Rock mass structure information intelligent extraction method based on point cloud data

Abstract

The invention relates to the technical field of rock mass engineering, in particular to a rock mass structure information intelligent extraction method based on point cloud data, and the method comprises the steps: reconstructing a three-dimensional point cloud model based on three-dimensional point cloud data; the optimal neighborhood of the target point cloud in the three-dimensional point cloud model is calculated to complete rock mass structural surface recognition; fitting the optimal plane of the target point cloud, calculating the normal vector of the target point, determining the optimal group of the dominant structural plane through the effectiveness evaluation index, and completing the clustering of the rock mass structural plane; and carrying out automatic extraction on rock mass structural plane information, carrying out reclassification on each group of clustered rock mass structural planes in combination with the distance from the optimal fitting plane of each point cloud to the original point, and constructing a three-dimensional structural plane network model of the rock mass according to the extracted rock mass structural plane information. The method can effectively ensure that the edge data of the structural surface is not missing, and avoids the problem that the clustering result of the structural surface is inaccurate due to the insufficiency of a local optimal solution.

Description

Technical Field

[0001] The present application relates to the field of rock mass engineering technology, and in particular to a method for intelligently extracting rock mass structure information based on point cloud data. Background Art

[0002] The occurrence of rock mass structural surfaces is fundamental to studying slope stability. Weak structural surfaces and the interruption of multiple groups of structural surfaces can directly lead to slope instability. Traditionally, structural surface measurements rely on manual contact measurements with a compass. This is difficult and dangerous for steep or inaccessible slopes, especially for rock masses with poor stability. Therefore, intelligent identification of rock mass structural surfaces has long been a key research focus in the geotechnical field. The goal is to directly determine the occurrence of structural surfaces through intelligent identification, eliminating the need for manual contact measurements, reducing the difficulty and risk of surveys, and saving exploration costs.

[0003] In the intelligent identification of rock mass structural surfaces, rapid and accurate rock mass data collection lays a solid foundation for subsequent research. Three-dimensional laser scanning technology has begun to be applied across various industries, and intelligent structural surface identification has begun to rely on 3D laser scanning. However, the 3D reconstruction achieved by 3D laser scanning often reveals unexposed structural surfaces. Scanning and stitching individual surfaces individually is laborious and prone to large errors. Invention patent CN113838005 proposes a method for intelligent identification of rock mass structural information based on dimensional conversion. This method reconstructs a 3D model from 2D information, minimizing image information loss and improving reconstruction accuracy. Wang Shuhong used an improved HDBSCAN algorithm to identify and group structural surfaces on a rock slope in the Dashidong limestone mine in Yuxi, Yunnan. Kong Xiali et al. combined drone photography with 3D laser scanning to propose a structural surface identification method based on multi-source data fusion. Wei Cui used drone photogrammetry to identify the bedrock of the Lianghekou Hydropower Station bank slope and conduct a preliminary stability assessment. Fu Dequan used conventional drone photography technology to acquire high-precision images of dangerous rocks on the slope of Jiguanling in Fangshan District, Beijing, which were used to build a three-dimensional point cloud model and calculate the volume of the dangerous rocks.

[0004] In recent years, the identification of structural surfaces has been mainly based on point cloud data of slope rock masses. Invention patent CN113743227 adopts a close-up photography data acquisition method and introduces a method for intelligent identification of rock structural surfaces based on the K-means algorithm. It solves the problems of low recognition accuracy in digital photography, the easy identification of units with large grayscale as structural surfaces, and the difficulty in identifying small cracks with small grayscale changes, effectively improving the efficiency of structural surface identification. Some scholars mainly identify structural surfaces based on the normal vectors of point cloud data, classifying point clouds with similar normal vectors into the same category to achieve the identification of different groups of structural surfaces. Wang Peitao et al. quickly identified the structural surfaces of slope rock masses based on three-dimensional point cloud data, conducted a sensitivity analysis of the nearest neighbor K value in the KNN algorithm, the angle threshold for normal vector classification, and the threshold for the number of point clouds for filtering interference surfaces, found the most suitable threshold, achieved the identification and classification of structural surfaces, and pointed out the problem that the KNN algorithm may cause missing data on the edges of structural surfaces. Summary of the Invention

[0005] To achieve the above objectives, this application provides the following technical solutions: According to a first aspect of the present invention, the present invention claims protection for a method for intelligently extracting rock mass structure information based on point cloud data, comprising: S1, configuring a flight method and a flight route of a multi-view UAV, wherein the multi-view UAV performs near-ground photogrammetry and collects three-dimensional point cloud data; S2, based on the three-dimensional point cloud data, reconstructing a three-dimensional point cloud model by using a joint algorithm of structure from motion sparse point cloud reconstruction and multi-view stereo vision dense point cloud reconstruction; S3, calculating the best neighborhood of the target point cloud in the three-dimensional point cloud model based on the adaptive KNN algorithm, and completing the rock mass structural surface identification through coplanarity detection; S4, using the PCA algorithm to fit the optimal plane of the target point cloud, calculating the normal vector of the target point, determining the optimal grouping of the dominant structural surfaces through the effectiveness evaluation index, and completing the clustering of the rock mass structural surfaces; S5, based on the algorithm for automatic extraction of structural surface spacing and trace length information, the rock structural surface information is automatically extracted, and the clustered rock structural surfaces are reclassified based on the distance from the best fitting plane of each point cloud to the origin. A three-dimensional structural surface network model of the rock mass is constructed based on the extracted rock structural surface information.

[0006] Furthermore, the S1 further includes: Use a multi-view UAV for oblique photography, fly close to the ground, configure the heading overlap rate and the side overlap rate, and obtain impact data with a preset resolution; Planning the shooting route of the multi-view UAV to ensure that the distance between the UAV and the slope remains constant when flying on the horizontal route and the camera optical axis remains perpendicular to the slope; When flying in the vertical direction, the angle is consistent with the foot of the slope, and the overall route is in a zigzag shape along the slope.

[0007] Furthermore, the S2 further includes: ContextCapture software is used to perform SFM-MVS 3D reconstruction. By determining the internal and external parameters of the 3D point cloud data collected by the multi-view UAV, the 3D coordinates of the feature points are determined, the 3D coordinates are optimized, and dense feature matching is completed to generate a 3D point cloud model of the slope.

[0008] Furthermore, the S3 further includes: Using the K value nearest neighbor search method in the KD-tree nearest neighbor search method, the first point cloud in the three-dimensional point cloud data (x i ,y i , z i ) and the second point cloud (x j ,y j , z j ) is calculated using formula (1): (1) Where: Target point cloud and any point distance; x i 、y i 、z i is the target point cloud coordinate; x j 、y j 、z j is the coordinate of any point cloud; Adopt adaptive K value calculation; In the KNN algorithm, each point in the point cloud data is related to the target point p i The distance is a reflection of the similarity between two points. The KNN algorithm uses Euclidean distance formula 1 as the target point. and any point The distance measure between .

[0009] Furthermore, the K value calculation of the K value nearest neighbor search method also includes: Determine the value range of the maximum and minimum density K in the three-dimensional point cloud data [K min , K max ], selecting target points in the three-dimensional point cloud data to form a point cloud set; Constructing a first matrix based on the three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set; Constructing a second matrix based on the three-dimensional coordinates of the point cloud center of the neighborhood of the target point in the point cloud set; Constructing a covariance matrix based on the three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set; Combining the eigenvectors and eigenvalues of the covariance matrix; Introducing the dispersion index and characteristic index entropy, within the K value range, the characteristic index entropy E will change with the increase of K. After continuous iterative calculation, the maximum value of the characteristic index entropy E is obtained. max , at this time E max The corresponding K value is the most suitable K value.

[0010] Furthermore, the S4 further includes: The deviation parameter η is used to represent the coplanarity of each target point cloud; Based on the deviation value of each target point in the target point cloud and the determined threshold, determining whether the neighborhood of the corresponding target point is in the same plane and making a deletion decision for the corresponding target point; Construct a plane equation for the point cloud that passes the coplanarity test; Get the normal vector information of each target point and obtain the corresponding occurrence information of each target point; The genetic algorithm GA and simulated annealing algorithm SA are combined with the FCM algorithm, and the GA algorithm and SA algorithm are used to optimize the FCM algorithm.

[0011] Furthermore, the S5 further includes: Infer the spatial size of the structural surface based on the trace length of the structural surface; The distribution law of the structural surface trace length is determined. Through the relationship between the structural surface radius and the trace length, the structural surface radius and distribution law are deduced, and the ratio of the average trace length of the structural surface to the mean diameter of the structural surface disk is obtained.

[0012] The present application relates to the field of rock mass engineering technology, and in particular to a method for intelligently extracting rock mass structural information based on point cloud data. The method comprises the following steps: reconstructing a three-dimensional point cloud model based on the three-dimensional point cloud data; calculating the optimal neighborhood of a target point cloud in the three-dimensional point cloud model to identify rock mass structural surfaces; fitting the optimal plane of the target point cloud, calculating the normal vector of the target point, determining the optimal grouping of dominant structural surfaces through an effectiveness evaluation index, and clustering the rock mass structural surfaces; automatically extracting rock mass structural surface information, reclassifying each group of clustered rock mass structural surfaces based on the distance from the best fitting plane of each point cloud to the origin, and constructing a three-dimensional structural surface network model of the rock mass based on the extracted rock mass structural surface information. The present invention can effectively ensure that the edge data of the structural surface is not missing, avoiding the problem of inaccurate structural surface clustering results caused by the deficiency of the local optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1A flowchart of a method for intelligently extracting rock mass structure information based on point cloud data as claimed in an embodiment of the present application; Figure 2 A schematic diagram of a near-ground photogrammetry measurement using an unmanned aerial vehicle (UAV) for an intelligent extraction method of rock mass structure information based on point cloud data as claimed in an embodiment of the present application; Figure 3 A schematic diagram of a KD-tree structure of a method for intelligently extracting rock mass structure information based on point cloud data as claimed in an embodiment of the present application; Figure 4 This is a SAGA-FCM algorithm flow chart of a method for intelligent extraction of rock structure information based on point cloud data, as claimed in an embodiment of the present application.

[0014] Figure 5 A three-dimensional point cloud model diagram of a method for intelligent extraction of rock structure information based on point cloud data, as claimed in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0016] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features specified as "first," "second," or "third" may explicitly or implicitly include at least one of such features. In the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are intended only to illustrate the relative positional relationships and movement of components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly. Furthermore, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements and may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to such process, method, product, or apparatus.

[0017] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0018] Drone photography technology is currently being adopted by many researchers due to its compact size, portability, adaptability, and high image resolution. Most rock mass structural surface identification algorithms are based on point clouds, calculating and classifying normal vectors within point cloud neighborhoods. The KNN algorithm is a common algorithm for point cloud neighborhood calculation. The key to the KNN algorithm lies in the selection of the K value. Traditional KNN algorithms often miss edge data on structural surfaces due to their fixed global K value. Furthermore, the fuzzy C-means clustering method, commonly used for point cloud normal vector clustering, can also easily fall into local optimal solutions, leading to inaccurate structural surface clustering results.

[0019] According to the first embodiment of the present invention, referring to Figure 1 The present invention claims protection for a method for intelligently extracting rock mass structure information based on point cloud data, comprising: S1, configuring a flight method and a flight route of a multi-view UAV, wherein the multi-view UAV performs near-ground photogrammetry and collects three-dimensional point cloud data; S2, based on the three-dimensional point cloud data, reconstructing a three-dimensional point cloud model by using a joint algorithm of structure from motion sparse point cloud reconstruction and multi-view stereo vision dense point cloud reconstruction; S3, calculating the best neighborhood of the target point cloud in the three-dimensional point cloud model based on the adaptive KNN algorithm, and completing the rock mass structural surface identification through coplanarity detection; S4, using the PCA algorithm to fit the optimal plane of the target point cloud, calculating the normal vector of the target point, determining the optimal grouping of the dominant structural surfaces through the effectiveness evaluation index, and completing the clustering of the rock mass structural surfaces; S5, based on the algorithm for automatic extraction of structural surface spacing and trace length information, the rock structural surface information is automatically extracted, and the clustered rock structural surfaces are reclassified based on the distance from the best fitting plane of each point cloud to the origin. A three-dimensional structural surface network model of the rock mass is constructed based on the extracted rock structural surface information.

[0020] Furthermore, the S1 further includes: Use a multi-view UAV for oblique photography, fly close to the ground, configure the heading overlap rate and the side overlap rate, and obtain impact data with a preset resolution; Planning the shooting route of the multi-view UAV to ensure that the distance between the UAV and the slope remains constant when flying on the horizontal route and the camera optical axis remains perpendicular to the slope; When flying in the vertical direction, the angle is consistent with the foot of the slope, and the overall route is in a zigzag shape along the slope.

[0021] In this embodiment, reference is made to Figure 2 A DJI M3E drone was used for oblique photography, flying close to the ground with an 80% heading overlap and a 75% lateral overlap, acquiring impact data with a resolution of 0.8cm. The drone's flight path was planned to ensure that the drone maintained a constant distance from the slope when flying horizontally, with the camera's optical axis perpendicular to the slope. When flying vertically, the drone's angle was aligned with the slope's toe, and the overall flight path formed a zigzag pattern along the slope.

[0022] Furthermore, the S2 further includes: ContextCapture software is used to perform SFM-MVS 3D reconstruction. By determining the internal and external parameters of the 3D point cloud data collected by the multi-view UAV, the 3D coordinates of the feature points are determined, the 3D coordinates are optimized, and dense feature matching is completed to generate a 3D point cloud model of the slope.

[0023] In this embodiment, ContextCapture software is used for SFM-MVS 3D reconstruction. By determining the internal and external parameters of the photo, determining the 3D coordinates of the feature points, optimizing the 3D coordinates, and completing dense feature matching, a 3D point cloud model of the slope is generated, and the model accuracy reaches 0.8 cm.

[0024] Furthermore, the S3 further includes: KD-tree will be divided into different dimensions at different levels, refer to Figure 3 Construct a two-dimensional point set. The root node (7,2) is the median of all points sorted by x and becomes the root node. The left subtree is sorted by y, and the median is (5,4), which becomes the second-level node. The left subtree is sorted by x again, and the median is (2,3), forming the third level. Similarly, construct the right subtree until all points are added.

[0025] Using the K value nearest neighbor search method in the KD-tree nearest neighbor search method, the first point cloud in the three-dimensional point cloud data (x i ,y i , z i ) and the second point cloud (x j ,y j , z j ) is calculated using formula (1): (1) Where: Target point cloud and any point distance; x i 、y i 、z i is the target point cloud coordinate; x j 、y j 、z j is the coordinate of any point cloud; Adopt adaptive K value calculation; In the KNN algorithm, each point in the point cloud data is related to the target point p i The distance is a reflection of the similarity between two points. The KNN algorithm uses Euclidean distance formula 1 as the target point. and any point The distance measure between .

[0026] In this embodiment, the selection of the K value in the traditional KNN algorithm is the key to the algorithm model. If the K value is too large, the results will tend to be averaged, and if the K value is too small, the results will be too sensitive. To address the difficulty in deciding the K value, an adaptive K value calculation is adopted as follows: Determine the maximum and minimum values of the 3D point cloud density as the value range of K [K min , K max ], select the target point p in the point cloud data i =(x i , y i , z i ), forming a point cloud set {P}={P a , P a+1 , ,P i ,P a+k}, where the coordinates of each data point are as shown in formula 2: (2) The three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set form the matrix A as shown in formula 3: (3) The three-dimensional coordinates of the center of the target point neighborhood in the point cloud set form the matrix B as shown in formula 4: (4) The three-dimensional coordinates of the target point neighborhood point cloud {P} in the point cloud set form the covariance matrix M as shown in Formula 5: (5) The eigenvectors and eigenvalues of matrix M form formula 6: λμ=Mμ; (6) Where: λ is the eigenvalue of matrix M; μ is the eigenvector of matrix M.

[0027] Add dispersion index S λ , surface index P λ , linear index L λ Satisfy the following formulas respectively: (7) (8) (9) Where: is the third eigenvalue of matrix M; is the second eigenvalue of matrix M; is the first eigenvalue of the matrix M.

[0028] Introduce the characteristic index entropy E as shown in formula 10: (10) (11) Within the K value range, each K value within the K value range is brought into the characteristic index entropy calculation formula for calculation. The characteristic index entropy E will change with the different values of K. After continuous iterative calculation, the maximum value of the characteristic index entropy E will be obtained. max , at this time E max The corresponding K value is the most suitable K value.

[0029] In this embodiment, using different values of K will cause the total amount of data in the point cloud set to change (K refers to the K data points surrounding the target point cloud). This means that matrices A and B in Equations 3 and 4 will change. Changes in matrices A and B will cause changes in matrix M, and thus the eigenvalues and eigenvectors of matrix M, which in turn cause changes in E. The characteristic exponential entropy corresponding to each K value is calculated, and the K value that achieves the maximum characteristic exponential entropy is the most appropriate K value.

[0030] Furthermore, the K value calculation of the K value nearest neighbor search method also includes: Determine the value range of the maximum and minimum density K in the three-dimensional point cloud data [K min ,K max ], selecting target points in the three-dimensional point cloud data to form a point cloud set; Constructing a first matrix based on the three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set; Constructing a second matrix based on the three-dimensional coordinates of the point cloud center of the neighborhood of the target point in the point cloud set; Constructing a covariance matrix based on the three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set; Combining the eigenvectors and eigenvalues of the covariance matrix; Introducing the dispersion index and characteristic index entropy, within the K value range, the characteristic index entropy E will change with the increase of K. After continuous iterative calculation, the maximum value of the characteristic index entropy E is obtained. max , at this time E max The corresponding K value is the most suitable K value.

[0031] Furthermore, the S4 further includes: The deviation parameter η is used to represent the coplanarity of each target point cloud; Based on the deviation value of each target point in the target point cloud and the determined threshold, determining whether the neighborhood of the corresponding target point is in the same plane and making a deletion decision for the corresponding target point; Construct a plane equation for the point cloud that passes the coplanarity test; Get the normal vector information of each target point and obtain the corresponding occurrence information of each target point; The genetic algorithm GA and simulated annealing algorithm SA are combined with the FCM algorithm, and the GA algorithm and SA algorithm are used to optimize the FCM algorithm.

[0032] Compared with other clustering algorithms, the FCM algorithm introduces the membership index to make the clustering results more accurate.

[0033] (12) (13) Where: n is the total number of point cloud normal vector data; k is the total number of cluster centers; u ji is the membership degree of the i-th point cloud to the j-th cluster center.

[0034] )(14) Where: m is the weighting parameter, usually m is 2; is the similarity distance between the normal vector of the i-th point cloud and the j-th cluster center.

[0035] The quality of the FCM algorithm clustering result depends on the calculation result of the objective function J(U,V) in formula 14. The objective function is the weighted sum of each point cloud data to each cluster center. The smaller the objective function value, the better the clustering result.

[0036] Genetic algorithm (GA for short) is an optimization algorithm based on natural selection and natural genetic mechanisms. The core idea is that individuals in a group produce new individuals through genetic mutation, and then select excellent individuals based on the principle of survival of the fittest, evolve from generation to generation, and finally obtain the optimal group. SA algorithm is a probability-based algorithm that finds the global optimal solution by simulating the change of internal energy during high-temperature solid annealing. SA algorithm includes the Metropolis criterion (Formula 15) and the annealing process, which correspond to the inner and outer cycles respectively. In short, when the temperature is reduced at a certain rate, at a certain temperature Ti, it must undergo multiple state transformations, and each state transformation generates After that, they will choose whether to accept it completely or with a certain probability P based on the Metropolis criterion. Become a new solution.

[0037] (15) Where: x is the previous solution; is the objective function corresponding to x; for The corresponding objective function.

[0038] Since the genetic simulated annealing algorithm (SAGA) and the fuzzy C-means clustering algorithm (FCM) can complement each other, a new algorithm, the fuzzy C-means clustering algorithm based on genetic simulated annealing (SAGA-FCM), was generated. It not only effectively overcomes the shortcomings of the GA algorithm's premature convergence, the SA algorithm's slow convergence, and the FCM algorithm's susceptibility to falling into local optimality, but also designs a genetic encoding method and fitness function according to the specific circumstances of the clustering problem, making the algorithm more effective and faster to converge to the global optimal solution.

[0039] Given a population size S, the maximum number of evolutions G max , the maximum number of iterations L max , mutation probability P m , initial annealing temperature T0, temperature cooling coefficient q, crossover probability P c , selection probability P a , termination temperature T end and the termination tolerance ΔJ of the objective function; Initialize k cluster centers, generate the initial population C, and calculate the initial membership of each sample for each cluster center; Set the initial evolution times gen=0; Select, cross, and mutate the old individuals in the initial population C to generate new individuals, calculate the k cluster centers and the membership of each sample for the generated new individuals, and calculate the objective function value J corresponding to the new population individuals I '; If the objective function J I '<ΔJ, then select a new individual to replace the old one, otherwise, accept the new individual with the probability P calculated by formula 15; If the number of evolutions gen <G max , then gen=gen+1, then initialize the evolution times gen=0, otherwise proceed to the next step; If the current temperature T i <T end , then the annealing process is completed and the current optimal cluster center V is output best , otherwise continue the annealing process, T i =qT i , and then reselect the old individuals, calculate the membership and calculate the objective function.

[0040] The separation coefficient PC, cross entropy PCE, Xie-Beni index XB and Fukuyama-Sugeno index FS were used to evaluate the clustering effect.

[0041] (16) (17) (18) (19) Where: is the average value of the cluster centers; is the jth cluster center To the average of the cluster centers distance.

[0042] A larger PC value indicates greater inter-group dispersion and a better clustering effect. Smaller PCE, Xie-Beni XB, and Fukuyama-Sugeno FS values indicate a better clustering effect. By combining the results of these four metrics, we can evaluate the clustering results and determine the optimal number of groups.

[0043] In this embodiment, after the KNN algorithm is completed, the deviation parameter η (Equation 20) is first used to represent the coplanarity of each target point cloud: (20) Each target point P in the point cloud i The corresponding deviation value η i , when the deviation value of the target point is less than the determined threshold η max , then the neighborhood of the corresponding target point is considered to be in the same plane; when the deviation value of the target point is greater than the determined threshold η max, then the neighborhood of the corresponding target point is considered not to be in the same plane, and the corresponding target point is deleted.

[0044] The plane equation of the point cloud after coplanarity detection is set as Equation 21: (twenty one) Where: a, b, c are plane parameters; d is the distance from the origin to the plane.

[0045] The plane parameters a, b, c satisfy 22 (twenty two) At this time, in order to make the fitted plane the best plane, it is necessary to satisfy the minimum sum of the squares of the distances (e) from the K points in the neighborhood point cloud set {P} to the plane, that is, Equation 23. i is any point p in the neighborhood point cloud set {P} m (x m ,y m ,z m ) to the plane distance, see Equation 24.

[0046] (twenty three) (twenty four) The optimal plane parameters and normal vector can be calculated based on the eigenvalues and eigenvectors of Equations 5 and 6. The minimum eigenvalue λ3 corresponds to the minimum value of (e), and the corresponding unit eigenvector µ3 is the plane parameters a, b, and c, which satisfy Equations 25 and 26. The plane parameter d is calculated using Equation 27, and the normal vector is the unit eigenvector µ3.

[0047] (25) (26) (27) Obtaining the normal vector information for each target point, that is, the corresponding occurrence information for each target point, is the next step in clustering target points with similar occurrences. The fuzzy C-means clustering algorithm (FCM) is commonly used for clustering data on rock mass structural planes. However, this algorithm also has shortcomings and is prone to getting stuck in local optima. To overcome this problem, this section improves the FCM algorithm by combining the genetic algorithm (GA) and the simulated annealing algorithm (SA). This optimization leverages the GA and SA algorithms' strong optimization capabilities and their ability to avoid getting stuck in local optima to optimize the FCM algorithm.

[0048] Furthermore, the S5 further includes: Infer the spatial size of the structural surface based on the trace length of the structural surface; The distribution law of the structural surface trace length is determined. Through the relationship between the structural surface radius and the trace length, the structural surface radius and distribution law are deduced, and the ratio of the average trace length of the structural surface to the mean diameter of the structural surface disk is obtained.

[0049] In this embodiment, the structural surface trace is the intersection of the rock mass structural surface and the exposed surface. The trace length is the length of the trace, which is an important indicator for describing the structural surface. The spatial size of the structural surface can be inferred based on the structural surface trace length. The structural surface trace length is calculated as shown in Formula 28: (28) In the formula i——a point on the structural surface, i=1, 2, 3, ..., n-1; j——another point on the structural surface except i, j=i+1, i+2,…,n.

[0050] To obtain the distribution pattern of the structure surface radius, it is necessary to first determine the distribution pattern of the structure surface trace length. Then, through the relationship between the structure surface radius and the trace length, the structure surface radius and distribution pattern can be derived. Assume that the distance from the center point of the structure surface disk to the center point of the structure surface trace is u, the disk radius is r, and half of the trace length is ch. Then the relationship between ch, r, and u satisfies Equation 29: (29) Referring to the “Study on Mechanical Parameters and Stability of Fractured Rock Masses”, the distribution function of the ratio of the structural surface trace length to the structural surface diameter (tc / D) is given by Equation 30: (30) The probability density function of tc / D is obtained by taking the derivative of the distribution function 31: (31) The expected value of the random variable tc / D can be obtained from the density function using Equation 32.

[0051] (32) That is, the ratio of the average trace length of the structural surface to the average diameter of the structural surface disk is 0.786, so the trace length of the structural surface is approximately equal to 1.572 times the radius of the structural surface disk.

[0052] In order to further verify the feasibility of the recognition algorithm of the present invention in actual engineering, a slope project was taken as an example, and the slope point cloud data information was collected using a DJI Mavic 3E drone, where the ground resolution was 0.8 cm and the flight mode adopted close-to-ground flight.

[0053] ContextCapture software was used to reconstruct a 3D model based on the SFM-MVS algorithm. During drone photography, the scale and rotation angle of photos can change slightly due to various factors. This necessitates extracting the same feature points from different photos to achieve matching between them. This algorithm uses the Scale-Invariant Feature Transform (SIFT) algorithm. The feature points extracted by the SIFT algorithm are local points of interest on the surface of an object. They are independent of the size and orientation of the photo and are unaffected by brightness, viewing angle, and noise. After extracting key feature points from different photos, feature matching is performed to create a 3D model. The Structure from Motion (SFM) algorithm is a commonly used 3D modeling method. Based on bundle adjustment, it uses matched feature points to incrementally reconstruct photos in 3D. Determine the internal and external parameters of the photos: The internal parameters of the photos are primarily camera parameters. UAV photogrammetry uses the same onboard camera, so the camera parameters and internal parameters are the same. The external parameters of the photos are primarily orthogonal rotation matrices and translation vectors between photos. Determine the three-dimensional coordinates of feature points: After determining the internal and external parameters of the photos, the three-dimensional coordinates are calculated using triangulation and epipolar geometry. Due to errors in feature point matching, the calculated three-dimensional coordinates of the feature points are inaccurate. Bundle adjustment is required. By rotating and translating the beams in mid-air, the optimal intersection of the common points between the photos is achieved. Finally, all feature point information is determined in the coordinate system, optimizing the three-dimensional spatial coordinates and generating a sparse point cloud model. For rock slopes, extracting structural surface information is clearly not possible with a sparse point cloud model alone. Therefore, dense reconstruction of the slope's three-dimensional point cloud model is required to improve the model's accuracy and meet engineering requirements. The multi-view stereo (MVS) algorithm is a patch-based algorithm that calculates the 3D point corresponding to each pixel in the photo pixel by pixel, generating a dense 3D point cloud model.

[0054] After the 3D model was reconstructed using the SFM-MVS algorithm, the 3D point cloud model generated by a high and steep slope rock was used as the basis. However, due to the large range of the 3D model, the large number of point clouds, the long overall recognition time and the limited computer computing power, the structural surface exposure and the key structural surface area were selected for structural surface identification and occurrence extraction. Figure 5As shown in FIG, according to the maximum value of the point cloud density distribution, the value range of the point cloud neighborhood K value is set to [5, 150]. The K value increment step size was set to 5. The maximum eigenvalue entropy of the target point cloud neighborhood at different K values was calculated. The adaptive K value for each point cloud was determined based on the maximum eigenvalue entropy, and then coplanarity detection was performed. Several commonly used K values were selected as fixed K values for control experiments, with K = 20, K = 30, and K = 40. The neighborhood coplanarity detection deviation threshold was set to 0.2. The number of point clouds processed by the adaptive KNN algorithm that passed the coplanarity detection was greater than that of the traditional KNN algorithm. The number of point clouds that passed the coplanarity detection and the point cloud pass rate of various algorithms are shown in Table 1. The total number of region recognition point clouds was 52,179, and the number of point clouds that passed the coplanarity detection under the adaptive K value algorithm was 52,033, with a point cloud pass rate of 99.72%. In contrast, the point cloud pass rates for the fixed K value recognition algorithm with K = 20, K = 30, and K = 40 were 98.41%, 97.66%, and 96.19%, respectively, which were inferior to the adaptive K value.

[0055] Table 1 Point cloud quantity and point cloud pass rate

[0056] After completing the coplanarity check for each point neighborhood, point cloud normals were calculated for the points that met the requirements. The SAGA-FCM algorithm was then used to perform cluster analysis based on these normals, with the initial parameters set as shown in Table 2. Based on the point cloud data for the structural surface, the clustering effectiveness evaluation indicators PC, PCE, XB, and FS were calculated for different groupings, with the optimal grouping determined. The calculation results for the clustering effectiveness evaluation indicators for the five groupings are shown in Table 3. It can be seen that while the effectiveness evaluation indicators PC and PCE are simple to calculate and fast to run, they exhibit monotonicity with increasing number of clusters: PC decreases monotonically and PCE increases monotonically, resulting in a decrease in effectiveness and loss of reference value. The curves for XB and FS show that both XB and FS reach their extreme values at k = 5, and both XB and FS reach their minimum values. Therefore, based on the four effectiveness evaluation indicators, the optimal number of groupings for the structural surface is 5.

[0057] Table 2 Initial parameter setting table

[0058] Table 3 Clustering effectiveness evaluation index calculation results

[0059] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0060] In addition, the functional units in the various embodiments of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

[0061] The above detailed description of the specific embodiments of the invention is intended only as an example, and the present application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions of the invention are also within the scope of the present application. Therefore, equivalent changes, modifications, and improvements made without departing from the spirit and scope of the present application should be included within the scope of the present application.

Claims

1. A method for intelligent extraction of rock mass structure information based on point cloud data, characterized in that: include: S1, configuring a flight method and flight route for a multi-view UAV, wherein the multi-view UAV performs near-ground photogrammetry and collects three-dimensional point cloud data; S2, based on the three-dimensional point cloud data, reconstructing a three-dimensional point cloud model by using a joint algorithm of structure from motion sparse point cloud reconstruction and multi-view stereo vision dense point cloud reconstruction; S3, calculating the best neighborhood of the target point cloud in the three-dimensional point cloud model based on the adaptive KNN algorithm, and completing the rock mass structural surface identification through coplanarity detection; S4, using the PCA algorithm to fit the optimal plane of the target point cloud, calculating the normal vector of the target point, determining the optimal grouping of the dominant structural surfaces through the effectiveness evaluation index, and completing the clustering of the rock mass structural surfaces; S5, based on the algorithm for automatic extraction of structural surface spacing and trace length information, the rock structural surface information is automatically extracted, and the clustered rock structural surfaces are reclassified based on the distance from the best fitting plane of each point cloud to the origin. A three-dimensional structural surface network model of the rock mass is constructed based on the extracted rock structural surface information.

2. The method for intelligently extracting rock mass structure information based on point cloud data according to claim 1, characterized in that: Said S1 further includes: Use a multi-view UAV for oblique photography, fly close to the ground, configure the heading overlap rate and the side overlap rate, and obtain impact data with a preset resolution; Planning the shooting route of the multi-view UAV to ensure that the distance between the UAV and the slope remains constant when flying on the horizontal route and the camera optical axis remains perpendicular to the slope; When flying in the vertical direction, the angle is consistent with the foot of the slope, and the overall route is in a zigzag shape along the slope.

3. The method for intelligently extracting rock mass structure information based on point cloud data according to claim 1, characterized in that: Said S2 further includes: ContextCapture software is used to perform SFM-MVS 3D reconstruction. By determining the internal and external parameters of the 3D point cloud data collected by the multi-view UAV, the 3D coordinates of the feature points are determined, the 3D coordinates are optimized, and dense feature matching is completed to generate a 3D point cloud model of the slope.

4. The method for intelligently extracting rock mass structure information based on point cloud data according to claim 1, characterized in that: Said S3 further includes: Using the K value nearest neighbor search method in the KD-tree nearest neighbor search method, the first point cloud in the three-dimensional point cloud data (x i ,y i , z i ) and the second point cloud (x j ,y j , z j ) is calculated using formula (1): (1) Where: Target point cloud and any point distance; x i 、y i 、z i is the target point cloud coordinate; x j 、y j 、z j is the coordinate of any point cloud; In the KNN algorithm, each point in the point cloud data is related to the target point p i The distance is a reflection of the similarity between two points. The KNN algorithm uses Euclidean distance formula 1 as the target point. and any point The distance measure between .

5. The method for intelligently extracting rock mass structure information based on point cloud data according to claim 4, characterized in that: The K value calculation of the K value nearest neighbor search method also includes: Determine the value range of the maximum and minimum density K in the three-dimensional point cloud data [K min ,K max ], selecting target points in the three-dimensional point cloud data to form a point cloud set; Constructing a first matrix based on the three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set; Constructing a second matrix based on the three-dimensional coordinates of the point cloud center of the neighborhood of the target point in the point cloud set; Constructing a covariance matrix based on the three-dimensional coordinates of the neighborhood point cloud of the target point in the point cloud set; Combining the eigenvectors and eigenvalues of the covariance matrix; Introducing the dispersion index and characteristic index entropy, within the K value range, the characteristic index entropy E will change with the increase of K. After continuous iterative calculation, the maximum value of the characteristic index entropy E is obtained. max , at this time E max The corresponding K value is the most suitable K value.

6. The method for intelligently extracting rock mass structure information based on point cloud data according to claim 1, characterized in that: Said S4 further includes: The deviation parameter η is used to represent the coplanarity of each target point cloud; Based on the deviation value of each target point in the target point cloud and the determined threshold, determining whether the neighborhood of the corresponding target point is in the same plane and making a deletion decision for the corresponding target point; Construct a plane equation for the point cloud that passes the coplanarity test; Get the normal vector information of each target point and obtain the corresponding occurrence information of each target point; Combine the genetic algorithm GA and the simulated annealing algorithm SA with the FCM algorithm, and use the GA algorithm and SA algorithm to optimize the FCM algorithm; Given a population size S, the maximum number of evolutions G max , the maximum number of iterations L max , mutation probability P m , initial annealing temperature T0, temperature cooling coefficient q, crossover probability P c , selection probability P a , termination temperature T end and the termination tolerance ΔJ of the objective function; Initialize k cluster centers, generate the initial population C, and calculate the initial membership of each sample for each cluster center; Set the initial evolution times gen=0; Select, cross, and mutate the old individuals in the initial population C to generate new individuals, calculate the k cluster centers and the membership of each sample for the generated new individuals, and calculate the objective function value J corresponding to the new population individuals I '; If the objective function J I '<ΔJ, then select a new individual to replace the old one, otherwise, accept the new individual with probability P; If the number of evolutions gen <G max , then gen=gen+1, reset the initial evolution times gen=0, otherwise proceed to the next step; If the current temperature T i <T end , then the annealing process is completed and the current optimal cluster center V is output best , otherwise continue the annealing process, T i =qT i , and then reselect the old individuals, calculate the membership and calculate the objective function; The separation coefficient PC, cross entropy PCE, Xie-Beni index XB and Fukuyama-Sugeno index FS were used to evaluate the clustering effect.

7. The method for intelligently extracting rock mass structure information based on point cloud data according to claim 1, characterized in that: Said S5 further includes: Infer the spatial size of the structural surface based on the trace length of the structural surface; The distribution law of the structural surface trace length is determined. Through the relationship between the structural surface radius and the trace length, the structural surface radius and distribution law are deduced, and the ratio of the average trace length of the structural surface to the mean diameter of the structural surface disk is obtained.

Images