Intelligent identification method and system of rock mass structural surface based on 3D laser scanning

Abstract

The present invention provides a method and system for intelligently identifying rock structural surfaces based on three-dimensional laser scanning, which relates to the field of geological engineering technology. The method comprises collecting point cloud data of a rock surface, performing multi-scale filtering processing based on the density distribution of the point cloud data, and obtaining multi-scale features; inputting the point cloud data and its multi-scale features after the multi-scale filtering processing into a pre-trained classification model to obtain a preliminary classification result of the rock structural surface; applying a RANSAC algorithm based on the preliminary classification result and combining the multi-scale features to perform plane fitting in the point cloud data; for each plane obtained by fitting, selecting a seed point from a plurality of inner points, and combining a growth criterion defined based on the multi-scale features, using a region growing method to gradually add neighboring points that meet the growth criterion to the current plane starting from the seed point to improve the boundary of the rock structural surface; and extracting the occurrence parameters, spacing, and ductility of the rock structural surface from the plane processed by the region growing method.

Description

Technical Field

[0001] The present invention relates to the field of geological engineering technology, and in particular to a method and system for intelligently identifying rock mass structural surfaces based on three-dimensional laser scanning. Background Art

[0002] Accurately identifying and analyzing rock mass structural surfaces is crucial in fields such as geological engineering, geotechnical engineering, and mining. Rock mass structural surfaces, such as joints, faults, and fissures, are important discontinuities in the rock mass, and their presence significantly affects the stability, mechanical properties, and permeability of the rock mass. Traditional methods for identifying rock mass structural surfaces rely primarily on manual field surveys, geological mapping, and simple measurement tools. These methods are not only time-consuming and labor-intensive, but also significantly affected by human factors, making it difficult to achieve comprehensive, accurate, and rapid identification of rock mass structural surfaces.

[0003] With the rapid development of 3D laser scanning technology, its application in geological and geotechnical engineering is becoming increasingly widespread. By emitting a laser beam and receiving the reflected signal, 3D laser scanning technology can rapidly acquire the 3D coordinates of a target object's surface, generating high-precision point cloud data. This point cloud data contains rich spatial geometric information and surface texture features, enabling the automatic identification of rock mass structural surfaces.

[0004] However, there are still many challenges in identifying rock structural surfaces directly from raw point cloud data. First, point cloud data on the rock surface often contain a lot of noise and redundant information, such as vegetation cover and the roughness of the rock surface, which will interfere with the identification of structural surfaces. Secondly, the morphology and distribution characteristics of rock structural surfaces are complex and diverse, including planes, curved surfaces, crosses, parallels and other forms, and structural surfaces of different scales appear differently in point cloud data, which increases the difficulty of identification. In addition, the identification of rock structural surfaces also needs to consider key information such as their occurrence parameters (such as dip, inclination, strike), spacing and ductility. This information is crucial for assessing the stability of the rock mass and conducting engineering design.

[0005] Therefore, it is necessary to provide an intelligent identification method for rock structure surfaces based on three-dimensional laser scanning and its system to solve the above technical problems. Summary of the Invention

[0006] To solve the above technical problems, the present invention provides a method and system for intelligently identifying rock mass structural surfaces based on three-dimensional laser scanning, which improves the recognition accuracy and efficiency of rock mass structural surfaces.

[0007] The present invention provides a method for intelligently identifying rock mass structural surfaces based on three-dimensional laser scanning, the method comprising the following steps:

[0008] Collecting point cloud data of the rock mass surface, and performing multi-scale filtering processing based on the density distribution of the point cloud data to obtain multi-scale features;

[0009] Inputting the multi-scale filtered point cloud data and its multi-scale features into a pre-trained classification model to obtain a preliminary classification result of the rock mass structural surface, wherein the classification model is a deep learning model trained on a sample set of point cloud data containing various rock mass structural surface features;

[0010] Based on the preliminary classification results, applying the RANSAC algorithm and combining the multi-scale features to perform plane fitting in the point cloud data, wherein each fitted plane contains multiple inliers;

[0011] For each fitted plane, a seed point is selected from the multiple interior points, and a growth criterion defined based on multi-scale features is combined to gradually add neighboring points that meet the growth criterion to the current plane starting from the seed point using a region growing method to improve the boundary of the rock mass structural surface;

[0012] The occurrence parameters, spacing and ductility of the rock mass structural planes are extracted from the plane processed by the region growing method.

[0013] Preferably, the collecting of point cloud data of the rock mass surface and performing multi-scale filtering processing based on the density distribution of the point cloud data to obtain multi-scale features include:

[0014] Calculating the local density of each point in the point cloud data to determine the density distribution of the point cloud data;

[0015] Determining filtering parameters of different scales according to the density distribution of the point cloud data, wherein the filtering parameters include a filtering window size and a threshold;

[0016] Applying a multi-scale Gaussian filtering method to process the point cloud data according to determined filtering parameters to extract feature points at different scales;

[0017] Feature points at different scales are fused to form multi-scale features.

[0018] Preferably, the step of inputting the multi-scale filtered point cloud data and its multi-scale features into a pre-trained classification model to obtain a preliminary classification result of the rock mass structural surface includes:

[0019] Combining the spatial position of the point cloud data and the multi-scale features to construct an enhanced feature vector;

[0020] Inputting the point cloud data and the enhanced feature vector into a pre-trained classification model, performing classification reasoning, outputting a classification label for each point, and identifying points belonging to rock mass structural surfaces;

[0021] The marked points belonging to the rock mass structural surface are post-processed to form the preliminary classification results of the rock mass structural surface.

[0022] Preferably, the step of combining the spatial position of the point cloud data and the multi-scale features to construct an enhanced feature vector comprises:

[0023] Extracting a spatial position feature of each point from the point cloud data, wherein the spatial position feature is a three-dimensional coordinate;

[0024] Aligning the spatial position feature of each point with the multi-scale feature;

[0025] The spatial position features of each aligned point and the multi-scale features are concatenated to form an enhanced feature vector.

[0026] Preferably, based on the preliminary classification results, the RANSAC algorithm is applied to the point cloud data in combination with the multi-scale features to perform plane fitting, wherein each fitted plane contains multiple inliers, including:

[0027] Extracting points identified as belonging to rock mass structural surfaces from the preliminary classification results to form a candidate point set, and performing feature enhancement processing in combination with the corresponding multi-scale features;

[0028] In the candidate point set, a RANSAC algorithm is applied to perform plane fitting, wherein the RANSAC algorithm randomly selects a set of points from the candidate point set in an iterative manner to fit a plane, and evaluates the degree of matching between the plane and other points in the candidate point set;

[0029] In each iteration, multi-scale features are combined to evaluate the quality of the fitted plane;

[0030] Setting an inlier threshold and obtaining inliers within the fitted plane, wherein the inliers refer to points whose distance from the fitted plane is less than the inlier threshold;

[0031] After each iteration, the quality of the currently fitted plane is evaluated based on the number and quality of the interior points, and the best fitted plane is selected as the result of the current iteration.

[0032] Preferably, for each plane obtained by fitting, a seed point is selected from the multiple interior points, and a growth criterion defined based on multi-scale features is combined, and a region growing method is used to gradually add neighboring points that meet the growth criterion to the current plane starting from the seed point to improve the boundary of the rock mass structural surface, including:

[0033] Select one or more points as seed points from the multiple interior points contained in the plane obtained by each fitting;

[0034] Based on the multi-scale features, a growth criterion is defined for determining whether a neighboring point should be added to the current plane, wherein the growth criterion includes a distance between the neighboring point and the seed point, consistency of the normal direction, and similarity of the multi-scale features;

[0035] Based on the selected seed point, the region growing method is used to gradually add neighboring points that meet the growth criteria to the current plane. In the process of adding points each time, the matching degree between the newly added points and the current plane is re-evaluated.

[0036] Through multiple iterations and growth, until no adjacent points that meet the growth criteria can be found.

[0037] Preferably, the step of extracting the occurrence parameters, spacing, and ductility of the rock mass structural plane from the plane processed by the region growing method includes:

[0038] Smoothing the plane processed by the region growing method;

[0039] Calculating the occurrence parameters of the rock mass structural surface using the smoothed plane, wherein the occurrence parameters include dip, inclination and strike;

[0040] Between confirmed rock mass structural surfaces, measuring the spacing between adjacent structural surfaces, wherein the spacing includes vertical spacing and horizontal spacing;

[0041] The ductility of the rock mass structural surface is obtained by calculating the length, width and connection parameters of the rock mass structural surface with other structural surfaces.

[0042] The present invention also provides a rock mass structural surface intelligent recognition system based on three-dimensional laser scanning, which is used to implement the rock mass structural surface intelligent recognition method based on three-dimensional laser scanning. The system includes:

[0043] A point cloud data multi-scale filtering module is used to collect point cloud data of the rock surface and perform multi-scale filtering based on the density distribution of the point cloud data to obtain multi-scale features;

[0044] A rock mass structural surface preliminary classification module is used to input the point cloud data and its multi-scale features processed by multi-scale filtering into a pre-trained classification model to obtain preliminary classification results of the rock mass structural surface, wherein the classification model is a deep learning model trained with a sample set of point cloud data containing multiple rock mass structural surface features;

[0045] A RANSAC plane fitting module is used to apply the RANSAC algorithm based on the preliminary classification results and perform plane fitting in the point cloud data in combination with the multi-scale features, wherein each fitted plane contains multiple inliers;

[0046] A region growing method boundary improvement module is used to select a seed point from the multiple interior points for each fitted plane, and use the region growing method to gradually add neighboring points that meet the growth criteria to the current plane starting from the seed point in combination with the growth criteria defined based on multi-scale features, so as to improve the boundary of the rock mass structural surface;

[0047] The structural surface parameter extraction module is used to extract the occurrence parameters, spacing and ductility of the rock structural surface from the plane processed by the region growing method.

[0048] Compared with related technologies, the intelligent identification method and system for rock mass structural surfaces based on three-dimensional laser scanning provided by the present invention have the following beneficial effects:

[0049] This method collects point cloud data from the rock mass surface and performs multiscale filtering to extract feature information at different scales. A pretrained deep learning model is then used to perform preliminary classification of the point cloud data. The RANSAC algorithm and region growing method are then used to further refine the identification results, ultimately extracting key parameters of the rock mass structural surface. This method not only effectively removes noise and redundant information, improving recognition accuracy and efficiency, but also comprehensively and accurately captures the morphology and distribution characteristics of the rock mass structural surface, providing powerful technical support for the fields of geological engineering and geotechnical engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of the method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning provided by the present invention;

[0051] Figure 2 This is a module structure diagram of the rock structure surface intelligent identification system based on three-dimensional laser scanning provided by the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all of the structures. Furthermore, the embodiments of the present invention and the features of the embodiments may be combined with one another unless there is a conflict.

[0053] It should also be noted that, for ease of description, only the part relevant to the present invention, rather than all of the content, is shown in the accompanying drawings. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processing or methods depicted as flow charts. Although flow charts describe various operations (or steps) as sequential processing, many operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of various operations can be rearranged. When its operation is completed, the processing can be terminated, but can also have additional steps not included in the accompanying drawings. The processing can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0054] Example 1

[0055] The present invention provides a method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning. Figure 1 As shown, reference Figure 1 As shown, the method includes the following steps:

[0056] S1: collecting point cloud data of the rock surface, and performing multi-scale filtering processing based on the density distribution of the point cloud data to obtain multi-scale features.

[0057] Specifically, step S1 includes the following steps:

[0058] S11: Calculating the local density of each point in the point cloud data to determine the density distribution of the point cloud data.

[0059] In 3D laser scanning, the point cloud data obtained from rock surfaces often has an uneven density distribution, which can be caused by variations in scanning angle, distance, or surface characteristics. To effectively identify and analyze rock structural surfaces, it's necessary to first understand the density characteristics of the point cloud data. To this end, in step S11, the local density around each point is calculated.

[0060] This process can be achieved by counting the number of points in each point's neighborhood. The neighborhood size can be defined as a spherical or cubic window, whose radius or side length is predetermined based on the resolution of the point cloud data and the expected scale of the structural surface. In this way, a density distribution map of the entire point cloud dataset can be obtained. This map can show which areas have dense point clouds and which areas have sparse point clouds, which is crucial for subsequent filtering.

[0061] S12: Determine filtering parameters of different scales according to the density distribution of the point cloud data, wherein the filtering parameters include a filtering window size and a threshold.

[0062] In this embodiment, based on the density distribution information obtained in step S11, filtering parameters can be set specifically to adapt to the data characteristics of different density areas. For high-density areas, a smaller filter window may be sufficient because these areas are richer in details; for low-density areas, a larger filter window is required to ensure that enough points participate in the filtering, thereby reducing the impact of noise.

[0063] In addition, a threshold is set to distinguish which points should be retained and which points should be removed as noise. The selection of this threshold depends on an understanding of the structural characteristics of the rock mass and experimental verification. Proper selection of filtering parameters helps improve the accuracy and efficiency of feature extraction.

[0064] S13: Applying a multi-scale Gaussian filtering method to process the point cloud data according to the determined filtering parameters to extract feature points at different scales.

[0065] In this embodiment, once the filtering parameters are determined, a multi-scale Gaussian filtering method can be used to smooth the point cloud data and highlight feature points at different scales. Gaussian filtering is a common noise reduction method that achieves a smoothing effect by replacing the origin with the weighted average of its neighboring points.

[0066] During this process, Gaussian functions of varying scales (i.e., widths) are applied to the point cloud to capture a variety of structural surface features, from minute to large. After filtering, features previously obscured by noise become more distinct, while unnecessary details are weakened or eliminated. This allows for more accurate identification of geometric feature points associated with rock mass structural surfaces.

[0067] S14: Fuse feature points at different scales to form multi-scale features.

[0068] In this embodiment, the last step is to establish a connection between the feature points extracted at different scales to form a multi-scale feature representation. This is because the rock structure surface exhibits different characteristics at different scales. For example, a crack may have both a macroscopic direction and a microscopic texture. Therefore, simply selecting feature points at a single scale is not enough to fully describe the structural surface. By fusing information from multiple scales, a more complete and robust feature representation can be constructed, which is very critical for subsequent classification and fitting steps. The fusion method can be to directly splice the feature vectors at each scale, or to automatically find the best combination through a learning algorithm. The multi-scale features finally formed not only contain spatial position information, but also shape and texture information at different scales, which greatly improves the ability to intelligently identify rock structure surfaces.

[0069] S2: Input the point cloud data processed by multi-scale filtering and its multi-scale features into a pre-trained classification model to obtain a preliminary classification result of the rock structure surface, wherein the classification model is a deep learning model trained with a point cloud data sample set containing multiple rock structure surface features.

[0070] Specifically, step S2 includes the following steps:

[0071] S21: Constructing an enhanced feature vector by combining the spatial position of the point cloud data and the multi-scale features.

[0072] In this embodiment, in order to enable the classification model to make full use of the information in the point cloud data, an enhanced feature vector is constructed in step S21. First, the spatial position features of each point, i.e., the three-dimensional coordinates (x, y, z), are extracted from the point cloud data. These coordinates provide the exact position of the point in space. Then, the spatial position features of each point and the multi-scale features obtained by step S1 are aligned. The alignment process ensures the correct association between the spatial position and the multi-scale features, so that the position information of each point matches its corresponding multi-scale features. Finally, the aligned spatial position features and multi-scale features are feature spliced ​​to form an enhanced feature vector. This vector contains the spatial position of the point as well as its features such as shape and texture at different scales, thereby providing rich information for subsequent classification tasks.

[0073] More specifically, step S21 includes the following steps:

[0074] a. Extracting the spatial position feature of each point from the point cloud data, wherein the spatial position feature is a three-dimensional coordinate.

[0075] In this embodiment, the spatial position features of each point are extracted from the point cloud data, and these features are expressed as three-dimensional coordinates. Spatial position features are crucial for understanding point cloud data because they directly determine the exact location of the point in three-dimensional space. Three-dimensional coordinates are the most basic component of point cloud data and the foundation for other advanced analyses. In this step, the entire point cloud dataset is traversed, and the three-dimensional coordinates of each point are recorded. The technical effect of this step is to provide precise position information for subsequent processing, allowing for further analysis and processing of the points based on geometric relationships.

[0076] b. Align the spatial position feature of each point with the multi-scale feature.

[0077] In this embodiment, the spatial position features of each point and the multi-scale features obtained by step S1 are aligned. The purpose of the alignment process is to ensure the correct association between the spatial position and the multi-scale features. Since point cloud data may be irregularly distributed and the multi-scale features are calculated independently at different scales, a process is required to ensure that the multi-scale features of each point correspond to its spatial position features. Algorithms such as nearest neighbor search can be used to find the closest spatial position feature and its corresponding multi-scale feature. Through the alignment process, the accuracy of the subsequent splicing operation can be ensured, thereby avoiding classification errors caused by feature mismatch.

[0078] c. Perform feature concatenation on the spatial position features of each aligned point and the multi-scale features to form an enhanced feature vector.

[0079] In this embodiment, the spatial position features and multi-scale features of each aligned point are feature spliced ​​to form an enhanced feature vector. Feature splicing refers to combining two or more different types of features into a new feature representation. Here, the three-dimensional coordinates of each point (spatial position features) are merged with the shape and texture features of the point at different scales (multi-scale features) to create a more powerful feature vector. This enhanced feature vector not only contains the original geometric information, but also incorporates the structured information obtained after filtering, greatly enhancing the expressive power of the feature. Such an enhanced feature vector can provide more contextual information to the deep learning classification model, which helps to improve the model's understanding and recognition accuracy of complex rock structure surface features.

[0080] S22: Input the point cloud data and the enhanced feature vector into a pre-trained classification model, perform classification reasoning, output a classification label for each point, and identify points belonging to rock mass structural surfaces.

[0081] In this embodiment, the point cloud data and enhanced feature vectors are input into a pre-trained deep learning classification model. The classification model is trained on a large number of point cloud data sample sets containing various rock structural surface features, and therefore has the ability to identify different types of rock structural surfaces. In this process, the classification model performs complex calculations and reasoning based on the input point cloud data and enhanced feature vectors, and outputs a classification label for each point. The classification label is used to identify which points belong to the rock structural surface and which do not. The result of this step is a preliminary classification result, in which each point is marked as possibly belonging to or not belonging to a part of the rock structural surface.

[0082] S23: Post-process the marked points belonging to the rock mass structural surface to form a preliminary classification result of the rock mass structural surface.

[0083] In this example, the identified points belonging to rock mass structural surfaces are post-processed to produce a more accurate and coherent preliminary classification of rock mass structural surfaces. This typically involves removing isolated points, filling small holes, and connecting disconnected structural surface segments. The goal is to improve the quality of the classification results, making the structural surface boundaries clearer and smoother, and reducing the possibility of misclassification.

[0084] In addition, morphological operations such as dilation and erosion can be applied to optimize the shape of the structural surface. Through these post-processing operations, a geometric form closer to the actual rock mass structural surface can be obtained, providing a reliable basis for subsequent plane fitting and other analyses.

[0085] S3: Based on the preliminary classification results, the RANSAC algorithm is applied, and the multi-scale features are combined to perform plane fitting in the point cloud data, wherein each fitted plane contains multiple inliers.

[0086] Specifically, step S3 includes the following steps:

[0087] S31: extracting points identified as belonging to rock mass structural surfaces from the preliminary classification results to form a candidate point set, and performing feature enhancement processing in combination with the corresponding multi-scale features.

[0088] In this embodiment, the preliminary classification results have roughly classified the point cloud data using the deep learning model, but these classifications may contain noise or misclassified points. To improve the accuracy of subsequent plane fitting, a purer point set is required as a basis.

[0089] Therefore, the task of this step is to select the points most likely to belong to the rock mass structural surface from the preliminary classification results. Simultaneously, these points are enhanced using the multi-scale features obtained in step S1 to ensure that each point not only contains its original spatial location information but also incorporates shape and texture features at different scales. The technical effect of this step is to provide a high-quality candidate point set, providing a reliable data foundation for the subsequent plane fitting.

[0090] S32: Applying the RANSAC algorithm to perform plane fitting in the candidate point set, wherein the RANSAC algorithm randomly selects a set of points from the candidate point set in an iterative manner to fit a plane, and evaluates the degree of matching between the plane and other points in the candidate point set.

[0091] In this example, RANSAC is an iterative algorithm used to estimate the parameters of a mathematical model, particularly when the data contains a large number of outliers. In this step, the RANSAC algorithm randomly selects a set of points from the candidate point set to attempt to fit a plane, and then evaluates how well this plane matches the rest of the candidate point set. This process is iterative, with each iteration generating a new plane hypothesis and calculating the fitness of that plane for all points.

[0092] Specifically, RANSAC selects a certain number of points (such as three non-collinear points) to define a plane, and then calculates the distance from the remaining points to the plane to determine which points can be considered "inliers", that is, points that meet the plane hypothesis. Through multiple iterations, RANSAC is able to find the optimal plane model even if there is a lot of noise or outliers in the data. The technical effect of this step is to find the plane model that best represents the rock mass structural surface through iterative optimization, thereby improving the accuracy and robustness of the fitting results.

[0093] S33: In each iteration, multi-scale features are combined to evaluate the quality of the fitted plane.

[0094] In this embodiment, the quality of the fitted plane is evaluated in combination with multi-scale features at each iteration. Because rock mass structural surfaces may exhibit different characteristics at different scales, when evaluating plane quality, in addition to considering the distance from the point to the plane, it is also necessary to combine multi-scale features to determine the rationality of the plane.

[0095] For example, by checking whether points on the fitted plane have similar multi-scale characteristics, or whether the plane's orientation is consistent with the normal direction of surrounding points, this comprehensive assessment method can more accurately identify planes that conform to rock mass structural surface characteristics and avoid misjudgments caused by single geometric features.

[0096] S34: setting an inlier threshold, and obtaining inliers that belong to the fitted plane, wherein the inliers refer to points whose distance from the fitted plane is less than the inlier threshold.

[0097] In this example, an inlier threshold is set and inliers are obtained that fall within the fitted plane. Inliers are those points whose distance from the fitted plane is less than the set threshold. The selection of this threshold depends on an understanding of the characteristics of the rock mass structural plane and experimental verification, and typically needs to be adjusted based on specific project requirements.

[0098] By setting a reasonable inlier threshold, points that do not conform to the plane assumption can be effectively filtered out, thereby improving the purity of the fitting results.

[0099] S35: After each iteration, the quality of the currently fitted plane is evaluated according to the number and quality of the interior points, and the best fitted plane is selected as the result of the current iteration.

[0100] In this embodiment, evaluation criteria include the number of inliers, the uniformity of their distribution, and the smoothness of the plane. By comparing the fitting results from different iterations, the plane that best reflects the characteristics of the rock mass structural surface is selected as the final fitting result. This step ensures that the final plane fitting result is not only geometrically reasonable but also closely resembles the actual rock mass structural surface in terms of feature expression, providing a reliable basis for subsequent analysis.

[0101] S4: For each plane obtained by fitting, a seed point is selected from the multiple internal points, and combined with the growth criterion defined based on multi-scale features, the region growing method is used to gradually add neighboring points that meet the growth criterion to the current plane starting from the seed point to improve the boundary of the rock structure surface.

[0102] Specifically, step S4 includes the following steps:

[0103] S41: Select one or more points as seed points from the multiple interior points contained in the plane obtained by each fitting.

[0104] In this embodiment, the selection of seed points is crucial for the subsequent region growing process because they are the starting points of growth. Ideal seed points should be located on the rock mass structural surface and have a high confidence level, that is, the probability of these points being correctly classified as structural surfaces is high.

[0105] To this end, we can select those points that are internal points in the RANSAC algorithm and have the highest degree of match with the fitting plane as seed points. In addition, we can also consider selecting multiple seed points distributed in different positions to ensure that the growth process can cover the entire structure surface.

[0106] S42: Based on the multi-scale features, a growth criterion is defined for determining whether a neighboring point should be added to the current plane, wherein the growth criterion includes the distance between the neighboring point and the seed point, the consistency of the normal direction, and the similarity of the multi-scale features.

[0107] In this embodiment, a growth criterion is defined based on multi-scale features to determine whether a neighboring point should be added to the current plane. The growth criterion is the core of the region growing method, which determines which neighboring points can be considered as part of the structural surface and added to the current plane. The growth criterion generally includes the following aspects:

[0108] The distance between neighboring points and the seed point: The closer the distance is, the more likely it is that the points belong to the same structural surface. Therefore, a maximum distance threshold can be set, and points exceeding this threshold will not be considered.

[0109] Normal Direction Consistency: Rock mass structural surfaces typically have relatively consistent normal directions. Therefore, if the normal direction of a neighboring point is similar to that of the seed point or the current plane, the point is more likely to belong to the same structural surface. Consistency can be assessed by calculating the angle between the two normals and setting an angle threshold.

[0110] Similarity of multi-scale features: Because rock mass structural surfaces exhibit different characteristics at different scales, the similarity of multi-scale features is also an important criterion. For example, if the shape and texture characteristics of neighboring points at multiple scales are similar to those of the seed point, then the points are more likely to belong to the same structural surface. This can be assessed by comparing the distance or similarity between multi-scale feature vectors.

[0111] By comprehensively considering the above three factors, a set of strict growth criteria can be defined to ensure that only points that truly belong to the structural surface will be added to the current plane.

[0112] S43: Based on the selected seed point, the region growing method is used to gradually add neighboring points that meet the growth criteria to the current plane, and each time a point is added, the matching degree between the newly added point and the current plane is re-evaluated.

[0113] In this embodiment, based on the selected seed point, its neighboring points are searched layer by layer, and it is determined whether these neighboring points should be added to the current plane according to the predefined growth criteria. Each time a new point is added, the parameters of the current plane (such as the normal direction, the center point, etc.) are recalculated, and the degree of matching between the newly added point and the updated plane is evaluated. If a neighboring point meets the growth criteria and has a high degree of matching with the current plane, it is added to the current plane; otherwise, the point is not added. This process is iterative until there are no more neighboring points that meet the growth criteria. By gradually expanding, the boundary of the rock mass structural surface is gradually improved, while maintaining the consistency of the geometric characteristics and multi-scale features of the structural surface.

[0114] S44: Through multiple iterations and growth, until no adjacent points that meet the growth criteria can be found.

[0115] In this embodiment, multiple iterations and growth are performed until no more adjacent points that meet the growth criteria can be found. In this process, as more and more points are added to the current plane, the boundary of the structural surface gradually becomes complete and clear. In order to ensure that the termination conditions of the growth process are reasonable, some additional termination conditions are set, including but not limited to the maximum number of iterations or the maximum growth range. When these conditions are reached, the growth process can be stopped even if there are still points that meet the growth criteria. This ensures that the growth process will neither over-expand nor terminate prematurely, thereby obtaining a complete and accurate boundary of the rock mass structural surface.

[0116] S5: Extract the occurrence parameters, spacing and ductility of the rock mass structural planes from the plane processed by the region growing method.

[0117] Specifically, step S5 includes the following steps:

[0118] S51: Smoothing the plane processed by the region growing method.

[0119] In this embodiment, because the region growing method may introduce some small geometric irregularities or noise, especially in boundary areas, these planes need to be smoothed before calculating the occurrence parameters to improve data quality. Smoothing can reduce small surface fluctuations, making subsequent parameter calculations more accurate and stable. Smoothing methods include but are not limited to moving average filtering or bilateral filtering.

[0120] S52: Calculate the occurrence parameters of the rock mass structural surface using the smoothed plane, wherein the occurrence parameters include dip, inclination, and strike.

[0121] In this example, the attitude parameter is an important indicator for describing the spatial posture of the rock mass structural surface, which is crucial for understanding the mechanical properties and stability of the rock mass. The specific calculation process is as follows:

[0122] Inclination: This refers to the angle between a structural plane and the horizontal. This angle can be determined by calculating the angle between the normal vector at each point and the vertical. For a given plane, the inclination is a constant.

[0123] Inclination: refers to the downward tilt of a structure, usually expressed as an angle ranging from 0° to 360°. It can be determined by calculating the projection direction of the normal vector of the structure surface onto the horizontal plane.

[0124] Strike: refers to the direction of projection of the intersection line of the structural surface onto the horizontal plane, usually expressed as an angle. It can be determined by calculating the projection direction of the perpendicular line of the structural surface normal vector onto the horizontal plane.

[0125] S53: Between the confirmed rock mass structural surfaces, measure the distance between adjacent structural surfaces, wherein the distance includes a vertical distance and a horizontal distance.

[0126] In this embodiment, the spacing between adjacent structural surfaces needs to be measured, including vertical and horizontal spacing. Vertical spacing refers to the shortest distance between two parallel structural surfaces, usually measured in a direction perpendicular to the structural surfaces; horizontal spacing refers to the shortest distance between two structural surfaces in the horizontal direction.

[0127] To accurately measure these spacings, algorithms including, but not limited to, nearest neighbor search can be used. The measurement results can provide insights into the distribution density and arrangement of structural planes, which is very useful for assessing the overall stability of the rock mass and potential sliding risks.

[0128] S54: Obtain the ductility of the rock mass structural surface by calculating the length, width and connection parameters of the rock mass structural surface with other structural surfaces.

[0129] In this example, the length and width of a structural plane can be directly measured from its boundaries, while connectivity parameters reflect the relationships between the structural plane and other structural planes. For example, if two structural planes intersect, the length and angle of their intersection are also important connectivity parameters. Through these calculations, we can fully understand the morphological characteristics of the structural planes and their position and function in the network.

[0130] Example 2

[0131] The present invention also provides a rock mass structure surface intelligent recognition system based on three-dimensional laser scanning, which is used to execute the rock mass structure surface intelligent recognition method based on three-dimensional laser scanning. Figure 2 As shown, the system includes:

[0132] The point cloud data multi-scale filtering module 100 is used to collect point cloud data of the rock surface and perform multi-scale filtering based on the density distribution of the point cloud data to obtain multi-scale features.

[0133] The rock structure surface preliminary classification module 200 is used to input the point cloud data and its multi-scale features that have been processed by multi-scale filtering into a pre-trained classification model to obtain preliminary classification results of the rock structure surface, wherein the classification model is a deep learning model trained with a point cloud data sample set containing multiple rock structure surface features.

[0134] The RANSAC plane fitting module 300 is used to apply the RANSAC algorithm based on the preliminary classification results and perform plane fitting in the point cloud data in combination with the multi-scale features, wherein each fitted plane contains multiple inliers.

[0135] The region growing method boundary improvement module 400 is used to select a seed point from the multiple internal points for each plane obtained by fitting, and combine the growth criteria defined based on multi-scale features to use the region growing method to gradually add neighboring points that meet the growth criteria to the current plane starting from the seed point to improve the boundary of the rock structure surface.

[0136] The structural surface parameter extraction module 500 is used to extract the occurrence parameters, spacing and ductility of the rock mass structural surface from the plane processed by the region growing method.

[0137] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0138] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0139] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

Claims

1. An intelligent identification method for rock mass structural surfaces based on 3D laser scanning, characterized in that: The method comprises the following steps: Collecting point cloud data of the rock mass surface, and performing multi-scale filtering processing based on the density distribution of the point cloud data to obtain multi-scale features; Inputting the multi-scale filtered point cloud data and its multi-scale features into a pre-trained classification model to obtain a preliminary classification result of the rock mass structural surface, wherein the classification model is a deep learning model trained on a sample set of point cloud data containing various rock mass structural surface features; Based on the preliminary classification results, applying the RANSAC algorithm and combining the multi-scale features to perform plane fitting in the point cloud data, wherein each fitted plane contains multiple inliers; For each fitted plane, a seed point is selected from the multiple interior points, and a growth criterion defined based on multi-scale features is combined to gradually add neighboring points that meet the growth criterion to the current plane starting from the seed point using a region growing method to improve the boundary of the rock mass structural surface; The occurrence parameters, spacing and ductility of the rock mass structural planes are extracted from the plane processed by the region growing method.

2. The method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to claim 1, characterized in that: The point cloud data of the rock surface is collected, and multi-scale filtering is performed based on the density distribution of the point cloud data to obtain multi-scale features, including: Calculating the local density of each point in the point cloud data to determine the density distribution of the point cloud data; Determining filtering parameters of different scales according to the density distribution of the point cloud data, wherein the filtering parameters include a filtering window size and a threshold; Applying a multi-scale Gaussian filtering method to process the point cloud data according to determined filtering parameters to extract feature points at different scales; Feature points at different scales are fused to form multi-scale features.

3. The method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to claim 2, characterized in that: The point cloud data processed by multi-scale filtering and its multi-scale features are input into the pre-trained classification model to obtain the preliminary classification results of the rock mass structural surface, including: Combining the spatial position of the point cloud data and the multi-scale features to construct an enhanced feature vector; Inputting the point cloud data and the enhanced feature vector into a pre-trained classification model, performing classification reasoning, outputting a classification label for each point, and identifying points belonging to rock mass structural surfaces; The marked points belonging to the rock mass structural surface are post-processed to form the preliminary classification results of the rock mass structural surface.

4. The method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to claim 3, characterized in that: The step of combining the spatial position of the point cloud data and the multi-scale features to construct an enhanced feature vector includes: Extracting a spatial position feature of each point from the point cloud data, wherein the spatial position feature is a three-dimensional coordinate; Aligning the spatial position feature of each point with the multi-scale feature; The spatial position features of each aligned point and the multi-scale features are concatenated to form an enhanced feature vector.

5. The method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to claim 4, characterized in that: Based on the preliminary classification results, the RANSAC algorithm is applied and the multi-scale features are combined to perform plane fitting on the point cloud data, wherein each fitted plane contains multiple inliers, including: Extracting points identified as belonging to rock mass structural surfaces from the preliminary classification results to form a candidate point set, and performing feature enhancement processing in combination with the corresponding multi-scale features; In the candidate point set, a RANSAC algorithm is applied to perform plane fitting, wherein the RANSAC algorithm randomly selects a set of points from the candidate point set in an iterative manner to fit a plane, and evaluates the degree of matching between the plane and other points in the candidate point set; In each iteration, multi-scale features are combined to evaluate the quality of the fitted plane; Setting an inlier threshold and obtaining inliers within the fitted plane, wherein the inliers refer to points whose distance from the fitted plane is less than the inlier threshold; After each iteration, the quality of the currently fitted plane is evaluated based on the number and quality of the interior points, and the best fitted plane is selected as the result of the current iteration.

6. The method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to claim 5, characterized in that: For each fitted plane, a seed point is selected from the multiple interior points, and a growth criterion defined based on multi-scale features is combined to gradually add neighboring points that meet the growth criterion to the current plane starting from the seed point using a region growing method to improve the boundary of the rock mass structural surface, including: Select one or more points as seed points from the multiple interior points contained in the plane obtained by each fitting; Based on the multi-scale features, a growth criterion is defined for determining whether a neighboring point should be added to the current plane, wherein the growth criterion includes a distance between the neighboring point and the seed point, consistency of the normal direction, and similarity of the multi-scale features; Based on the selected seed point, the region growing method is used to gradually add neighboring points that meet the growth criteria to the current plane. In the process of adding points each time, the matching degree between the newly added points and the current plane is re-evaluated. Through multiple iterations and growth, until no adjacent points that meet the growth criteria can be found.

7. The method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to claim 6, characterized in that: The extraction of the occurrence parameters, spacing and ductility of the rock mass structural plane from the plane processed by the region growing method includes: Smoothing the plane processed by the region growing method; Calculating the occurrence parameters of the rock mass structural surface using the smoothed plane, wherein the occurrence parameters include dip, inclination and strike; Between confirmed rock mass structural surfaces, measuring the spacing between adjacent structural surfaces, wherein the spacing includes vertical spacing and horizontal spacing; The ductility of the rock mass structural surface is obtained by calculating the length, width and connection parameters of the rock mass structural surface with other structural surfaces.

8. A system for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning, for executing the method for intelligent identification of rock mass structural surfaces based on three-dimensional laser scanning according to any one of claims 1 to 7, characterized in that: The system comprises: A point cloud data multi-scale filtering module is used to collect point cloud data of the rock surface and perform multi-scale filtering based on the density distribution of the point cloud data to obtain multi-scale features; A rock mass structural surface preliminary classification module is used to input the point cloud data and its multi-scale features processed by multi-scale filtering into a pre-trained classification model to obtain preliminary classification results of the rock mass structural surface, wherein the classification model is a deep learning model trained with a sample set of point cloud data containing multiple rock mass structural surface features; A RANSAC plane fitting module is used to apply the RANSAC algorithm based on the preliminary classification results and perform plane fitting in the point cloud data in combination with the multi-scale features, wherein each fitted plane contains multiple inliers; A region growing method boundary improvement module is used to select a seed point from the multiple interior points for each fitted plane, and use the region growing method to gradually add neighboring points that meet the growth criteria to the current plane starting from the seed point in combination with the growth criteria defined based on multi-scale features, so as to improve the boundary of the rock mass structural surface; The structural surface parameter extraction module is used to extract the occurrence parameters, spacing and ductility of the rock structural surface from the plane processed by the region growing method.

Images