CN117876581A - High-precision rock JRC value prediction method and system based on point cloud processing - Google Patents

High-precision rock JRC value prediction method and system based on point cloud processing Download PDF

Info

Publication number
CN117876581A
CN117876581A CN202311781392.5A CN202311781392A CN117876581A CN 117876581 A CN117876581 A CN 117876581A CN 202311781392 A CN202311781392 A CN 202311781392A CN 117876581 A CN117876581 A CN 117876581A
Authority
CN
China
Prior art keywords
rock
fracture
model
value
point cloud
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
Application number
CN202311781392.5A
Other languages
Chinese (zh)
Inventor
杜雪明
梁佳森
方宏远
薛冰寒
赵小华
狄丹阳
李斌
王念念
孙明明
杜明瑞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhengzhou University
Original Assignee
Zhengzhou University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhengzhou University filed Critical Zhengzhou University
Priority to CN202311781392.5A priority Critical patent/CN117876581A/en
Publication of CN117876581A publication Critical patent/CN117876581A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30132—Masonry; Concrete

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Quality & Reliability (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)

Abstract

The invention discloses a high-precision rock JRC value prediction method and a system based on point cloud processing, comprising the following steps: s1, shooting rock fracture images with different roughness, collecting rock fracture damage information in the rock fracture images, and registering depth frames of the images to generate three-dimensional point clouds of the surface of the rough fracture, so as to obtain a natural rough rock fracture geometric model; s2, generating a synthetic rock fracture geometric model through a fractal Brownian motion method; s3, merging point cloud data sets of the natural rough rock fracture geometric model and the synthetic rock fracture combination model to obtain a rock fracture expansion data set; s4, dividing the rock fracture expansion data set into a training set and a testing set, training the support vector machine model, and constructing to obtain a rock fracture prediction model; s5, inputting the newly shot rock fracture image data into a rock fracture prediction model, and outputting the prediction of the rock fracture JRC value. The invention realizes the comprehensive analysis and prediction of the fracture properties.

Description

High-precision rock JRC value prediction method and system based on point cloud processing
Technical Field
The invention relates to a high-precision rock JRC value prediction method and system based on point cloud processing, and belongs to rock fracture analysis technology.
Background
The characteristics of rock cracks are critical to the fields of geological exploration, rock mechanics research, engineering geology, geotechnical engineering and the like. The rock joint surface characteristics and roughness coefficient (JRC) are one of the key parameters for measuring the surface roughness of rock cracks, and have important effects on the mechanical properties and hydrogeologic conditions of the rock. Therefore, high-precision measurement of rock fracture characteristics and accurate prediction of JRC values are critical to research and applications in the relevant fields.
Conventional rock fracture characterization methods typically involve manual measurements, which have subjective and instability problems and are time consuming in processing large amounts of data. Furthermore, existing JRC prediction methods typically rely on empirical formulas, which have limited accuracy.
Along with the rapid development of artificial intelligence, more and more detection analysis projects can predict analysis results by constructing a model and a learning algorithm, but the rock JRC value is not only used for judging the type of rock fracture plane images, but also used for comprehensively judging three-dimensional information such as rock fracture depth, roughness and the like, so that the traditional industrial camera shoots the processed rock fracture images to learn the model algorithm, and the intelligent judgment result of the rock JRC value cannot be accurately output.
Disclosure of Invention
The invention solves the technical problems that: aiming at the problem that the existing artificial intelligent algorithm model is insufficient in accuracy in judging the characteristics and roughness of the joint surfaces of the rock, the high-accuracy rock JRC value prediction method and system based on point cloud processing are provided.
The invention is realized by adopting the following technical scheme:
the invention firstly discloses a high-precision rock JRC value prediction method based on point cloud processing, which comprises the following steps:
s1, shooting rock fracture images with different roughness, collecting rock fracture damage information in the rock fracture images, and registering depth frames of the images to generate three-dimensional point clouds of the surface of the rough fracture, so as to obtain a natural rough rock fracture geometric model;
s2, generating a synthetic rock fracture geometric model through a fractal Brownian motion method;
s3, merging the point cloud data set of the natural rough rock fracture geometric model obtained in the step S1 and the point cloud data set of the synthetic rock fracture combination model generated in the step S2 to obtain a rock fracture expansion data set for constructing a rock fracture prediction model;
s4, dividing the rock fracture expansion data set into a training set and a testing set, inputting fracture geometric parameters of the training set and the testing set, and training the support vector machine model to obtain a rock fracture prediction model;
S5, inputting the newly shot rock fracture image data into a rock fracture prediction model, and outputting the prediction of the rock fracture JRC value.
In the high-precision rock JRC value prediction method based on point cloud processing, specifically, the step S1 comprises the following sub-steps:
s11, rock materials with the same size are used as a basis for measuring the volume of the rough cracks, the rock materials are randomly cut, a series of rock natural crack injuries covering the values of the Baton standard section line curves are obtained, actual JRC values corresponding to the rock natural cracks are obtained, and a depth camera is used for shooting the selected rock natural crack injuries with different JRC values;
s12, carrying out coordinate system conversion on original depth image data acquired by shooting natural fracture damage of the rock to obtain three-dimensional point cloud data;
and S13, storing three-dimensional point cloud data as a natural rough rock fracture geometric model, and acquiring average height difference and peak-valley height of corresponding fractures and lengths, widths and depths of the fractures through the three-dimensional point cloud data.
In a high-precision rock JRC value prediction method based on point cloud processing according to the present invention, further, in the substep S12,
the position relationship between any spatial point Q (x, y, z) in the original depth image data acquired by the depth camera and the mapping point Q (u, v, d) of the depth camera on the original depth image is as follows:
d＝z·s，
Wherein x, y and z respectively represent the coordinates of any spatial point in three-dimensional space, f x And f y Representing the focal length of the depth camera in its image coordinate system on the x-axis and y-axis, u and v being the pixel coordinates of the depth camera in its image coordinate system, d being the depth data of the depth camera, representing the depth values of the spatial point Q (x, y, z) in the depth cypress image coordinate system, c x And c y Representing the position of the center point of the aperture of the depth camera, s is an original depth image scaling factor, and the mapping relation from the space point to the original depth image is converted into a matrix expression:
wherein the matrixIs an internal reference matrix of the depth camera, +.>Representing homogeneous coordinates in the depth camera image coordinate system, assuming that the origin of the world coordinate system coincides with the origin of the camera, the rotation matrix is +.> Translation vector->And converting the original depth data into a three-dimensional point cloud in a world coordinate system from a two-dimensional image in a depth camera coordinate system through the conversion matrix.
In the high-precision rock JRC value prediction method based on point cloud processing, specifically, the step S2 comprises the following sub-steps:
s21, setting the construction area of the rock fracture to be formed by four corner points A 0 、B 0 、C 0 And D 0 Initial coordinate determination in world coordinate system, initial coordinates of four corner points define an initial plane, and vertical coordinate value Z in each point coordinate value in the initial plane obeys N (0, sigma) 2 ) Wherein N (0, sigma) 2 ) Representing a mean value of 0 and a raw variance of sigma 2 Is a gaussian distribution of (c);
s22, carrying out one-dimensional fractal Brownian motion interpolation on the initial plane, and carrying out four corner points A 0 、B 0 、C 0 And D 0 The Z values of (2) and the values of adjacent nodes are respectively averaged to respectively obtain a central point A of four corner points 1 Four midpoints B connected with adjacent nodes 1 、C 1 、D 1 、E 1 Linear interpolation to generate a new point a 2 、B 2 、C 2 、D 2 And E is 2 From the first variance ofGaussian distribution>The random value is added to the corresponding new point, the first variance +.>Is calculated as follows:
wherein H is Hurst index, and the value range is 0-1, sigma 2 Is the original variance of the gaussian distribution;
s23, adopting the step S22 as a basic recursion process, and continuing to interpolate the A 2 、B 2 、C 2 、D 2 And E is 2 Generates new points from the values of (2) from the second varianceGaussian distribution>The extracted random value is added to the generated new point, the second variance +.>Is calculated as follows:
s24, repeating the step S23, and linearly inserting newly generated nodes each timeFrom the nth variance after the valueGaussian distributionThe sum of the extracted random values, n is the number of linear interpolations, n variance +.>Is calculated as follows:
finally, a node number (2) n +1) 2 Size 2 n ×2 n And a three-dimensional curved surface with fractal characteristics is used for simulating and synthesizing a rock fracture geometric model.
In the high-precision rock JRC value prediction method based on point cloud processing, further, the crack roughness is adjusted by changing the Hurst index of fractal Brownian motion in the generation process of the synthetic rock crack geometric model.
In the high-precision rock JRC value prediction method based on point cloud processing of the present invention, specifically, the point cloud data set in step S3 further includes:
the data format is consistent, and the point cloud formats of the two model point cloud data sets are consistent;
converting a coordinate system, wherein the point cloud data of the two models are in the same coordinate system;
data are spliced, and point cloud data sets of the two models are spliced together according to the iteration of the nearest points;
and adding the data source identifier to the combined point cloud data.
In the high-precision rock JRC value prediction method based on point cloud processing, specifically, the step S4 includes the following sub-steps:
s41, preparing data, namely preparing the data structures of a training set and a testing set in a rock fracture prediction model as follows:
the data of the training set comprises the width, length, depth, average height difference, peak-to-valley height and actual JRC value of the fracture model, wherein the width, length, depth, average height difference and peak-to-valley height of the fracture are input features X of model training train The data of the test set comprises the width, length, depth, average height difference, peak-to-valley height and actual JRC value of the fracture model, wherein the width, length, depth, average height difference and peak-to-valley height of the fracture are input features X of model training test ；
Using z-score normalization to give a mean value of 0 and a standard deviation of 1 for each data feature in the training set and test set except for JRC values;
s42, selecting Gaussian kernel as kernel function of the support vector machine model to train the support vector machine model, and inputting the characteristic X in the training set data train Inputting a support vector machine model for training, wherein in the training process, the support vector machine model adjusts the weight and bias of the support vector machine model through the following formula, and searches an optimal decision boundary, so that the actual JRC value of an output label and a crack of the support vector machine model is gradually reduced, and training data are maximized at intervals and correctly classified;
wherein w is the weight of the support vector machine model, b is the bias of the support vector machine model, ζ is the relaxation variable of the support vector machine model, C is the regularization parameter of the support vector machine model, and N is the number of training samples;
s43, after training is completed, evaluating the performance of the model by using the test set, and inputting the characteristic X in the training set data test Inputting a support vector machine model with training completed, comparing the JRC value output by the model with the actual JRC value of the sample fracture damage in the test set, and calculating the average absolute error, average deviation error and average error between the JRC value output by the model and the actual JRC value by the following formulaSquare root error, mean square error, and R square to evaluate the performance of the training model:
m is the number of samples of the rock fracture model in the test set, Y test,i For the actual JRC value, Y, of the fracture of the ith rock fracture sample in the training set pred,i Outputting a JRC value for an ith rock fracture sample in the test set;
MAE is Y test,i And Y pred,i The average absolute error between the two is in the range of [0, + ], the smaller the MAE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; MBE is Y test,i And Y pred,i The average deviation error between the two models is in the range of [0, + ] and the smaller the MBE value is, the better the model prediction performance of the support vector machine is, and 0 represents perfect prediction; RMSE is Y test,i And Y pred,i The root mean square error between the two is [0, + ], the smaller the RMSE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; MSE is Y test,i And Y pred,i The mean square error between the two is in the range of [0, + ], the smaller the MSE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; r_squared is Y test,i And Y pred,i R square, range betweenAround (- ≡1)]R2 = 1 represents a perfect fit, R square<0, which indicates that the prediction performance of the support vector machine model is poor, and 1, which indicates that the support vector machine model is perfectly fitted.
In the method for predicting the high-precision rock JRC value based on the point cloud processing, specifically, in the step S5, firstly, image processing is carried out on a newly shot rock fracture image, including noise removal, picture size cutting, picture brightness and contrast adjustment, then fracture characteristic data including fracture length, width, depth, average height difference and peak-valley height are extracted from the processed image, a rock fracture prediction model is input, and a predicted rock fracture JRC value is predicted.
The invention also discloses a high-precision rock JRC value prediction system based on point cloud processing, which comprises the following steps:
the depth camera is used for shooting and collecting rock fracture images with different roughness;
the computer equipment is used for storing and executing the high-precision rock JRC value prediction method based on the point cloud processing in the claims 1-8, the depth camera is in communication connection with the computer equipment, and the predicted JRC value is output according to the rock fracture image shot and collected by the depth camera.
Finally, the invention also discloses a storage medium, and the storage medium is stored with the rock fracture prediction model constructed by the high-precision rock JRC value prediction method based on the point cloud processing.
The invention comprehensively combines the depth camera point cloud processing technology, the fractal analysis method and the machine learning algorithm of the support vector machine, realizes the high-precision characteristic acquisition of rock cracks and the accurate prediction of the JRC (Joint Roughness Coefficient) value of the rock, and provides the acquisition of the non-contact, accurate and repeatable roughness coefficient of the joint surface of the rock in the fields of geology and engineering. The invention has the following advantages:
(1) And the rock fracture image features are acquired with higher precision. According to the invention, by using a depth camera and a point cloud processing technology, high-precision three-dimensional modeling of the rock fracture can be realized. Compared with the traditional mode of judging and measuring through an empirical value main pipe, the method can acquire the shape, the size and the distribution information of the crack more accurately, and provides a more reliable data base for the JRC value prediction output by the subsequent support vector machine model.
(2) The geometric model of the rock fracture is more fully expanded. The invention introduces a fractal analysis method to construct a synthetic rock fracture geometric model, combines morphological characteristics of the fracture with the Brownian motion fractal dimension, and can describe the complexity of the fracture more comprehensively. The method can be applied to generation of various rock types and crack shapes, expands the data set for shooting and obtaining the natural rough rock crack geometric model, improves the data universality for realizing the JRC value prediction training, and further improves the accuracy of the support vector machine model for predicting the rock JRC value.
(3) The prediction precision of the machine learning prediction model by adopting the support vector machine is higher. The invention uses a support vector machine regression model (SVM) to improve the prediction accuracy of the JRC value. In the rock fracture data, nonlinear association may exist between the rock fracture shape and the JRC value, and the nonlinear kernel function of the SVM has advantages in modeling of nonlinear relations, so that complex relations between the rock fracture characteristics and the JRC value can be captured better, and the JRC value of the rock fracture can be predicted accurately.
In summary, the invention provides a technology for acquiring rock fracture characteristics with high precision and predicting the JRC value based on an image point cloud processing technology and a machine learning model, can more accurately describe the morphological characteristics of the rock fracture and predict the JRC value, realizes comprehensive analysis and prediction of fracture properties, provides a powerful tool for the field of rock engineering, and has wide application prospect.
The invention is further described below with reference to the drawings and detailed description.
Drawings
Fig. 1 is a flow chart of a high-precision rock JRC value prediction method based on point cloud processing.
FIG. 2 is a partial rock fracture image acquired.
Fig. 3 is a schematic diagram of a mapping relation of coordinate system conversion of an original depth image of a rock natural fracture injury acquired through shooting in the invention.
FIG. 4 is a fractal Brownian motion interpolation schematic diagram of a curved surface of a geometric model of a synthetic rock fracture generated in the invention.
FIG. 5 is a schematic view of a synthetic rock fracture geometry model surface generated in the present invention.
Fig. 6 is a schematic diagram of a high-precision rock JRC value prediction system based on point cloud processing according to the present invention.
Reference numerals in the drawings: 100-depth camera, 200-computer.
Detailed Description
Example 1
Referring to fig. 1, a flow chart of a high-precision rock JRC value prediction method based on point cloud processing according to the present invention is illustrated, including the following steps:
s1, shooting rock fracture images with different roughness, collecting rock fracture damage information in the rock fracture images, and registering depth frames of the images to generate a three-dimensional point cloud of the surface of the rough fracture, so as to obtain a natural rough rock fracture geometric model.
Step S1 comprises the following sub-steps:
and S11, adopting rock materials with the same size as a rough fracture volume measurement basis, randomly cutting the rock materials, taking a series of rock natural fracture injuries covering the Baton standard section line curve values, obtaining JRC values corresponding to the rock natural fractures, and adopting a depth camera to shoot the selected rock natural fracture injuries with different JRC values.
In this example, limestone with dimensions of 40cm by 20cm by 5cm was used as the base material for the measurement of the volume of coarse cracks. And (3) obtaining a series of natural fracture injuries with different Baton joint surface roughness coefficients JRC by repeatedly and randomly cutting the limestone plate for a plurality of times and comparing 10 standard section typical surface line curves proposed by Baton to obtain partial injury images as shown in figure 2.
In the embodiment, a Microsoft Azure Kinect DK depth camera is adopted to shoot rock cracks with different acquired roughness, the device provides 3840×2160 pixel RGB images and 1024×1024 pixel depth images, and the depth camera realizes a time-of-flight (TOF) concept and consists of an infrared emitter and an infrared sensor. The infrared emitter emits pulses of light onto the object to be observed, and the infrared sensor receives pulses of light reflected back from the object, the distance between the infrared sensor and the infrared emitter being known, so that the device is able to determine the three-dimensional coordinates of each pixel of the sensor from the time it takes for infrared light to travel from the emitter to the sensor. Because of the space limitation of the rock cracks, the damage shooting position is mainly set according to the angle and the light condition, the depth camera lens is always moved up and down to a certain distance along the direction parallel to the crack, the shooting distance is set to be 100-200 cm, and an infrared device on the head of the depth camera is aligned to the crack damage center position.
And S12, carrying out coordinate system conversion on original depth image data acquired by shooting the natural fracture damage of the rock to obtain three-dimensional point cloud data.
As shown in fig. 3, the depth camera captures any spatial point Q (x, y, z) in the acquired original depth image data, and its mapping point Q (u, v, d) on the original depth image has the following positional relationship:
d＝z·s，
wherein x, y and z respectively represent the coordinates of any spatial point in three-dimensional space, f x And f y Representing the focal length of the depth camera in its image coordinate system on the x-axis and y-axis, u and v are pixel coordinates in the depth camera image coordinate system that represent the projected position of the point Q (x, y, z) on the depth camera image. Specifically, u denotes pixel coordinates in the horizontal direction, and v denotes pixel coordinates in the vertical direction. d is depth data of the depth camera, representing depth values of Q (x, y, z) in a depth camera coordinate system. This depth value is acquired by means of a depth camera,for representing the distance of the point Q relative to the depth camera, c x And c y S represents the original depth image scaling factor and represents the depth camera aperture center point position.
The mapping relation from the space point to the original depth image is converted into a matrix expression:
Wherein the matrixIs an internal reference matrix of the depth camera, +.>Representing homogeneous coordinates in the depth camera image coordinate system, S is a scale factor in the homogeneous coordinates, ensuring that the third component of the homogeneous coordinates is 1, and setting the origin of the world coordinate system to coincide with the origin of the camera, thus rotating the matrix->Translation vector->And converting the original depth data into a three-dimensional point cloud in a world coordinate system from a two-dimensional image in a depth camera coordinate system through the conversion matrix.
And S13, storing three-dimensional point cloud data as a natural rough rock fracture geometric model, and acquiring average height difference, peak-valley height, length, width and depth of corresponding fractures through the three-dimensional point cloud data.
In step S1 of the method, a depth camera is used for photographing, the principle of which is based on infrared or Time-of-Flight (Time-of-Flight) technology, by emitting light pulses and measuring their return Time to calculate the distance of the object surface. These principles will be explained in detail and are associated with the shooting process. The depth camera may measure distance information of the object surface through a sensor and capture point cloud data. These point cloud data can be used for subsequent analysis and processing, etc., and after the point cloud data is successfully segmented by step S12, a series of parameters corresponding to the fracture will be calculated: average height difference, peak-to-valley height, and length, width, and depth of the fissures.
The peak-to-valley height (Rz) is the vertical distance between the highest point and the lowest point in the point cloud, and its mathematical calculation formula is as follows:
Rz＝max(Z)-min(Z)
wherein: z is the height value in the point cloud. By calculating the height difference between the highest point and the lowest point, the peak-to-valley height (Rz) of the point cloud data can be obtained. The peak-to-valley height (also referred to as peak-to-valley value) is the vertical distance between the highest point and the lowest point on the fracture surface. It represents the surface irregularities and roughness of the fissures. A larger peak-to-valley height means a more irregular surface that may have an impact on friction, landslide, soil erosion, and the like.
The average height difference is the average value of the height differences of all points on the fracture surface, and the mathematical calculation formula is as follows:
wherein: n is the number of points in the point cloud, Z i Is the height value of the i-th point. By calculating the average of the elevation values of all the vertices, the average elevation difference can be obtained. The average height difference is the average of the height differences at all points on the fracture surface. It describes the degree of irregularity and roughness of the fracture surface. A high average height differential indicates that the fracture surface is uneven, possibly resulting in soil erosion and water flow disturbance.
Model dimensions include the Length, width, and Height of the Model, and their mathematical formulas are as follows:
Model Length＝max(X)-min(X)、
Model Width＝max(Y)-min(Y)、
Model Height＝max(Z)-min(Z)、
Wherein: x, Y, Z are the x, y, z coordinates in the point cloud, respectively.
The model center coordinate (CenterX, centerY, centerZ) is the center position coordinate of the point cloud model in the three-dimensional space, and the mathematical calculation formula is as follows:
the bounding box size of the crack point cloud data can be determined by calculating the range of X, Y, Z coordinates of the point cloud model in the three-dimensional space, wherein the bounding box size comprises the length, the width and the depth of the crack. The width of the crack represents the span of the crack in the horizontal direction, which describes the horizontal extent of the crack and can reflect the size of the crack. Larger fracture widths and lengths generally mean more open space and may have significant impact on hydrogeology and geotechnical engineering. The fracture depth represents the vertical position of the fracture, i.e. the distance of the bottom of the fracture from the rock surface, and also reflects the vertical position and possible inclination angle of the fracture, which can affect aspects such as groundwater flow, geotechnical engineering, and geological risk assessment.
The fracture geometry model may exhibit specific geometries over different frequency ranges, while the Power Spectral Density (PSD) of the point cloud data may help analyze these frequency components to provide information about the fracture geometry. The present embodiment also calculates the Power Spectral Density (PSD) of the point cloud data by Discrete Fourier Transform (DFT), and the result is stored in the PSD variable in units of decibels (dB), representing the spectral characteristics of the point cloud data.
The width, length, depth, average height differential and peak-to-valley height of the fracture can be directly extracted from the point cloud data of the rock fracture with a clear physical meaning without additional experimentation or measurement, typically without the need to destroy or alter the original fracture sample, by analyzing these features, insight into the shape, size, roughness and distribution of the fracture can be obtained, and using these features for prediction and analysis helps to better understand the characteristics and potential impact of the rock fracture.
S2, generating a synthetic rock fracture geometric model through a fractal Brownian motion method.
The method aims at expanding a data set and generating a synthetic rock fracture by introducing a fractal Brownian motion method. This method simulates the fractal characteristics of the fracture surface of the rock and generates additional fracture models. We will describe in detail the fractal brownian motion generation method including fractal parameters, size, depth, resolution, etc. Since fBm (fractal brownian motion) is defined as a series of values with fractal features that follow a gaussian random process, it is usually a mathematical model describing irregular and complex geometries with random fractal features in nature from a generated random fractal curve or surface.
The method specifically comprises the following substeps:
substep S21, as shown in FIG. 4, assume that the structural area of the composite rock fracture is defined by four corner points A 0 、B 0 、C 0 And D 0 Initial coordinate determination in world coordinate system, initial coordinates of four corner points defining an initial plane, Z value of each point in the initial plane obeying N (0, sigma) 2 ) Wherein N (0, sigma) 2 ) Representing a mean value of 0 and a raw variance of sigma 2 Is a gaussian distribution of (c).
The Z value represents a vertical coordinate value among coordinate values of each point in the initial plane. Specifically, the coordinates are (x, y, Z) for each point in the plane. The Z value represents the coordinate of this point in the vertical direction, i.e. the height. The gaussian distribution (normal distribution) probability density function to which the height value of each point in the initial plane follows is:
wherein μ is the mean value, σ 2 Is the variance. Here, the Z values obey a mean of 0 and a variance of σ 2 Gaussian distribution of (c), namely:
this means that the Z value distribution in the initial plane is a mean value of 0 and a variance of sigma 2 Is a gaussian distribution of (c).
S22, carrying out one-dimensional fractal Brownian motion interpolation on the initial plane, and carrying out four corner points A 0 、B 0 、C 0 And D 0 The values of (a) and the values of neighboring nodes are averaged, respectively, the neighboring nodes refer to the midpoints of the four corner neighbors, as shown in fig. 4, e.g., A1 is the midpoint of A0 and B0, B1 is the midpoint of A0, B0 and C0, and so on. The values of these midpoints are obtained by averaging the values of neighboring nodes. The process is iterated continuously, new nodes are generated, and a three-dimensional curved surface with fractal characteristics is formed. Center point A of four corner points respectively 1 Four midpoints B connected with adjacent nodes 1 、C 1 、D 1 、E 1 Linear interpolation to generate a new point a 2 、B 2 、C 2 、D 2 And E is 2 From the first variance ofGaussian distribution>Is added to the corresponding new point, the first variance +.>Is calculated as follows;
wherein H is Hurst index, and the value rangeAround 0-1, sigma 2 Is the original variance of the gaussian distribution.
The purpose of this random value is to introduce some randomness so that the resulting three-dimensional surface has certain irregularities and complexity. Statistically, the random values of the gaussian distribution are generated by a normal distribution, and a random number generation algorithm (Box-Muller conversion), which is a method of generating a normal distribution random number based on a transformation in the form of polar coordinates, may be used to extract the random values from the normal distribution. The method comprises the following specific steps:
(1) Generating two independent random numbers U from a uniform distribution [0,1 ] 1 And U 2 。
(2) Calculating variable Z 0 And Z 1 Wherein Z is 0 And Z 1 Is a standard normal distribution random number (average value is O, variance is 1) with independent same distribution
By this method, two independent random numbers Z can be generated from the standard normal distribution 0 And Z 1 . If it is required to generate a signal with other mean mu and variance lambda 2 Can simply distribute Z 0 Or Z is 1 Zoom and pan as follows:
X＝μ+λZ
wherein X is the final normal distribution random number and Z is Z 0 Or Z is 1 μ is the desired mean and λ is the standard deviation representing the normal distribution.
The result of the addition of the random values is that new points after linear interpolation are added with random values of Gaussian distribution, and some random disturbance is introduced to the new points, so that the finally formed three-dimensional curved surface is more real and has fractal characteristics.
Substep S23, pickingContinuing interpolation A using step S22 as basic recursion procedure 2 、B 2 、C 2 、D 2 And E is 2 Generating new points from the values of the neighboring nodes, from the second varianceGaussian distribution>The extracted random value is added to the generated new point, the second variance +.>Is calculated as follows:
sub-step S24, repeating step S23, and linearly interpolating the newly generated node each time, and then changing from the nth variance to the third varianceGaussian distribution->The sum of the extracted random values, n is the number of linear interpolations, n variance +.>Is calculated as follows:
finally, a node number (2) n +1) 2 Size 2 n ×2 n And a three-dimensional curved surface with fractal characteristics is used for simulating and synthesizing a rock fracture geometric model.
Step S2 creates a three-dimensional surface featuring rough rock fractures, as shown in fig. 5. The process of this step can be summarized as follows:
(1) Initializing: the initial coordinates of the four corner points A0, B0, C0 and D0 are initialized, which define a plane. The Z value of each point of the initial plane obeys a mean value of 0 and a variance of sigma 2 Is a gaussian distribution of (c).
(2) One-dimensional fractal brownian motion (fBm) operation: for the initial plane, one-dimensional fBm interpolation is first performed, interpolating each row and each column of the plane into a curve containing more points. This operation adds one dimension, forming a higher dimension surface.
(3) Recursive interpolation and randomness: for each boundary and midpoint, a linear interpolation operation is performed to generate a new point. This will expand the initial plane to a larger surface. Variance in this processIs calculated by a formula, which is subject to the Hurst index H and the variance sigma of the previous step 2 Influence.
(4) Increasing the dimension: the dimensions are sequentially increased by repeating the one-dimensional fBm interpolation operation and adding randomness. Each operation generates a higher dimensional surface based on the previous operation.
(5) Repeating the recursion: repeating the above recursive process increases one dimension for each interpolation and randomness operation. Finally, by multiple recursions, a three-dimensional surface can be generated in which the coordinates (X, Y, Z) of each point are based on the Hurst index H and the variance sigma 2 Calculated.
This step is in fact a multidimensional fractal interpolation process, by means of stepwise interpolation and recursive operations, a three-dimensional surface with fractal characteristics can be generated for simulating the complex shape of a rock fracture. Interpolation and randomness of each dimension introduces more detail and complexity, creating more dimensions. This method allows the generation of three-dimensional surfaces that are highly self-similar and have fractal properties to simulate structures such as rock fractures in nature. The roughness of the generated fracture can be controlled by changing the Hurst index of fractal brownian motion during the generation process. The greater the Hurst index, the smoother the fracture surface; the smaller the Hurst index, the rougher the fracture surface.
And S3, merging the point cloud data set of the natural rough rock fracture geometric model obtained in the step S1 and the point cloud data set of the synthetic rock fracture combination model generated in the step S2 to obtain a rock fracture expansion data set for constructing a rock fracture prediction model.
This step aims at expanding the data samples for model training and verification. The present embodiment uses software PCL (Point Cloud Library) to read, convert, merge the point cloud data of the two rock fracture geometric models, which will involve mapping and combining of the point cloud coordinates to ensure consistency and data integrity. The merging of the point cloud data may be performed as follows:
(1) Data format unification: ensuring that the point cloud formats of the two point cloud data sets are consistent typically involves coordinate axes, data types, attributes of the points, and the like.
(2) Coordinate system conversion: if the point cloud data come from different coordinate systems, it is necessary to ensure that they are in the same coordinate system. The coordinate system of one of the data sets is transformed so that they are aligned under the same coordinate system.
(3) And (3) data splicing: the two point cloud data sets are stitched together according to a closest point iterative algorithm (ICP). Points of both data sets are added to a new data set.
(4) Identifying the source of the data: for the consolidated data, an attribute or identifier is added to the rock fracture expansion dataset to represent the source of the data, and an additional identifier is added to each point of the point cloud data to indicate from which dataset the point came for differentiation in subsequent analysis.
S4, dividing the rock fracture expansion data set into a training set and a testing set, inputting fracture geometric parameters of the training set and the testing set to train the support vector machine model, outputting a predicted JRC value of the fracture, comparing the predicted JRC value with an actual JRC value of the fracture, and constructing to obtain the rock fracture prediction model.
The method comprises the following substeps:
Substep S41, data preparation. In the training process of the support vector machine model, a training set and a testing set containing fracture characteristics and corresponding JRC values need to be prepared first. These data sets are extracted or generated by previous steps, wherein the training set is used to train the model and the test set is used to evaluate the performance of the model. The data structures of the training set and the test set in the rock fracture prediction model are prepared as follows:
the data of the training set comprises the width, length, depth, average height difference, peak-to-valley height and actual JRC value of the fracture model, wherein the width, length, depth, average height difference and peak-to-valley height of the fracture are input features X of model training train The data of the test set comprises the width, length, depth, average height difference, peak-to-valley height and actual JRC value of the fracture model, wherein the width, length, depth, average height difference and peak-to-valley height of the fracture are input features X of model training test 。
Normalizing the input features is a common step in support of vector machine model training. This ensures that the dimensions of the different input features are consistent, avoiding the dominant impact of certain input features on the model. The present example uses z-score normalization to make the training set and test set include a mean value of 0 for each feature including fracture width, length, depth, average height difference, peak-to-valley height, and standard deviation of 1.
Classification of the rock fracture expansion dataset was performed by the HoldOut method using cross-validation (cvpart). Firstly, dividing a rock fracture expansion data set into a training set and a testing set through a cvpartition function; the rng function is then used to set a random number seed to ensure repeatability of the results. The same data set segmentation is ensured each time the code is run by setting the same random seed. Finally, indexes of the training set and the testing set are obtained through cv.training and cv.test, and the indexes are used for extracting the corresponding training set and testing set from the original data.
Step S42, selecting Gaussian kernel or polynomial kernel (Polynomial kernel) as kernel function of the support vector machine model for training the support vector machine model, and inputting the training set into the feature X train Inputting a support vector machine model, wherein the support vector machine model is communicated in the training processThe weight and bias of the support vector machine model are adjusted through the following formula, the optimal decision boundary is searched, the actual JRC value of the output label and the crack of the support vector machine model is gradually reduced, the training data is maximally spaced and correctly classified,
wherein w is the weight of the support vector machine model, b is the bias of the support vector machine model, ζ is the relaxation variable of the support vector machine model, C is the regularization parameter of the support vector machine model, and N is the number of training samples.
In a support vector machine model, the selection of the appropriate kernel function is critical to the performance of the model. For rock fracture prediction, different kernel functions, such as Gaussian kernels (Gaussian kernel) or polynomial kernels (Polynomial kernel), may be tried to select a kernel function that is more efficient in capturing complex relationships in the data.
The training process is a standard implementation of a Support Vector Machine (SVM), which is part of the prior art. The following is an explanation of the training process:
(1) Selecting a kernel function: in support vector machines, kernel functions are used to map input features into a high-dimensional space. Gaussian kernel (Gaussian), also called radial basis function (Radial Basis Function, RBF), is chosen herein.
(2) Model training: the input feature x_train of the training set is input to the support vector machine model. During training, the support vector machine model minimizes the loss function by adjusting weights and offsets. This optimization problem typically includes a regularization term to balance maximizing spacing and minimizing training errors.
(3) Optimization target: the support vector machine adjusts the model parameters to find the optimal decision boundaries by optimizing the targets:
where w is the weight, b is the bias, ζ is the relaxation variable, C is the regularization parameter, and N is the number of training samples.
(4) Maximizing the interval: the goal of a Support Vector Machine (SVM) is to find a decision boundary such that the distance of the sample point to the decision boundary is maximized. This is called maximizing the separation, which is achieved by optimizing w in the target.
(5) Correct classification: meanwhile, the training process of the support vector machine pursues correct classification of training data. This is introduced by a relaxation variable xi which allows some sample points to appear within the space or on the misclassified side.
After the training is completed, the performance of the model is evaluated by using the test set, and the standardized test set is input into the feature X test Inputting a support vector machine model after training, comparing a JRC value output by the model with an actual JRC value of sample fracture damage in a test set, and evaluating the performance of the training model by calculating an average absolute error, an average deviation error, a root mean square error, a mean square error and an R square between the JRC value output by the model and the actual JRC value through the following formula:
wherein,m is the number of samples of the rock fracture model in the test set, Y test,i For the actual JRC value, Y, of the fracture of the ith rock fracture sample in the training set pred,i To output JRC values for the ith rock fracture sample in the test set, MAE is Y test,i And Y pred,i Average absolute error between MBE is Y test,i And Y pred,i Average deviation error between RMSE is Y test,i And Y pred,i Root mean square error between MSE and Y test,i And Y pred,i The mean square error between R_squared is Y test,i And Y pred,i And R square in between, and is used for evaluating the fitting degree of a regression model to observed data.
Specifically, the range of MAE values is [0, + ] and the smaller the MAE value, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; the range of MBE values is [0, + ], the smaller the MBE value is, the better the model prediction performance of the support vector machine is, and 0 represents perfect prediction; the range of the RMSE value is [0, + ], the smaller the RMSE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; the range of MSE values is [0, + ], the smaller the MSE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; the range of R squared values is (- ≡1], with less than 0 indicating poor predictive performance of the support vector machine model and 1 indicating perfect fit of the support vector machine model.
S5, inputting the newly shot rock fracture image data into a rock fracture prediction model, and outputting the prediction of the rock fracture JRC value.
Firstly, performing image processing on a newly shot rock fracture image, including removing noise, cutting the image size, adjusting the brightness and contrast of the image, then extracting fracture characteristic data including fracture length, width, depth, average height difference and peak-valley height from the processed image through converting three-dimensional point cloud data, inputting the fracture characteristic data into a rock fracture prediction model, and predicting the rock fracture prediction model according to the input fracture characteristic data to obtain a predicted rock fracture JRC value.
Example two
As shown in fig. 6, the embodiment discloses a high-precision rock JRC value prediction system based on point cloud processing, which comprises a depth camera 100 and a computer device 200, wherein the depth camera 100 is used for shooting and collecting rock fracture images with different roughness, and transmitting the shot rock fracture images with different roughness to the computer device 200 in a communication manner, and the computer device 200 is used for storing and executing the high-precision rock JRC value prediction method based on point cloud processing in the first embodiment, and outputting predicted JRC values according to the rock fracture images shot and collected by the depth camera.
The specific operation procedure of this embodiment is as follows:
first, an image of the rock fracture is taken using the depth camera 100. These images are input into the computer device 200, and the processes of predicting the JRC value of the rock fracture stored in the computer device 200 involve the procedures of image acquisition, preprocessing, feature extraction, model calling, model prediction, result display and the like. The computer device 200 pre-processes the captured image data to ensure that it is suitable for the predictive model stored therein. Preprocessing includes removing image noise, cropping image size, adjusting image brightness and contrast. The necessary rock fracture features, including fracture surface width, length, depth, average height difference, and peak-to-valley height, are then extracted from the image, which will constitute the input data for the model. According to the requirements of the support vector machine model stored in the computer equipment 200, the extracted features are standardized or normalized, so that the input features and the data used in model training are ensured to have similar data distribution. The computer device 200 runs the constructed support vector machine model, transmits the normalized input features to the support vector machine model, and predicts the model based on the input features to obtain an estimate of the fracture JRC value, which will be the output of the real-time prediction. The crack JRC values predicted by the support vector machine model are displayed on a computer screen or recorded in a file for viewing or further analysis by the user.
The present embodiments may enable real-time estimation and monitoring of rock fracture roughness in rock engineering and other applications.
Example III
The embodiment discloses a storage medium, and the storage medium stores a rock fracture prediction model constructed by the high-precision rock JRC value prediction method based on point cloud processing, which can be used for the rock JRC value prediction by the computer equipment in the second embodiment through program calling. The specific implementation process of the storage medium is basically the same as the high-precision rock JRC value prediction method based on the point cloud processing in the second embodiment, and will not be described herein.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.

Claims (10)

1. The high-precision rock JRC value prediction method based on the point cloud processing is characterized by comprising the following steps of:
s1, shooting rock fracture images with different roughness, collecting rock fracture damage information in the rock fracture images, and registering depth frames of the images to generate three-dimensional point clouds of the surface of the rough fracture, so as to obtain a natural rough rock fracture geometric model;
s2, generating a synthetic rock fracture geometric model through a fractal Brownian motion method;
s3, merging the point cloud data set of the natural rough rock fracture geometric model obtained in the step S1 and the point cloud data set of the synthetic rock fracture combination model generated in the step S2 to obtain a rock fracture expansion data set for constructing a rock fracture prediction model;
s4, dividing the rock fracture expansion data set into a training set and a testing set, inputting fracture geometric parameters of the training set and the testing set, and training the support vector machine model to obtain a rock fracture prediction model;
s5, inputting the newly shot rock fracture image data into a rock fracture prediction model, and outputting the prediction of the rock fracture JRC value.
2. The method for predicting the JRC value of the rock with high precision based on the point cloud processing according to claim 1, wherein the step S1 comprises the following sub-steps:
S11, rock materials with the same size are used as a basis for measuring the volume of the rough cracks, the rock materials are randomly cut, a series of rock natural crack injuries covering the values of the Baton standard section line curves are obtained, actual JRC values corresponding to the rock natural cracks are obtained, and a depth camera is used for shooting the selected rock natural crack injuries with different JRC values;
s12, carrying out coordinate system conversion on original depth image data acquired by shooting natural fracture damage of the rock to obtain three-dimensional point cloud data;
and S13, storing three-dimensional point cloud data as a natural rough rock fracture geometric model, and acquiring average height difference and peak-valley height of corresponding fractures and lengths, widths and depths of the fractures through the three-dimensional point cloud data.
3. A method for predicting a high-precision rock JRC value based on point cloud processing as recited in claim 2, wherein, in the substep S12,
the position relationship between any spatial point Q (x, y, z) in the original depth image data acquired by the depth camera and the mapping point Q (u, v, d) of the depth camera on the original depth image is as follows:
d＝z·S
wherein x, y and z respectively represent the coordinates of any spatial point in three-dimensional space, f x And f y Representing the focal length of the depth camera in its image coordinate system on the x-axis and y-axis, u and v being the pixel coordinates of the depth camera in its image coordinate system, d being the depth data of the depth camera, representing the depth values of the spatial point Q (x, y, z) in the depth cypress image coordinate system, c x And c y Representing the position of the center point of the aperture of the depth camera, s is an original depth image scaling factor, and the mapping relation from the space point to the original depth image is converted into a matrix expression:
wherein the matrixIs an internal reference matrix of the depth camera, +.>Representing homogeneous coordinates in the depth camera image coordinate system, assuming that the origin of the world coordinate system coincides with the origin of the camera, the rotation matrix is +.> Translation vectorAnd converting the original depth data into a three-dimensional point cloud in a world coordinate system from a two-dimensional image in a depth camera coordinate system through the conversion matrix.
4. The method for predicting the JRC value of the rock with high precision based on the point cloud processing according to claim 1, wherein the step S2 comprises the following sub-steps:
s21, setting the construction area of the rock fracture to be formed by four corner points A 0 、B 0 、C 0 And D 0 Initial coordinate determination in world coordinate system, initial coordinates of four corner points define an initial plane, and vertical coordinate value Z in each point coordinate value in the initial plane obeys N (0, sigma) 2 ) Wherein N (0, sigma) 2 ) Representing a mean value of 0 and a raw variance of sigma 2 Is a gaussian distribution of (c);
s22, carrying out one-dimensional fractal Brownian motion interpolation on the initial plane, and carrying out four corner points A 0 、B 0 、C 0 And D 0 The Z values of (2) and the values of adjacent nodes are respectively averaged to respectively obtain a central point A of four corner points 1 Four midpoints B connected with adjacent nodes 1 、C 1 、D 1 、E 1 Linear interpolation to generate a new point a 2 、B 2 、C 2 、D 2 And E is 2 From the first variance ofGaussian of (2)Distribution->The random value is added to the corresponding new point, the first variance +.>Is calculated as follows:
wherein H is Hurst index, and the value range is 0-1, sigma 2 Is the original variance of the gaussian distribution;
s23, adopting the step S22 as a basic recursion process, and continuing to interpolate the A 2 、B 2 、C 2 、D 2 And E is 2 Generates new points from the values of (2) from the second varianceGaussian distribution>The extracted random value is added to the generated new point, the second variance +.>Is calculated as follows:
s24, repeating the step S23, and performing linear interpolation on the newly generated node each time and then taking the n variance as the n varianceGaussian distributionThe sum of the extracted random values, n is the number of linear interpolations, n variance +.>Is calculated as follows:
finally, a node number (2) n +1) 2 Size 2 n ×2 n And a three-dimensional curved surface with fractal characteristics is used for simulating and synthesizing a rock fracture geometric model.
5. The method for predicting the JRC value of the high-precision rock based on the point cloud processing according to claim 4, wherein the crack roughness is adjusted by changing the Hurst index of fractal Brownian motion in the generation process of the synthetic rock crack geometric model.
6. The method for predicting the JRC value of the rock with high precision based on the point cloud processing according to claim 1, wherein the point cloud data set in step S3 further comprises:
the data format is consistent, and the point cloud formats of the two model point cloud data sets are consistent;
converting a coordinate system, wherein the point cloud data of the two models are in the same coordinate system;
data are spliced, and point cloud data sets of the two models are spliced together according to the iteration of the nearest points;
and adding the data source identifier to the combined point cloud data.
7. The method for predicting the JRC value of the rock with high precision based on the point cloud processing according to claim 1, wherein the step S4 comprises the following sub-steps:
s41, preparing data, namely preparing the data structures of a training set and a testing set in a rock fracture prediction model as follows:
the data of the training set comprises the width, length, depth, average height difference, peak-to-valley height and actual JRC value of the fracture model, wherein the width, length, depth, average height difference and peak-to-valley height of the fracture are input features X of model training train The data of the test set comprises the width, length, depth, average height difference, peak-to-valley height and actual JRC value of the fracture model, wherein the width, length, depth, average height difference and peak-to-valley height of the fracture are input features X of model training test ；
Using z-score normalization to give a mean value of 0 and a standard deviation of 1 for each data feature in the training set and test set except for JRC values;
s42, selecting Gaussian kernel as kernel function of the support vector machine model to train the support vector machine model, and inputting the characteristic X in the training set data train Inputting a support vector machine model for training, wherein in the training process, the support vector machine model adjusts the weight and bias of the support vector machine model through the following formula, and searches an optimal decision boundary, so that the actual JRC value of an output label and a crack of the support vector machine model is gradually reduced, and training data are maximized at intervals and correctly classified;
wherein w is the weight of the support vector machine model, b is the bias of the support vector machine model, ζ is the relaxation variable of the support vector machine model, C is the regularization parameter of the support vector machine model, and N is the number of training samples;
s43, after training is completed, evaluating the performance of the model by using the test set, and inputting the characteristic X in the training set data test Inputting a support vector machine model with training completed, comparing the JRC value output by the model with the actual JRC value of the sample fracture damage in the test set, and calculating the average absolute error, average deviation error, root mean square error, mean square error and sum of the JRC value output by the model and the actual JRC value by the following formula R square to evaluate the performance of the training model:
m is the number of samples of the rock fracture model in the test set, Y test,i For the actual JRC value, Y, of the fracture of the ith rock fracture sample in the training set pred,i Outputting a JRC value for an ith rock fracture sample in the test set;
MAE is Y test,i And Y pred,i The average absolute error between the two is in the range of [0, + ], the smaller the MAE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; MBE is Y test,i And Y pred,i The average deviation error between the two models is in the range of [0, + ] and the smaller the MBE value is, the better the model prediction performance of the support vector machine is, and 0 represents perfect prediction; RMSE is Y test,i And Y pred,i The root mean square error between the two is [0, + ], the smaller the RMSE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; MSE is Y test,i And Y pred,i The mean square error between the two is in the range of [0, + ], the smaller the MSE value is, the better the prediction performance of the support vector machine model is, and 0 represents perfect prediction; r_squared is Y test,i And Y pred,i R square in between, in the range (- ≡1)]A value less than 0 indicates poor predictive performance of the support vector machine model, and a value of 1 indicates perfect fit of the support vector machine model.
8. The method for predicting the JRC value of the rock with high precision based on the point cloud processing according to claim 1, wherein in the step S5, the image processing is performed on the newly shot rock fracture image, including removing noise, cutting the image size, adjusting the image brightness and contrast, then the fracture characteristic data including the fracture length, width, depth, average height difference and peak-valley height are extracted from the processed image, and the fracture characteristic data is input into a rock fracture prediction model to predict the JRC value of the rock fracture to be predicted.
9. A high-precision rock JRC value prediction system based on point cloud processing is characterized by comprising:
the depth camera is used for shooting and collecting rock fracture images with different roughness;
the computer equipment is used for storing and executing the high-precision rock JRC value prediction method based on the point cloud processing in the claims 1-8, the depth camera is in communication connection with the computer equipment, and the predicted JRC value is output according to the rock fracture image shot and collected by the depth camera.
10. A storage medium, wherein the storage medium stores thereon a rock fracture prediction model constructed by the high-precision rock JRC value prediction method based on the point cloud processing according to any one of claims 1 to 8.
CN202311781392.5A 2023-12-22 2023-12-22 High-precision rock JRC value prediction method and system based on point cloud processing Pending CN117876581A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311781392.5A CN117876581A (en) 2023-12-22 2023-12-22 High-precision rock JRC value prediction method and system based on point cloud processing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311781392.5A CN117876581A (en) 2023-12-22 2023-12-22 High-precision rock JRC value prediction method and system based on point cloud processing

Publications (1)

Publication Number Publication Date
CN117876581A true CN117876581A (en) 2024-04-12

Family

ID=90593900

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311781392.5A Pending CN117876581A (en) 2023-12-22 2023-12-22 High-precision rock JRC value prediction method and system based on point cloud processing

Country Status (1)

Country Link
CN (1) CN117876581A (en)

Similar Documents

Publication Publication Date Title
US8462987B2 (en) Detecting multiple moving objects in crowded environments with coherent motion regions
CN109740588B (en) X-ray picture contraband positioning method based on weak supervision and deep response redistribution
Gong et al. Data processing for real-time construction site spatial modeling
CA3062802C (en) System and methods for 3d model evaluation
CN110866531A (en) Building feature extraction method and system based on three-dimensional modeling and storage medium
CN111191629A (en) Multi-target-based image visibility detection method
CN113470090A (en) Multi-solid-state laser radar external reference calibration method based on SIFT-SHOT characteristics
CN112347550A (en) Coupling type indoor three-dimensional semantic graph building and modeling method
CN113916130B (en) Building position measuring method based on least square method
CN112991424B (en) Fractal dimension calculation method and device based on octree algorithm
Daghigh et al. A critical review of discontinuity plane extraction from 3D point cloud data of rock mass surfaces
CN113012063B (en) Dynamic point cloud repairing method and device and computer equipment
CN115761303A (en) Ground object classification method based on airborne laser radar point cloud and remote sensing image data
CN117876381B (en) AI visual detection method and system for identifying and analyzing concrete structure cracks
Sánchez-Aparicio et al. Detection of damage in heritage constructions based on 3D point clouds. A systematic review
CN115222884A (en) Space object analysis and modeling optimization method based on artificial intelligence
Vincke et al. Vision based metric for quality control by comparing built reality to BIM
CN115953604B (en) Real estate geographic information mapping data acquisition method
Srivastava et al. Drought stress classification using 3D plant models
Budianti et al. Background blurring and removal for 3d modelling of cultural heritage objects
CN117876581A (en) High-precision rock JRC value prediction method and system based on point cloud processing
Kuçak The feature extraction from point clouds using geometric features and ransac algorithm
CN118172242B (en) Unmanned plane water surface image stitching method, medium and system
Lee A hybrid model for DTM generation from LIDAR data
Wang et al. High-Quality Contour Line Generation from LiDAR Point Clouds for the Area of Forests

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