CN116403124A - Structural plane intelligent recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet - Google Patents
Structural plane intelligent recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet Download PDFInfo
- Publication number
- CN116403124A CN116403124A CN202211521091.4A CN202211521091A CN116403124A CN 116403124 A CN116403124 A CN 116403124A CN 202211521091 A CN202211521091 A CN 202211521091A CN 116403124 A CN116403124 A CN 116403124A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- surrounding rock
- pointnet
- data
- dresnet
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000011435 rock Substances 0.000 title claims abstract description 98
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000013135 deep learning Methods 0.000 claims abstract description 32
- 238000012360 testing method Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 20
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000002372 labelling Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 4
- 239000010410 layer Substances 0.000 description 16
- 238000010276 construction Methods 0.000 description 9
- 238000001514 detection method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000011229 interlayer Substances 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000003595 mist Substances 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 239000003381 stabilizer Substances 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an intelligent structural surface identification method of three-dimensional point cloud surrounding rock based on DResNet-PointNet, which comprises the following steps: s1, acquiring point cloud data of a tunnel portal rock mass by using an unmanned aerial vehicle-mounted three-dimensional laser scanner; s2, preprocessing point cloud data, and dividing the point cloud data into training data and test data; s3, training the DResNet-PointNet point cloud deep learning frame through training data to obtain a trained DResNet-PointNet deep learning frame; s4, predicting the structural surface of the point cloud surrounding rock according to the trained DResNet-PointNet deep learning frame, and extracting the structural surface geometric information of the predicted surrounding rock to obtain the structural surface geometric information of the surrounding rock. The method provided by the invention can be simply realized, is flexible to operate, has obvious effect and meets the application requirement.
Description
Technical Field
The invention belongs to the technical field of engineering digitization, and particularly relates to a structural plane intelligent identification method of a three-dimensional point cloud surrounding rock based on DResNet-PointNet.
Background
In recent years, along with the construction of side slopes and project foundations, underground mining, underground treatment libraries of nuclear waste, and rock mass engineering such as a highway, the research on the characterization of joints and mechanical properties is more and more urgent. The research in the aspect has great significance on the safety evaluation of various rock mass engineering, and is also a mainstream trend of the development of the rock mass engineering at present.
At present, the calculation and analysis methods of the stability of the buried rock mass are more, the basic assumption and the theoretical basis of different kinds of calculation methods are different, and the method has different applicability and limitation in the analysis of the stability of the rock mass. The intelligent recognition of the structural surface of the surrounding rock of the rock mass mainly adopts algorithms such as clustering in mathematical statistics, the whole is still in a preliminary research stage, and the algorithms are required to be continuously enriched and improved due to higher requirements on the accuracy and the working efficiency of the algorithms, so that the method has a wide lifting space. With the development of deep learning technology, it is also desirable for three-dimensional point clouds to be able to solve such problems as: classification, identification, segmentation, complementation, registration and the like, so that three-dimensional point cloud classification identification under a deep learning framework has higher engineering adaptability.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a structural plane intelligent identification method of a three-dimensional point cloud surrounding rock based on DResNet-PointNet.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a structural plane intelligent identification method of three-dimensional point cloud surrounding rock based on DResNet-PointNet comprises the following steps:
s1, acquiring point cloud data of a tunnel portal rock mass by using an unmanned aerial vehicle-mounted three-dimensional laser scanner;
s2, preprocessing the point cloud data obtained in the step S1 to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data;
s3, training the DResNet-PointNet point cloud deep learning frame through the training data obtained in the step S2 to obtain a trained DResNet-PointNet deep learning frame;
s4, according to the trained DResNet-PointNet deep learning frame obtained in the step S3, carrying out structural plane prediction of the point cloud surrounding rock on the test data obtained in the step S2, and extracting structural plane geometric information of the predicted surrounding rock to obtain structural plane geometric information of the surrounding rock.
Preferably, step S2 comprises the steps of:
s21, performing gridding treatment on the point cloud data obtained in the step S1, and then calculating each node normal vector in the treated point cloud data to obtain each node normal vector;
s22, denoising the point cloud data processed in the step S21 by combining the adjacent node normal vector obtained in the step S21 to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data.
Preferably, the step S22 specifically includes the following steps:
s221, performing integral scanning on the point cloud data processed in the step S21 in the x and y directions respectively by adopting a traversing mode to obtain scanning data;
s222, calculating an included angle theta 1 between the kth-1 node of a certain row or a certain column and the normal vector of the kth node by using the normal vector of the adjacent node obtained in the step S21, if the included angle theta 1 exceeds a preset threshold value
The kth point of the row or column is marked as 1 and is out of range;
s223, repeating the step S221 and the step S221 until all the points are traversed for 2 times to obtain an edge matrix, then removing noise data in the edge matrix to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data.
Preferably, in step S223, the noise data is point cloud data, where the point cloud and an edge point located at the boundary of the structural surface of the surrounding rock in the edge matrix cannot form the structural surface of the surrounding rock.
Preferably, step S3 comprises the steps of:
s31, using CloudCompare software to carry out corresponding marking on the training data obtained in the step S2 according to the structural plane grade of the surrounding rock, so as to obtain a training sample;
s32, inputting the training sample obtained in the step S31 into a DResNet-PointNet deep learning frame for training, and obtaining the trained DResNet-PointNet deep learning frame.
Preferably, step S31 comprises the steps of:
s311, dividing the tunnel portal side slope rock mass into four types to obtain the structural plane grade of surrounding rock;
s312, performing data augmentation on the partially preprocessed point cloud data obtained in the step S2, and then inputting the data augmented with the data into CloudCompare software to perform corresponding labeling according to the structural plane grade of the surrounding rock obtained in the step S311, so as to obtain a training sample.
Preferably, in step S32, the dresent-PointNet deep learning framework is obtained by the following method:
s321, constructing three residual multi-layer perceptrons ResMLP-1 on the basis of a PointNet framework, wherein each ResMLP-1 comprises two residual blocks, and each residual block comprises two convolution layers, so as to obtain a ResMLP-1 framework;
s322, constructing three residual multi-layer perceptrons ResMLP-2 on the basis of the ResMLP-1 framework obtained in the step S321, wherein each ResMLP-2 comprises three residual blocks, and each residual block comprises two convolution layers to obtain a ResMLP-2 framework;
s323, constructing three residual multi-layer perceptrons ResMLP-3 on the basis of the ResMLP-2 framework obtained in the step S322, wherein each ResMLP-2 comprises three residual blocks, each residual block comprises two convolution layers, and the ResMLP-3 framework, namely the DResNet-PointNet deep learning framework is obtained.
Preferably, the step S4 specifically includes the following steps:
s41, inputting the test data obtained in the step S2 into the trained DResNet-PointNet deep learning frame obtained in the step S3 to predict the structural surface of the point cloud surrounding rock, and obtaining the predicted structural surface of the surrounding rock;
s42, fitting points on the predicted structural surface of the surrounding rock obtained in the step S41 to obtain a general expression of a plane equation: ax+by+cz+d=0, wherein a, B, C, D are parameters, and then a structural face occurrence parameter calculation formula of the surrounding rock is obtained, and geometric information of the predicted structural face of the surrounding rock is extracted based on the formula to obtain structural face geometric information of the surrounding rock.
Compared with the prior art, the invention has the beneficial effects that:
the intelligent structural surface identification method of the three-dimensional point cloud surrounding rock based on the DResNet-PointNet provided by the invention can observe a full airspace through an unmanned aerial vehicle three-dimensional laser scanning technology; the surface real-time point cloud data is automatically acquired by carrying out omnibearing observation on the side slope rock mass; based on CloudCompare, carrying out data annotation on partial rock mass point cloud data as a training sample; the method comprises the steps of carrying out deep learning training on training data by using DResNet-PointNet, then carrying out partial prediction data set labeling, carrying out structural surface prediction of point cloud surrounding rocks on other point cloud data, calculating relevant geometric parameters such as relevant occurrence, roughness and the like of the point cloud surrounding rocks by using the structural surfaces of the point cloud surrounding rocks, efficiently evaluating the rock mass of the region, providing important data support for slope stability evaluation, carrying out processing mainly by a computer in the whole process, and having the advantages of small manual requirement, high efficiency and high quality.
Drawings
FIG. 1 is a flow chart of a structural plane intelligent identification method of a three-dimensional point cloud surrounding rock based on DResNet-PointNet, which is provided by the embodiment of the invention;
FIG. 2 is a schematic diagram of a node normal vector calculation process according to an embodiment of the present invention;
FIG. 3 is a diagram of a prior art PointNet deep learning framework;
FIG. 4 is a DResNet-PointNet deep learning improvement framework in an embodiment of the invention;
FIG. 5 is a schematic diagram of a convolution residual error of a DResNet-PointNet in an embodiment of the present invention;
FIG. 6 is a training diagram of point cloud samples in an embodiment of the present invention;
FIG. 7 is a partial point cloud semantic segmentation prediction graph of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
As shown in fig. 1, the embodiment of the invention provides a structural plane intelligent identification method of a three-dimensional point cloud surrounding rock based on dresent-Pointnet, which specifically comprises the following steps:
s1, acquiring point cloud data of a tunnel portal rock mass by using an unmanned aerial vehicle-mounted three-dimensional laser scanner, wherein the method specifically comprises the following steps of:
c1: collecting project data and determining project detection requirements;
project data is collected as comprehensively as possible, the fine detection requirements of all parties such as construction parties, construction parties and the like are known, relevant laws and regulations are known, and project detection schemes are formulated.
C2: in-field investigation to evaluate field conditions and possible risk factors, including in particular potential flight disturbances, wind speed effects, electromagnetic interference, mist dust, etc.;
and C3: applying for a flight permit to a national authorities;
before flying, the unmanned aerial vehicle should apply for flight permission to relevant government parts, negotiate with construction units, construction units and the like, and strive for assistance of the construction units and the construction units.
And C4: according to the detection requirements and the field environment, selecting a proper unmanned aerial vehicle and matched equipment, training an unmanned aerial vehicle operator, and determining the heavy difficulty of a task and other details to be emphasized; in order to ensure that the unmanned aerial vehicle can reliably obtain image data, the embodiment specifically adopts a multi-rotor unmanned aerial vehicle system, and the unmanned aerial vehicle system has the characteristics of low flight, flexible flight, stable hovering, multi-lens carrying, onboard stabilizer, low-frequency anti-shake and the like.
C5: making an unmanned aerial vehicle flight plan, including a take-off and landing place, flight time, a flight track, a detection control point, a detection key area and the like of the unmanned aerial vehicle;
c6: and (3) carrying out real flight, obtaining point cloud data, and transmitting, storing and preprocessing the point cloud data.
S2, preprocessing the point cloud data obtained in the step S1 to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data (the dividing ratio of the training data to the test data is 1:9), wherein the method specifically comprises the following steps of:
s21, in order to improve the operation speed and facilitate the realization of algorithm programming, gridding processing is carried out on the point cloud data obtained in the step S1, the point cloud data with higher scanning accuracy can be realized by adopting a three-dimensional difference method, and then each node normal vector in the processed point cloud data is calculated to obtain each node normal vector; in general, most point normal vectors are fitted from 5 points, but fewer than 5 fitting points at the boundary are fitted from 4 points or 3 points. In order to embody the influence difference of the computing node and the 4 adjacent points on the computing node, the influence of the discrete points on the interpolation points is considered to be reduced along with the increase of the distance by referring to an inverse distance weighting method (IDW) in the interpolation algorithm. Therefore, for point normal vector calculation, the influence of the calculation node is greater than that of the 4 adjacent points, the calculation node can be given a weight of 2, the weight of the 4 adjacent points is 1, each node normal vector in the point cloud data is obtained through cyclic calculation, the node normal vector is used as a discrimination index, and the algorithm schematic diagram is shown in fig. 2.
S22, denoising the point cloud data processed in the step S21 by combining the adjacent node normal vector obtained in the step S21 to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data, wherein the method specifically comprises the following steps of:
s221, performing integral scanning on the point cloud data processed in the step S21 in the x and y directions respectively by adopting a traversing mode to obtain scanning data;
s222, calculating an included angle theta 1 between the kth-1 node of a certain row or a certain column and the normal vector of the kth node by using the normal vector of the adjacent node obtained in the step S21, if the included angle theta 1 exceeds a preset threshold value
The kth point of the row or column is marked as 1 and is out of range;
s223, repeating the step S221 and the step S221 until all the points are traversed for 2 times to obtain an edge matrix, then removing noise data in the edge matrix to obtain preprocessed point cloud data, dividing the preprocessed point cloud data into training data and test data, wherein the noise data is edge points positioned at the boundary of the structural surface of the surrounding rock in the edge matrix and point cloud data where the point cloud is positioned cannot form the structural surface of the surrounding rock.
S3, training the DResNet-PointNet point cloud deep learning frame through the training data obtained in the step S2 to obtain a trained DResNet-PointNet deep learning frame, wherein the training data specifically comprises the following steps:
s31, using CloudCompare software to correspondingly mark the training data obtained in the step S2 according to the structural plane grade of the surrounding rock to obtain a training sample, and specifically comprises the following steps:
s311, dividing the tunnel portal side slope rock mass into four types to obtain the structural plane grade of surrounding rock; the method for classifying rock mass structure types is proposed by Gu Dezhen teaching of the national academy of sciences of China, and according to the rock mass integrity, rock masses are divided into four types of integral block structures, layered structures, broken structures and discrete structures, wherein the rock mass of the integral block structures is lithology single, and the rock mass is formed by giant thick sedimentary rock, metamorphic rock and volcanic lava with slight structural deformation, and the fire invades the rock mass. The surrounding rock has a small structural surface, generally no more than three groups, extremely poor continuity and a closed state, and is free of filling or contains a small amount of scraps. The layered structure is a medium-thick layer rock body damaged by the structure or a thin layered rock body with strong buckling and interlayer dislocation under the structure effect, the structural surface of surrounding rock is 2-3 groups, the layered sheets are developed, the original soft interlayer and interlayer dislocation occur at time, the interlayer binding force is poor, and the structural surface of the surrounding rock is mostly filled with mud films, fragments and mud. The fracture structure generally develops in areas with strong construction activities, the surrounding rock has a large number of structural surface groups, high density and poor continuity, and the structural fracture is developed. The discrete structure develops in the construction of a broken belt, a strong weathered belt, cracks and joints are well developed, the structure classification of specific surrounding rocks is shown in the following table 1;
TABLE 1 rock mass structure type classification
S312, performing data addition on the training data obtained in the step S2, and then inputting the data subjected to data addition into CloudCompare software to perform corresponding labeling according to the structural plane grade of the surrounding rock obtained in the step S11, so as to obtain a training sample, as shown in FIG. 6;
s32, inputting the training sample obtained in the step S31 into a DResNet-PointNet deep learning frame for training, and obtaining the trained DResNet-PointNet deep learning frame.
The DResNet-PointNet deep learning framework is obtained by the following method:
(1) As shown in fig. 3, three residual multi-layer perceptron ResMLP-1 is constructed on the basis of the PointNet framework to raise or lower the dimension of the point cloud data and extract features, so as to obtain a ResMLP-1 framework; each ResMLP-1 contains two survivors, each survivor containing two convolutional layers. ResMLP-1 promotes the n×3-dimensional point cloud data to n×64 dimensions, and extracts point cloud features. ResMLP-1 has a depth that is twice the depth of the multilayer film and the corresponding position of the sensor in the dot network. Therefore, the ResMLP-1 can obtain more point cloud characteristic information than the multi-layer perceptron of the point network;
(2) Constructing three residual multi-layer perceptrons ResMLP-2 on the basis of the ResMLP-1 framework to obtain a ResMLP-2 framework; the N x 64-dimensional feature matrices are lifted to N x 1024-dimensional feature matrices by ResMLP-2 with three residual blocks and six convolutional layers. ResMLP-2 is deeper than ResMLP-1, and the dimensions of the ResMLP-2 point cloud feature are greatly improved. This is because some feature information is lost in extracting global features using maximum pooling after ResMLP-2. Only if the dimension of the feature is greatly improved (from N multiplied by 64 to N multiplied by 1024), accurate and rich feature information can be obtained;
(3) Three residual multi-layer perceptrons ResMLP-3 are constructed on the basis of the ResMLP-2 framework, each ResMLP-3 comprises three residual blocks, each residual block comprises two convolution layers, and the ResMLP-3 framework, namely the DResNet-PointNet deep learning framework is obtained, as shown in figure 4. ResMLP-3 functions to gradually reduce the dimensions of the combined features in the DResNet-PointNet deep learning framework (including 1024-dimensional global features and 64-dimensional local features), i.e., from N×1088 dimensions to N×64 dimensions. As shown in fig. 5, three residuum blocks are followed by a convolution layer mapping features to m semantic tags. Also, the depth of ResMLP-3 is twice the depth of the multi-layer perceptron at the corresponding location in the PointNet.
S4, extracting structural surface geometric information of the surrounding rock from the semantic division point cloud data obtained in the step S3 to obtain the structural surface geometric information of the surrounding rock, wherein the method specifically comprises the following steps of:
s41, inputting the test data obtained in the step S2 into the trained DResNet-PointNet deep learning frame obtained in the step S3 to predict the structural surface of the point cloud surrounding rock, so as to obtain the predicted structural surface of the surrounding rock, as shown in FIG. 7;
s42, fitting points on the predicted structural surface of the surrounding rock obtained in the step S41 to obtain a general expression of a plane equation: ax+by+Cz+D=0, wherein A, B, C and D are parameters, then a structural surface occurrence parameter calculation formula of the surrounding rock is obtained, and geometric information of the predicted structural surface of the surrounding rock is extracted based on the formula to obtain structural surface geometric information of the surrounding rock;
the structural face occurrence parameter calculation formula of the surrounding rock is as follows:
wherein S, N, E, W is north and south, respectively.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (8)
1. The intelligent structural surface identification method for the three-dimensional point cloud surrounding rock based on the DResNet-PointNet is characterized by comprising the following steps of:
s1, acquiring point cloud data of a tunnel portal rock mass by using an unmanned aerial vehicle-mounted three-dimensional laser scanner;
s2, preprocessing the point cloud data obtained in the step S1 to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data;
s3, training the DResNet-PointNet point cloud deep learning frame through the training data obtained in the step S2 to obtain a trained DResNet-PointNet deep learning frame;
s4, according to the trained DResNet-PointNet deep learning frame obtained in the step S3, carrying out structural plane prediction of the point cloud surrounding rock on the test data obtained in the step S2, and extracting structural plane geometric information of the predicted surrounding rock to obtain structural plane geometric information of the surrounding rock.
2. The intelligent structural surface recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet as set forth in claim 1, wherein the step S2 comprises the steps of:
s21, performing gridding treatment on the point cloud data obtained in the step S1, and then calculating each node normal vector in the treated point cloud data to obtain each node normal vector;
s22, denoising the point cloud data processed in the step S21 by combining the adjacent node normal vector obtained in the step S21 to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data.
3. The intelligent structural surface recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet as set forth in claim 2, wherein the step S22 specifically comprises the following steps:
s221, performing integral scanning on the point cloud data processed in the step S21 in the x and y directions respectively by adopting a traversing mode to obtain scanning data;
s222, calculating an included angle theta 1 between the kth-1 node of a certain row or a certain column and the normal vector of the kth node by using the normal vector of the adjacent node obtained in the step S21, if the included angle theta 1 exceeds a preset threshold value
The kth point of the row or column is marked as 1 and is out of range;
s223, repeating the step S221 and the step S221 until all the points are traversed for 2 times to obtain an edge matrix, then removing noise data in the edge matrix to obtain preprocessed point cloud data, and dividing the preprocessed point cloud data into training data and test data.
4. The method for intelligently identifying a structural surface of a three-dimensional point cloud surrounding rock based on dresent-PointNet according to claim 3, wherein in step S223, the noise data is point cloud data of edge points located at the boundary of the structural surface of the surrounding rock and where the point cloud is located, which cannot form the structural surface of the surrounding rock.
5. The intelligent structural surface recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet as claimed in claim 1, wherein the step S3 comprises the following steps:
s31, using CloudCompare software to carry out corresponding marking on the training data obtained in the step S2 according to the structural plane grade of the surrounding rock, so as to obtain a training sample;
s32, inputting the training sample obtained in the step S31 into a DResNet-PointNet deep learning frame for training, and obtaining the trained DResNet-PointNet deep learning frame.
6. The method for intelligently identifying a structural surface of a three-dimensional point cloud surrounding rock based on DResNet-PointNet as set forth in claim 5, wherein the step S31 comprises the steps of:
s311, dividing the tunnel portal side slope rock mass into four types to obtain the structural plane grade of surrounding rock;
s312, performing data addition on the training data obtained in the step S2, and then inputting the data subjected to data addition into CloudCompare software to perform corresponding labeling according to the structural plane grade of the surrounding rock obtained in the step S311, so as to obtain a training sample.
7. The method for intelligently identifying structural surfaces of three-dimensional point cloud surrounding rocks based on dresent-PointNet according to claim 5, wherein in step S32, the dresent-PointNet deep learning framework is obtained by the following method:
s321, constructing three residual multi-layer perceptrons ResMLP-1 on the basis of a PointNet framework, wherein each ResMLP-1 comprises two residual blocks, and each residual block comprises two convolution layers, so as to obtain a ResMLP-1 framework;
s322, constructing three residual multi-layer perceptrons ResMLP-2 on the basis of the ResMLP-1 framework obtained in the step S321, wherein each ResMLP-2 comprises three residual blocks, and each residual block comprises two convolution layers to obtain a ResMLP-2 framework;
s323, constructing three residual multi-layer perceptrons ResMLP-3 on the basis of the ResMLP-2 framework obtained in the step S322, wherein each ResMLP-2 comprises three residual blocks, each residual block comprises two convolution layers, and the ResMLP-3 framework, namely the DResNet-PointNet deep learning framework is obtained.
8. The intelligent structural surface recognition method of the three-dimensional point cloud surrounding rock based on the DResNet-PointNet as set forth in claim 1, wherein the step S4 specifically includes the following steps:
s41, inputting the test data obtained in the step S2 into the trained DResNet-PointNet deep learning frame obtained in the step S3 to predict the structural surface of the point cloud surrounding rock, and obtaining the predicted structural surface of the surrounding rock;
s42, fitting points on the predicted structural surface of the surrounding rock obtained in the step S41 to obtain a general expression of a plane equation: ax+by+cz+d=0, wherein a, B, C, D are parameters, and then a structural face occurrence parameter calculation formula of the surrounding rock is obtained, and geometric information of the predicted structural face of the surrounding rock is extracted based on the formula to obtain structural face geometric information of the surrounding rock.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211521091.4A CN116403124A (en) | 2022-11-30 | 2022-11-30 | Structural plane intelligent recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211521091.4A CN116403124A (en) | 2022-11-30 | 2022-11-30 | Structural plane intelligent recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116403124A true CN116403124A (en) | 2023-07-07 |
Family
ID=87018501
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211521091.4A Pending CN116403124A (en) | 2022-11-30 | 2022-11-30 | Structural plane intelligent recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116403124A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117556703A (en) * | 2023-11-23 | 2024-02-13 | 中铁大桥局集团有限公司 | Method, device and equipment for identifying rock mass structural surface of side slope and readable storage medium |
| CN117854060A (en) * | 2024-03-07 | 2024-04-09 | 山东大学 | Tunnel rock body planar crack identification method and system based on deep learning |
-
2022
- 2022-11-30 CN CN202211521091.4A patent/CN116403124A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117556703A (en) * | 2023-11-23 | 2024-02-13 | 中铁大桥局集团有限公司 | Method, device and equipment for identifying rock mass structural surface of side slope and readable storage medium |
| CN117854060A (en) * | 2024-03-07 | 2024-04-09 | 山东大学 | Tunnel rock body planar crack identification method and system based on deep learning |
| CN117854060B (en) * | 2024-03-07 | 2024-05-03 | 山东大学 | Tunnel rock body planar crack identification method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111091105B (en) | Remote sensing image target detection method based on new frame regression loss function | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| CN116403124A (en) | Structural plane intelligent recognition method of three-dimensional point cloud surrounding rock based on DResNet-PointNet | |
| CN110532889B (en) | Track foreign matter detection method based on rotor unmanned aerial vehicle and YOLOv3 | |
| CN112462346A (en) | Ground penetrating radar roadbed defect target detection method based on convolutional neural network | |
| CN107316064B (en) | Asphalt pavement crack classification and identification method based on convolutional neural network | |
| CN114693615A (en) | Deep learning concrete bridge crack real-time detection method based on domain adaptation | |
| CN109596078A (en) | Multi-information fusion spectrum of road surface roughness real-time testing system and test method | |
| CN110189304A (en) | Remote sensing image target on-line quick detection method based on artificial intelligence | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| US20220315243A1 (en) | Method for identification and recognition of aircraft take-off and landing runway based on pspnet network | |
| CN114972384A (en) | Tunnel rock intelligent rapid regional grading method based on deep learning | |
| CN105741267A (en) | Multi-source image change detection method based on clustering guided deep neural network classification | |
| CN111460919A (en) | Monocular vision road target detection and distance estimation method based on improved YO L Ov3 | |
| CN103871062A (en) | Lunar surface rock detection method based on super-pixel description | |
| CN110147714A (en) | Coal mine gob crack identification method and detection system based on unmanned plane | |
| CN113919226B (en) | Mining vegetation ecological cumulative effect disturbance range identification method based on weight | |
| CN112348758B (en) | Optical remote sensing image data enhancement method and target identification method | |
| CN113343563A (en) | Landslide susceptibility evaluation method based on automatic sample selection and surface deformation rate | |
| CN113450573A (en) | Traffic monitoring method and traffic monitoring system based on unmanned aerial vehicle image recognition | |
| CN113569921A (en) | Ship classification and identification method and device based on GNN | |
| CN112465057A (en) | Target detection and identification method based on deep convolutional neural network | |
| CN111508252A (en) | Method and device for updating high-definition map by using V2X information integration technology | |
| Malini et al. | An automatic assessment of road condition from aerial imagery using modified VGG architecture in faster-RCNN framework | |
| CN117288094B (en) | Real-time positioning system of heading machine based on laser sensor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |