Method, apparatus, equipment and storage medium for identifying rock mass structural surfaces based on point cloud.

Abstract

This invention provides a method, apparatus, device, and storage medium for identifying rock mass structural surfaces based on point clouds, relating to the field of data processing technology. The method includes: processing dense three-dimensional point cloud data of a rock mass to obtain processed three-dimensional point cloud data; uniformly downsampling the processed three-dimensional point cloud data to obtain sampled processed point cloud data and unsampled processed point cloud data; training a neural network model based on the sampled processed point cloud data to obtain a rock mass structural surface identification model; inputting the unsampled processed point cloud data into the rock mass structural surface identification model to obtain a corresponding unsampled processed point cloud label set; and obtaining the rock mass structural surface attitude based on the unsampled processed point cloud label set. This invention improves the identification speed while maintaining the accuracy of rock mass structural surface identification.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method, apparatus, device, and storage medium for identifying rock mass structural surfaces based on point clouds. Background Technology

[0002] Rock mass structural plane identification is of paramount importance in fields such as geology and engineering. Structural planes refer to discontinuities in rocks, such as joints, fissures, and faults, which have a decisive influence on the physical and mechanical properties of rock masses. Current methods for automatically identifying rock structural planes from point cloud data are overly complex, often requiring long processing times for massive datasets, and are prone to errors. Summary of the Invention

[0003] This invention solves the problem of improving the recognition rate while ensuring the accuracy of rock mass structure surface identification.

[0004] To address the aforementioned problems, this invention provides a method, apparatus, device, and storage medium for identifying rock mass structural surfaces based on point clouds.

[0005] In a first aspect, the present invention provides a method for identifying rock mass structural surfaces based on point clouds, comprising:

[0006] The dense three-dimensional point cloud data of the rock mass is processed to obtain the processed three-dimensional point cloud data;

[0007] The processed 3D point cloud data and the unsampled point cloud data are obtained by uniformly downsampling the processed 3D point cloud data.

[0008] A rock mass structure surface recognition model is obtained by training a neural network model based on the sampled point cloud data.

[0009] The unsampled point cloud data is input into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud label set;

[0010] The rock mass structural surface attitude is obtained based on the unsampled point cloud label set and the unsampled point cloud data.

[0011] Optionally, the method of training a neural network model based on the sampled point cloud data to obtain a rock mass structure surface recognition model includes:

[0012] Based on the agglomerative hierarchical clustering algorithm, the sampled point cloud data is classified to obtain the corresponding sampled point cloud label set.

[0013] Based on the sampled point cloud data and the corresponding sampled point cloud label set, a point cloud data training set and a point cloud data test set are obtained;

[0014] The neural network model is trained using the point cloud data training set until the neural network converges, thus obtaining an initial neural network model.

[0015] The initial neural network model is tested using the point cloud data test set to test the test set accuracy of the initial neural network model. If the test set accuracy does not meet the accuracy requirement, more data is acquired for retraining until the accuracy requirement is met, thus obtaining the rock mass structure surface recognition model.

[0016] The accuracy of the test set includes:

[0017] ,

[0018] Where AC represents the accuracy of the test set. The number of point clouds for accurate classification, The number of point clouds in the point cloud data test set.

[0019] Optionally, the step of classifying the sampled point cloud data to obtain a corresponding sampled point cloud label set includes:

[0020] The similarity metric is obtained by processing the point cloud data through the sampling process.

[0021] The similarity metric includes:

[0022] ,

[0023] in, For the aforementioned similarity metric, Let be the normal vector of the i-th point cloud. Let be the transpose of the normal vector of the j-th point cloud;

[0024] The sampled point cloud data is grouped into structural surface categories based on the similarity metric to obtain the corresponding sampled point cloud label set.

[0025] Optionally, obtaining the rock mass structural plane attitude based on the unsampled point cloud label set and the unsampled point cloud data includes:

[0026] A density-based noisy spatial clustering algorithm is used to separate the unsampled point cloud data according to the unsampled point cloud label set to obtain multiple rock mass structural surfaces;

[0027] Based on the least squares method, the corresponding plane normal vectors are obtained by fitting data to the rock mass structural surfaces respectively.

[0028] The dip angle and dip direction of the rock mass structural surface are obtained from the plane normal vector;

[0029] The occurrence of the rock mass structural planes includes:

[0030] ,

[0031] ,

[0032] in, The dip angle of the rock mass structural surface. Let (A, B, C) be the dip direction of the rock mass structural plane, and (B, C) be the plane normal vector.

[0033] The attitude of the rock mass structural surface is obtained based on the dip angle and dip direction of the rock mass structural surface.

[0034] Optionally, obtaining sampled point cloud data and unsampled point cloud data by uniformly downsampling the processed 3D point cloud data includes:

[0035] The processed 3D point cloud data is divided into multiple 3D meshes with fixed dimensions;

[0036] The center point of each of the three-dimensional meshes is obtained respectively;

[0037] All the center points of the grid are used as the sampled point cloud data;

[0038] The unsampled point cloud data is obtained from the data in the processed 3D point cloud data excluding the sampled point cloud data.

[0039] Optionally, the dense three-dimensional point cloud data of the rock mass includes multiple point clouds. Processing the dense three-dimensional point cloud data of the rock mass yields processed three-dimensional point cloud data, including:

[0040] Based on the KD-tree algorithm, the corresponding neighborhood point sets are obtained from the point cloud respectively;

[0041] The processed 3D point cloud data is obtained by performing principal component analysis on the neighborhood point set.

[0042] Optionally, it also includes:

[0043] Structural surface images are obtained by collecting rock structural surface data;

[0044] Based on the SfM-MVS algorithm, the structural surface image is converted into point cloud data to obtain the dense three-dimensional point cloud data of the rock mass.

[0045] Secondly, the present invention provides a rock mass structure surface identification device based on point cloud, comprising: a data processing module for processing dense three-dimensional point cloud data of rock mass to obtain processed three-dimensional point cloud data;

[0046] The downsampling module is used to obtain sampled point cloud data and unsampled point cloud data by uniformly downsampling the processed 3D point cloud data.

[0047] The model training module is used to train a neural network model based on the sampled point cloud data to obtain a rock mass structure surface recognition model;

[0048] The module for obtaining the unsampled point cloud tag set is used to input the unsampled point cloud data into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud tag set.

[0049] The rock mass structural plane attitude acquisition module is used to obtain the rock mass structural plane attitude based on the unsampled point cloud label set and the unsampled point cloud data.

[0050] Thirdly, the present invention provides an electronic device, including a memory and a processor;

[0051] The memory is used to store computer programs;

[0052] The processor is configured to, when executing the computer program, implement the point cloud-based rock mass structure surface identification method as described in the first aspect.

[0053] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the point cloud-based rock mass structure surface identification method as described in the first aspect.

[0054] The beneficial effects of the point cloud-based rock mass structure surface identification method, apparatus, equipment, and storage medium of the present invention are as follows: By processing dense three-dimensional point cloud data of the rock mass, the accuracy of the data is improved, laying the foundation for subsequent data processing. Uniform downsampling of the processed three-dimensional point cloud data reduces the impact of redundant data on training speed, shortens data processing time, and preserves key geometric features of the rock mass surface, thus improving the accuracy of rock mass structure surface identification. A neural network model is trained using the sampled point cloud data to obtain a rock mass structure surface identification model. The neural network model enhances the structural understanding ability, enabling the model to converge quickly using simplified data. Inputting unsampled point cloud data into the rock mass structure surface identification model yields the corresponding unsampled point cloud label set, further revealing the rock mass structure surface attitude, which improves the identification rate while maintaining the accuracy of rock mass structure surface identification. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating a method for identifying rock mass structure surfaces based on point clouds, according to an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of the structure of a neural network model according to an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of a point cloud-based rock mass structure surface recognition device according to an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0059] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0060] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0061] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0062] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0063] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0064] like Figure 1As shown in the figure, an embodiment of the present invention provides a method for identifying rock mass structure surfaces based on point clouds, comprising:

[0065] Step 110: Process the dense three-dimensional point cloud data of the rock mass to obtain processed three-dimensional point cloud data.

[0066] Specifically, by using drones or high-definition cameras, data on the rock structure surface at the site are collected, structural surface photos are obtained, and point cloud data is converted into dense three-dimensional point cloud data of the rock mass.

[0067] Step 120: Obtain sampled point cloud data and unsampled point cloud data by uniformly downsampling the processed 3D point cloud data.

[0068] Specifically, uniform downsampling is a point cloud simplification method that selects representative points at fixed spatial intervals to remove redundant information. The data obtained from uniform downsampling is used as the sampled point cloud data, and the remaining data in the processed 3D point cloud data is used as the unsampled point cloud data.

[0069] Step 130: Train a neural network model based on the sampled point cloud data to obtain a rock mass structure surface recognition model.

[0070] Specifically, the sampled point cloud data is labeled, and an agglomerative hierarchical clustering algorithm is used to classify the sampled point clouds, achieving automatic labeling of the point clouds. A neural network model is then trained using the labeled data to obtain a rock mass structure surface recognition model.

[0071] Step 140: Input the unsampled point cloud data into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud label set.

[0072] Specifically, the unsampled point cloud data is input into the trained rock mass structure surface recognition model for automatic classification, and the classification labels of all point clouds, i.e. structure surface groups, are obtained. The point clouds can be automatically assigned to different structure surface groups.

[0073] Step 150: Obtain the rock mass structural surface attitude based on the unsampled point cloud label set and the unsampled point cloud data.

[0074] Specifically, in engineering geology and rock mechanics, the attitude of a rock mass structural plane refers to its geometric orientation and dip in three-dimensional space. It is usually uniquely determined by the dip angle and dip direction of the rock mass structural plane.

[0075] In this embodiment, the dense 3D point cloud data of the rock mass is processed to improve data accuracy and lay the foundation for subsequent data processing. Uniform downsampling of the processed 3D point cloud data reduces the impact of redundant data on training speed, shortens data processing time, and preserves key geometric features of the rock mass surface, improving the accuracy of rock mass structural surface identification. A neural network model is trained using the sampled point cloud data to obtain a rock mass structural surface identification model. This neural network model enhances structural understanding capabilities, enabling the model to converge quickly using simplified data. Inputting unsampled point cloud data into the rock mass structural surface identification model yields the corresponding unsampled point cloud label set, further revealing the rock mass structural surface attitude. This approach improves the identification speed while maintaining accuracy.

[0076] Optionally, the step of training a neural network model based on the sampled point cloud data to obtain a rock mass structure surface recognition model includes:

[0077] Based on the agglomerative hierarchical clustering algorithm, the sampled point cloud data is classified to obtain the corresponding sampled point cloud label set.

[0078] Based on the sampled point cloud data and the corresponding sampled point cloud label set, a point cloud data training set and a point cloud data test set are obtained;

[0079] The neural network model is trained using the point cloud data training set until the neural network converges, thus obtaining an initial neural network model.

[0080] The initial neural network model is tested using the point cloud data test set to test the test set accuracy of the initial neural network model. If the test set accuracy does not meet the accuracy requirement, more data is acquired for retraining until the accuracy requirement is met, thus obtaining the rock mass structure surface recognition model.

[0081] The accuracy of the test set includes:

[0082] ,

[0083] Where AC represents the accuracy of the test set. The number of point clouds for accurate classification, The number of point clouds in the point cloud data test set.

[0084] Specifically, in combination Figure 2As shown, the neural network model includes a 1×6 input layer, five convolutional layers, two 4096 fully connected layers, and one output layer. The 1×6 input layer, Conv1, takes point cloud coordinates and point cloud normal vectors as input. The five convolutional layers, Conv2, Conv3, Conv4, Conv5, and Conv6, have 96, 256, 384, 384, and 256 output channels respectively, with a kernel size of 1×6. There are two 4096 fully connected layers, and the output layer outputs the label of the point cloud.

[0085] In some more specific embodiments, hyperparameters of the neural network model are set, including activation function, batch size, learning rate, epoch, and optimization algorithm. Here, batch size represents the number of samples used in each iteration; learning rate is the rate at which the model updates weights in each iteration; epoch refers to the number of times the network processes the entire training dataset; and optimization algorithm is a model used to adjust model parameters to minimize the loss function. The model is fine-tuned using an iterative approach to optimize the training process until the training accuracy exceeds 90%. The sampled dataset is divided into a point cloud training set and a point cloud test set in an 8:2 ratio. The point cloud training set is input into the neural network model for training, and the point cloud test set is used to test the model, verifying the test set accuracy (AC). When the test set accuracy exceeds 90%, the trained rock mass structure surface recognition model is saved.

[0086] In this optional embodiment, an agglomerative hierarchical clustering algorithm is used to classify the sampled point cloud data to obtain corresponding point cloud labels. The point cloud data with cluster labels is then divided into training and testing sets for subsequent training and validation of the neural network model.

[0087] Optionally, the step of classifying the sampled point cloud data to obtain a corresponding sampled point cloud label set includes:

[0088] The similarity metric is obtained by processing the point cloud data through the sampling process.

[0089] The similarity metric includes:

[0090] ,

[0091] in, For the aforementioned similarity metric, Let be the normal vector of the i-th point cloud. Let be the transpose of the normal vector of the j-th point cloud;

[0092] The sampled point cloud data is grouped into structural surface categories based on the similarity metric to obtain the corresponding sampled point cloud label set.

[0093] Specifically, Agglomerative Hierarchical Clustering (AHC) is a classic unsupervised clustering method, belonging to bottom-up hierarchical clustering. It constructs a tree-like structure by continuously merging highly similar sample points or clusters, thus revealing the inherent hierarchical structure of the data. Using AHC to classify sampled point clouds achieves automatic labeling of the point clouds, treating each object as a cluster, and then merging these clusters into larger clusters until a certain termination condition is met. Currently, using the square of the cosine value between unit normal vectors is the most efficient way to cluster the attitude of rock mass structural surfaces. Determine how many categories or structural surface groups the point cloud should be divided into, setting the number of clusters M, where M is an integer representing the final number of clusters. Calculate the cosine distance of all point clouds, i.e., the similarity measure, where:

[0094] ,

[0095] ,

[0096] in, For the aforementioned similarity metric, Let be the normal vector of the i-th point cloud. Let be the transpose of the normal vector of the j-th point cloud. The acute angle between the normal vectors of two point clouds is defined. Point clouds with small distance differences are automatically assigned to the same label L (L=1,2,...,M), initially grouping points belonging to the same structural surface into one category, forming several candidate regions. These labeling results are then manually reviewed and corrected by professional technicians. For example, for regions that are mis-clustered due to sparse or occluded point clouds, operators can manually adjust the labels of the point clouds to ensure that the point set corresponding to each label accurately represents an independent rock mass structural surface.

[0097] In this optional embodiment, point clouds are classified based on agglomerative hierarchical clustering algorithm. Agglomerative hierarchical clustering has good topological structure preservation ability and is suitable for point cloud data classification tasks with complex rock surface structures and diverse morphologies.

[0098] Optionally, obtaining the rock mass structural plane attitude based on the unsampled point cloud label set and the unsampled point cloud data includes:

[0099] A density-based noisy spatial clustering algorithm is used to separate the unsampled point cloud data according to the unsampled point cloud label set to obtain multiple rock mass structural surfaces;

[0100] Based on the least squares method, the corresponding plane normal vectors are obtained by fitting data to the rock mass structural surfaces respectively.

[0101] The dip angle and dip direction of the rock mass structural surface are obtained from the plane normal vector;

[0102] The occurrence of the rock mass structural planes includes:

[0103] ,

[0104] ,

[0105] in, The dip angle of the rock mass structural surface. Let (A, B, C) be the dip direction of the rock mass structural plane, and (B, C) be the plane normal vector.

[0106] The attitude of the rock mass structural surface is obtained based on the dip angle and dip direction of the rock mass structural surface.

[0107] Specifically, after obtaining the category label for each point, to further distinguish structural surfaces belonging to the same category but located at different positions (e.g., multiple parallel joint surfaces), cluster analysis needs to be performed on the points of each category. The density-based noisy spatial clustering (DBSCAN) algorithm is used to separate individual rock mass structural surfaces, and then the attitude of each structural surface is further obtained. The normal vector of the rock mass structural surface is calculated based on least squares fitting. Assuming the plane equation is Ax + By + Cz + D = 0, any given set of 3D point cloud data... The goal is to find suitable coefficients (A, B, C, D) that minimize the sum of the squared distances from all points to the plane. This plane is then considered a satisfactory plane. The plane's normal vector is (A, B, C). The attitude of a rock mass structural plane is used to represent its orientation in three-dimensional space, typically described by its dip direction and dip angle. The dip direction indicates the maximum downward tilt perpendicular to the strike line, pointing to the downward-dipping side of the structural plane. The dip angle indicates the maximum angle of inclination of the structural plane relative to the horizontal plane, usually expressed in degrees, ranging from 0° (completely horizontal) to 90° (completely vertical). The attitude of a rock mass structural plane can be described by the ratio of its dip direction to its dip angle.

[0108] Optionally, obtaining sampled point cloud data and unsampled point cloud data by uniformly downsampling the processed 3D point cloud data includes:

[0109] The processed 3D point cloud data is divided into multiple 3D meshes with fixed dimensions;

[0110] The center point of each of the three-dimensional meshes is obtained respectively;

[0111] All the center points of the grid are used as the sampled point cloud data;

[0112] The unsampled point cloud data is obtained from the data in the processed 3D point cloud data excluding the sampled point cloud data.

[0113] Specifically, uniform downsampling of dense point clouds is a point cloud data processing technique. Its purpose is to reduce the number of points in the point cloud while maintaining its overall geometry and distribution characteristics, thereby improving data processing efficiency or reducing storage requirements. The specific implementation is as follows: the point cloud is divided into a fixed-size three-dimensional grid (voxels); within each voxel, only one representative point (such as the center point or mean point) is retained.

[0114] Optionally, the dense three-dimensional point cloud data of the rock mass includes multiple point clouds. Processing the dense three-dimensional point cloud data of the rock mass yields processed three-dimensional point cloud data, including:

[0115] Based on the KD-tree algorithm, the corresponding neighborhood point sets are obtained from the point cloud respectively;

[0116] The processed 3D point cloud data is obtained by performing principal component analysis on the neighborhood point set.

[0117] Specifically, the KD-tree method is used to compute the neighborhood point set of any point cloud. A KD-tree is a data structure where each node represents a hyperrectangular region in k-dimensional space. It can be viewed as a generalization of the Binary Search Tree (BST) in multidimensional space. The process of constructing a KD-tree based on point cloud data recursively divides the space into two parts until a stopping condition is met. For a given point, the neighborhood point set is quickly obtained using the constructed KD-tree. Based on the neighborhood point set of the point cloud, its covariance matrix is ​​constructed. PCA calculation eigenvalues and eigenvectors Calculate each element in the eigenvector separately. The normal vector at a point is determined by comparing its modulus with the corresponding modulus values. If ,but That is, the normal vector of that point. .

[0118] .

[0119] Optionally, it also includes:

[0120] Structural surface images are obtained by collecting rock structural surface data;

[0121] Based on the SfM-MVS algorithm, the structural surface image is converted into point cloud data to obtain the dense three-dimensional point cloud data of the rock mass.

[0122] Specifically, SfM (Structure from Motion) and MVS (Multi-View Stereo) are two techniques for reconstructing 3D models from a sequence of 2D images. They are often used together to generate high-quality 3D point clouds or mesh models from a series of uncalibrated 2D images.

[0123] like Figure 3 As shown in the figure, an embodiment of the present invention provides a rock mass structure surface identification device based on point cloud, comprising:

[0124] Data processing module 10 is used to process dense three-dimensional point cloud data of rock mass to obtain processed three-dimensional point cloud data;

[0125] The downsampling module 20 is used to obtain sampled point cloud data and unsampled point cloud data by uniformly downsampling the processed three-dimensional point cloud data.

[0126] Model training module 30 is used to train a neural network model based on the sampled point cloud data to obtain a rock mass structure surface recognition model;

[0127] The unsampled point cloud tag set acquisition module 40 is used to input the unsampled point cloud data into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud tag set.

[0128] The rock mass structural plane attitude acquisition module 50 is used to obtain the rock mass structural plane attitude based on the unsampled point cloud label set and the unsampled point cloud data.

[0129] The point cloud-based rock mass structure surface recognition device of this embodiment is used to implement the point cloud-based rock mass structure surface recognition method as described above. Its advantages over the prior art are the same as those of the point cloud-based rock mass structure surface recognition method over the prior art, and will not be repeated here.

[0130] Optionally, the model training module 30 is specifically used to: classify the sampled point cloud data based on the agglomerative hierarchical clustering algorithm to obtain the corresponding sampled point cloud label set;

[0131] Based on the sampled point cloud data and the corresponding sampled point cloud label set, a point cloud data training set and a point cloud data test set are obtained;

[0132] The neural network model is trained using the point cloud data training set until the neural network converges, thus obtaining an initial neural network model.

[0133] The initial neural network model is tested using the point cloud data test set to test the test set accuracy of the initial neural network model. If the test set accuracy does not meet the accuracy requirement, more data is acquired for retraining until the accuracy requirement is met, thus obtaining the rock mass structure surface recognition model.

[0134] The accuracy of the test set includes:

[0135] ,

[0136] Where AC represents the accuracy of the test set. The number of point clouds for accurate classification, The number of point clouds in the point cloud data test set.

[0137] Optionally, the model training module 30 is specifically used to: obtain a similarity measure through the sampled point cloud data;

[0138] The similarity metric includes:

[0139] ,

[0140] in, For the aforementioned similarity metric, Let be the normal vector of the i-th point cloud. Let be the transpose of the normal vector of the j-th point cloud;

[0141] The sampled point cloud data is grouped into structural surface categories based on the similarity metric to obtain the corresponding sampled point cloud label set.

[0142] Optionally, the rock mass structure surface attitude acquisition module 50 is specifically used for: using a density-based noisy spatial clustering algorithm to separate the unsampled point cloud data according to the unsampled point cloud label set to obtain multiple rock mass structure surfaces;

[0143] Based on the least squares method, the corresponding plane normal vectors are obtained by fitting data to the rock mass structural surfaces respectively.

[0144] The dip angle and dip direction of the rock mass structural surface are obtained from the plane normal vector;

[0145] The occurrence of the rock mass structural planes includes:

[0146] ,

[0147] ,

[0148] in, The dip angle of the rock mass structural surface. Let (A, B, C) be the dip direction of the rock mass structural plane, and (B, C) be the plane normal vector.

[0149] The attitude of the rock mass structural surface is obtained based on the dip angle and dip direction of the rock mass structural surface.

[0150] Optionally, the downsampling module 20 is specifically used to: divide the processed 3D point cloud data into multiple 3D meshes with a fixed size;

[0151] The center point of each of the three-dimensional meshes is obtained respectively;

[0152] All the center points of the grid are used as the sampled point cloud data;

[0153] The unsampled point cloud data is obtained from the data in the processed 3D point cloud data excluding the sampled point cloud data.

[0154] Optionally, the data processing module 10 is specifically used to: obtain the corresponding neighborhood point sets based on the point cloud according to the KD tree algorithm;

[0155] The processed 3D point cloud data is obtained by performing principal component analysis on the neighborhood point set.

[0156] Optionally, the point cloud-based rock mass structure surface recognition device further includes a data acquisition module, which is used to: obtain a structure surface image by collecting rock structure surface data;

[0157] Based on the SfM-MVS algorithm, the structural surface image is converted into point cloud data to obtain the dense three-dimensional point cloud data of the rock mass.

[0158] like Figure 4 As shown, an electronic device 400 provided in this embodiment of the invention includes a memory 410 and a processor 420; the memory 410 is used to store a computer program; the processor 420 is used to implement the point cloud-based rock mass structure surface recognition method as described above when the computer program is executed.

[0159] Alternatively, an electronic device 400 includes a memory 410 and a processor 420 coupled to the memory 410; the memory 410 is configured to store a computer program; and the processor 420 is configured to perform the following operations when the computer program is executed:

[0160] The dense three-dimensional point cloud data of the rock mass is processed to obtain the processed three-dimensional point cloud data;

[0161] The processed 3D point cloud data and the unsampled point cloud data are obtained by uniformly downsampling the processed 3D point cloud data.

[0162] A rock mass structure surface recognition model is obtained by training a neural network model based on the sampled point cloud data.

[0163] The unsampled point cloud data is input into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud label set;

[0164] The rock mass structural surface attitude is obtained based on the unsampled point cloud label set and the unsampled point cloud data.

[0165] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the point cloud-based rock mass structure surface identification method described above.

[0166] Alternatively, a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the following operations:

[0167] The dense three-dimensional point cloud data of the rock mass is processed to obtain the processed three-dimensional point cloud data;

[0168] The processed 3D point cloud data and the unsampled point cloud data are obtained by uniformly downsampling the processed 3D point cloud data.

[0169] A rock mass structure surface recognition model is obtained by training a neural network model based on the sampled point cloud data.

[0170] The unsampled point cloud data is input into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud label set;

[0171] The rock mass structural surface attitude is obtained based on the unsampled point cloud label set and the unsampled point cloud data.

[0172] The present invention will now be described an electronic device 400 that can serve as a server or client of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. Electronic device 400 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 400 can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0173] Electronic device 400 includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or a computer program loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0174] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. In this application, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention according to actual needs. Furthermore, the functional units in the various embodiments of the present invention 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 integrated units can be implemented in hardware or as software functional units.

[0175] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for identifying rock mass structural surfaces based on point clouds, characterized in that, include: The dense three-dimensional point cloud data of the rock mass is processed to obtain the processed three-dimensional point cloud data; The processed 3D point cloud data and the unsampled point cloud data are obtained by uniformly downsampling the processed 3D point cloud data. A rock mass structure surface recognition model is obtained by training a neural network model based on the sampled point cloud data, including: Based on the agglomerative hierarchical clustering algorithm, the sampled point cloud data is classified to obtain the corresponding sampled point cloud label set. Based on the sampled point cloud data and the corresponding sampled point cloud label set, a point cloud data training set and a point cloud data test set are obtained; The neural network model is trained using the point cloud data training set until the neural network converges, thus obtaining an initial neural network model. The initial neural network model is tested using the point cloud data test set to test the test set accuracy of the initial neural network model. If the test set accuracy does not meet the accuracy requirement, more data is acquired for retraining until the accuracy requirement is met, thus obtaining the rock mass structure surface recognition model. The accuracy of the test set includes: ， Where AC represents the accuracy of the test set. The number of point clouds for accurate classification, The number of point clouds in the point cloud data test set; The unsampled point cloud data is input into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud label set; The rock mass structural plane attitude is obtained based on the unsampled point cloud label set and the unsampled point cloud data, including: A density-based noisy spatial clustering algorithm separates the unsampled point cloud data into multiple categories based on the unsampled point cloud label set, and performs cluster analysis on the point cloud data of each category to obtain a single rock mass structural surface. Based on the least squares method, the corresponding plane normal vectors are obtained by fitting data to the rock mass structural surfaces respectively. The dip angle and dip direction of the rock mass structural surface are obtained from the plane normal vector; The occurrence of the rock mass structural planes includes: ， ， in, The dip angle of the rock mass structural surface. Let (A, B, C) be the dip direction of the rock mass structural plane, and (B, C) be the plane normal vector. The attitude of the rock mass structural surface is obtained based on the dip angle and dip direction of the rock mass structural surface.

2. The method for identifying rock mass structural surfaces based on point clouds according to claim 1, characterized in that, The process of classifying the sampled point cloud data to obtain a corresponding sampled point cloud label set includes: The similarity metric is obtained by processing the point cloud data through the sampling process. The similarity metric includes: ， in, For the aforementioned similarity metric, Let be the normal vector of the i-th point cloud. Let be the transpose of the normal vector of the j-th point cloud; The sampled point cloud data is grouped into structural surface categories based on the similarity metric to obtain the corresponding sampled point cloud label set.

3. The method for identifying rock mass structural surfaces based on point clouds according to claim 1, characterized in that, The process of uniformly downsampling the processed 3D point cloud data to obtain sampled point cloud data and unsampled point cloud data includes: The processed 3D point cloud data is divided into multiple 3D meshes with fixed dimensions; The center point of each of the three-dimensional meshes is obtained respectively; All the center points of the grid are used as the sampled point cloud data; The unsampled point cloud data is obtained from the data in the processed 3D point cloud data excluding the sampled point cloud data.

4. The method for identifying rock mass structural surfaces based on point clouds according to claim 1, characterized in that, The dense 3D point cloud data of the rock mass includes multiple point clouds. Processing the dense 3D point cloud data of the rock mass yields processed 3D point cloud data, including: Based on the KD-tree algorithm, the corresponding neighborhood point sets are obtained from the point cloud respectively; The processed 3D point cloud data is obtained by performing principal component analysis on the neighborhood point set.

5. The method for identifying rock mass structural surfaces based on point clouds according to claim 4, characterized in that, Also includes: Structural surface images are obtained by collecting rock structural surface data; Based on the SfM-MVS algorithm, the structural surface image is converted into point cloud data to obtain the dense three-dimensional point cloud data of the rock mass.

6. A rock mass structure surface identification device based on point cloud, characterized in that, include: The data processing module is used to process the dense three-dimensional point cloud data of the rock mass to obtain processed three-dimensional point cloud data. The downsampling module is used to obtain sampled point cloud data and unsampled point cloud data by uniformly downsampling the processed 3D point cloud data. The model training module is used to train a neural network model based on the sampled point cloud data to obtain a rock mass structure surface recognition model, including: Based on the agglomerative hierarchical clustering algorithm, the sampled point cloud data is classified to obtain the corresponding sampled point cloud label set. Based on the sampled point cloud data and the corresponding sampled point cloud label set, a point cloud data training set and a point cloud data test set are obtained; The neural network model is trained using the point cloud data training set until the neural network converges, thus obtaining an initial neural network model. The initial neural network model is tested using the point cloud data test set to test the test set accuracy of the initial neural network model. If the test set accuracy does not meet the accuracy requirement, more data is acquired for retraining until the accuracy requirement is met, thus obtaining the rock mass structure surface recognition model. The accuracy of the test set includes: ， Where AC represents the accuracy of the test set. The number of point clouds for accurate classification, The number of point clouds in the point cloud data test set; The module for obtaining the unsampled point cloud tag set is used to input the unsampled point cloud data into the rock mass structure surface recognition model to obtain the corresponding unsampled point cloud tag set. The rock mass structural plane attitude acquisition module is used to obtain the rock mass structural plane attitude based on the unsampled point cloud label set and the unsampled point cloud data, including: A density-based noisy spatial clustering algorithm separates the unsampled point cloud data into multiple categories based on the unsampled point cloud label set, and performs cluster analysis on the point cloud data of each category to obtain a single rock mass structural surface. Based on the least squares method, the corresponding plane normal vectors are obtained by fitting data to the rock mass structural surfaces respectively. The dip angle and dip direction of the rock mass structural surface are obtained from the plane normal vector; The occurrence of the rock mass structural planes includes: ， ， in, The dip angle of the rock mass structural surface. Let (A, B, C) be the dip direction of the rock mass structural plane, and (B, C) be the plane normal vector. The attitude of the rock mass structural surface is obtained based on the dip angle and dip direction of the rock mass structural surface.

7. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the point cloud-based rock mass structure surface recognition method as described in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the point cloud-based rock mass structure surface identification method as described in any one of claims 1 to 5.

Images