of the fitness function is the current optimal whale individual; Sf, enter the main loop of the WOA algorithm, update the position of the whale individual through the whale predation mechanism, and the whale population completes one update, iteration t=t+1; Sg, evaluate the fitness function of the whole whale population individual again, find the global optimal whale individual and its position; if the termination condition of the algorithm is met or the maximum iteration number is met, execute Sh; if not, execute Sd, continue iteration; Sh, output the global optimal whale individual, take the k unit vectors of the whale individual as the optimal cluster center, and output the optimal cluster center and its cluster division.

[0051] Specifically, the global optimization clustering based on the WOA algorithm: The cluster center determined by the FCM clustering algorithm is taken as an input value of the population of the WOA algorithm, and the global optimal clustering result is obtained through the iterative update of the WOA algorithm. In the process of applying the WOA algorithm to the clustering grouping of the rock mass structure surface occurrence, any whale individual represents a possible solution to the problem to be optimized, that is, each whale individual has k cluster centers, and the optimal solution is obtained by continuously optimizing the whale individual to find the best whale individual.

[0052] Suppose the j cluster centers of the i whale individual are ( , ,…, ), then the cluster grouping corresponding to the whale individual is . The unit vector of the structure surface occurrence is three-dimensional data (i.e. d=3), so when j=4, the whale individual can be represented as a four-row three-column matrix composed of four cluster (cluster) center normal vectors (l, m, n) as shown in the formula.

[0053] ; In the implementation of the WOA algorithm for clustering grouping of structure surface occurrence, in order to guide the whale individual to find the position of the best whale more accurately in the optimization process, the fitness function in the algorithm implementation is defined as follows: ; In the formula, is the cluster corresponding to the j cluster center of the whale individual, is the cluster center of the cluster , j=1,2,…,k; is a cluster is a structural plane occurrence unit normal vector sample; k is the number of clusters, j = 1, 2, …, k. The smaller the value, the better the corresponding clustering effect.

[0054] The WOA algorithm is used for clustering analysis, and the clustering grouping results and the clustering center are output, and the specific steps are as follows: (1) Initialize algorithm parameters, mainly including whale individual population size M, clustering grouping number k, and maximum iteration .

[0055] (2) Randomly generate M whale individuals representing j (j = 1, 2, …, k) clustering centers according to the characteristics of the structural plane occurrence unit normal vector and the clustering grouping number; randomly replace one of the whale individuals with the clustering center unit normal vector obtained by the FCM algorithm.

[0056] (3) Calculate the distance between sample data set (i = 1, 2, …, N) and the jth clustering center of the whale individual, and for each whale individual, N x j distances will be obtained.

[0057] (4) According to the minimum distance principle, each sample is divided into a cluster corresponding to the clustering center of each whale individual, and all samples are divided into the sub-cluster corresponding to the clustering center of the whale individual.

[0058] (5) Calculate the fitness function of the whale population individual , wherein The minimum value corresponds to the current optimal whale individual . The FCM-WOA algorithm clustering grouping includes the following steps: Sa, import the structural plane occurrence data and convert it to a unit normal vector; Sb, set the noise data elimination threshold, execute the noise data elimination algorithm to clean up the structural plane occurrence data to eliminate noise data, and form a structural plane occurrence unit normal vector data sample set; Sc, select k initial clustering centers from the sample set composed of structural plane unit normal vectors using the improved maximum and minimum distance method; Sd, set the iteration number of the FCM algorithm, update the clustering center using the FCM algorithm, and obtain the clustering center; Se, set the whale population size, set the WOA algorithm iteration number for global optimization clustering analysis, and obtain the global optimization clustering center and clustering grouping result; Sf, assign k = k + 1, and return to Sc to start a new loop;​ Sg, after all k values are calculated according to the foregoing steps, the elbow method is used to determine the optimal number of clusters, and the clustering grouping result is output.

[0059] The structural planes are extracted using the Compass plug-in of the CloudCompare software. These structural planes are marked in different colors in the following figure. The attitude of these structural planes measured manually and the results of the method proposed in this paper are shown in Table 1, and the absolute value of the evaluation deviation of the dip and the dip angle is 5.2° and 3.43° respectively. The attitude of these structural planes has a large deviation from the manual fitting result, which is caused by the roughness of the structural surface or the small size of the manual fitting.

[0060] Table 1 Angle deviation of the attitude of the marked structural planes measured by the method proposed in this paper and the manual fitting result The present application aims at the problems of low search efficiency, inaccurate normal vector calculation, strong parameter sensitivity and other problems of the traditional region growing algorithm method in processing outcrop point cloud. The efficient organization method of point cloud data is researched to improve the calculation efficiency, the robust estimation method of normal vector is optimized to improve the efficiency and accuracy of rock mass structural plane region growing, and the region growing algorithm is improved to optimize the initial structural plane identification result, so as to realize the fast and accurate identification of rock mass structural plane from the massive point cloud data of rock mass outcrop.

Claims

1. A method for rapid identification of engineering geological features of altered rock masses, characterized in that, Includes the following steps: S1. Obtain topographic data of the area to be identified for rock mass features, and obtain a surface plane model by fitting the topographic data. Generate a three-dimensional flight path based on the obtained surface plane model. S2. Control the UAV to fly close to the rock surface through the three-dimensional flight path and take pictures according to the preset position; S3. Perform data preprocessing on the captured images, segment the preprocessed images, and construct a structural surface attitude dataset. S4. Based on the structural plane attitude dataset, data processing is performed, and the FCM-WOA algorithm for automatic clustering and grouping of structural plane attitudes is used to obtain the rock mass structural plane attitude clustering groups.

2. The method for rapid identification of engineering geological features of altered rock masses according to claim 1, characterized in that, The UAV acquires terrain data of the area to be identified for rock mass features, and obtains a surface plane model by fitting the terrain data; a three-dimensional flight path is generated based on the acquired surface plane model. The UAV is a rotary-wing UAV, the camera system is configured below the rotary-wing UAV, and the flight path planning software is installed in the UAV system.

3. The method for rapid identification of engineering geological features of altered rock masses according to claim 2, characterized in that, During the initial flight of the rotary-wing UAV, the camera system acquires the terrain trend of the rock feature identification area and uses flight path planning software to fit a terrain trend plane 3-15m above the ground surface. When the shooting target is determined, the camera system performs vertical shooting perpendicular to the fitted plane and rotates left and right at preset angles.

4. The method for rapid identification of engineering geological features of altered rock masses according to claim 1, characterized in that, In step S3, the captured image is processed to obtain point cloud data, and preprocessed. A neighborhood set of the point cloud is constructed using KNN nearest neighbor search. A Kd tree structure of the point cloud data is constructed and a neighborhood index is established. The normal vector and curvature of each point are obtained. A region growing algorithm is constructed with the angle between the point cloud normal vectors and the curvature as the growth rules and a threshold is set. The point with the smallest curvature value is selected as the seed point for planar growth. Since the point cloud segmentation will undergrow when the threshold of the angle between the normal vectors is small, a large number of small-scale point cloud planar subsets will be generated. After the structural surfaces are extracted by the region growing algorithm, smoothness constraints and coplanarity constraints are used to merge these structural surfaces.

5. The method for rapid identification of engineering geological features of altered rock masses according to claim 4, characterized in that, The specific steps of constructing the Kd tree structure for point cloud data and establishing a neighborhood index include: Sa: Determine if the neighborhood set of the point cloud is empty. If it is empty, create an empty Kd tree structure; otherwise, execute Sb. Sb, calculate the variance of the X, Y and Z coordinate axes of the data points, and select the axes as the dimensional directions; Sc. Based on the values ​​along the dimension, sort the points along the dimension according to their coordinates to determine the data corresponding to the segmentation dimension. Using this as the center point, it will be greater than Points smaller than 1 are included in the left subspace. The point is included in the right subspace.

6. Sd, Repeat steps Sa to Sc on the point cloud data, traversing the entire dataset until each space contains only one data point.

7. The method for rapid identification of engineering geological features of altered rock masses according to claim 5, characterized in that, The variances of the data points along the X, Y, and Z coordinate axes are calculated using the following formula: ； In the formula, , , , respectively, are the variances of the data points along the X, Y, and Z coordinate axes, and n represents the number of data points.

8. The method for rapid identification of engineering geological features of altered rock masses according to claim 4, characterized in that, The normal vector is calculated using the following formula: ； ； In the formula, k is The number of domain points; Let be the three-dimensional centroid of this neighborhood; eigenvalues ​​of the covariance matrix C; Eigenvalues The corresponding feature vector; Curvature is calculated using the following formula: ； In the formula, , , These are the eigenvalues.

9. The method for rapid identification of engineering geological features of altered rock masses according to claim 1, characterized in that, In step S4, the FCM-WOA algorithm includes an FCM algorithm for optimizing initial cluster centers and a global optimization clustering algorithm based on WOA. The FCM algorithm for optimizing initial cluster centers specifically includes the following steps: Sa, Initialize empty data set Store k initial cluster centers; Sb, When k=1, calculate the average similarity of each structural surface sample in the structural surface attitude dataset, and take the initial cluster center of the structural surface with the highest average similarity. When Sc and k=2, calculate the distance from any structural surface sample in set S to the remaining N-1 structural surface samples, and denote the maximum value as max; move the two structural surfaces that make the maximum value of max as the initial cluster centers to the initial cluster set; Sd, When k=3, first determine the first two initial cluster centers according to the method of Sb, then calculate the distance from the remaining N-2 samples to the initial cluster centers. The structure surface that makes the minimum value take the maximum value is the third initial cluster center, and move it to the initial cluster set. Se, if k>3, repeat the above steps until k initial cluster centers are selected to form an initial cluster center set.

10. A method for rapid identification of engineering geological features of altered rock masses according to claim 8, characterized in that, The global optimization clustering based on the WOA algorithm includes the following steps: Sa, initialization algorithm parameters, including whale population size M, number of clusters k, and maximum number of iterations. ; Sb. Based on the characteristics of the unit normal vector of the rock mass structural plane and the number of cluster groups, randomly generate M whale individuals representing the cluster center of j; randomly replace one of the whale individuals with the cluster center unit normal vector obtained by the FCM algorithm; Sc, calculate the sample dataset sequentially. The distance to the j-th cluster center of an individual whale will be N×j distances for each individual whale; Sd, sequentially assign each sample to the cluster corresponding to the cluster center of each whale individual according to the principle of minimum distance, until all samples are assigned to the sub-cluster corresponding to the cluster center of each whale individual; Se, calculate the fitness function of individual whales in the population, where the whale individual corresponding to the minimum value of the fitness function is the current best whale individual; Sf, enter the main loop of the WOA algorithm, update the position of individual whales through the whale predation mechanism, and the whale population completes one update, iteration t=t+1; Sg: Evaluate the fitness function of the entire whale population again to find the globally optimal whale individual and its position; if the termination condition or the maximum number of iterations is met, execute Sh; if not, execute Sd and continue iterating. Sh, output the globally optimal whale individual, taking the k unit normal vectors of this whale individual as the optimal cluster center, and output the optimal cluster center and its cluster partition.

11. A method for rapid identification of engineering geological features of altered rock masses according to claim 9, characterized in that, The FCM-WOA algorithm for clustering and grouping includes the following steps: Sa, import the structural surface attitude data and convert it into a unit normal vector; Sb. Set a noise data removal threshold, execute a noise data removal algorithm to clean the structural surface attitude data to remove noise data, and form a sample set of structural surface attitude unit normal vector data. Sc. Select k initial cluster centers from the sample set consisting of unit normal vectors of the structural surface using an improved maximum-minimum distance method; Sd: Set the number of iterations for the FCM algorithm, and use the FCM algorithm to update the cluster centers to obtain the cluster centers; Se, set the whale population size, set the number of WOA algorithm iterations to perform global optimization clustering analysis, and obtain the global optimization cluster centers and clustering results; Sf, assign k=k+1, return to Sc to start a new loop; Sg. After traversing and calculating all k values ​​according to the aforementioned steps, the elbow method is used to determine the optimal number of clusters, and the clustering results are output.

Images