Underground engineering blasting hole internal crack identification and charging intelligent matching design method
Through deep learning and image processing technology, the rock mass cracks in underground engineering are identified and the blasting design is optimized, which solves the problems of rock mass damage and low explosive utilization in traditional blasting technology, and achieves a more efficient and safe blasting effect.
Patent Information
- Application Number
- CN202411020935.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-29
AI Technical Summary
In underground engineering, traditional blasting technology causes great damage to the rock mass, low explosive utilization rate, and manual crack measurement takes a long time, which poses high cost and safety risks.
Deep learning and image processing technology are used to build a rock fracture detection system and a loading intelligent matching design system. By obtaining rock fracture images, pre-processing and identification are performed, using fuzzy C-mean clustering, least squares method, and Zhang-Suen fines. Techniques such as synthesis algorithms and SURF algorithms extract crack features and perform splicing and edge detection, calculate crack area and length, establish a loading intelligent matching design system, and optimize blasting design.
It improves the energy utilization rate of explosives, reduces rock mass damage, reduces construction costs, and improves blasting efficiency and safety.
Smart Images

Figure CN118940525A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of tunnel and underground engineering equipment, and in particular to a design method for crack identification and charge intelligent matching in underground engineering blasting blastholes. Background Art
[0002] With the rapid development of social economy, my country has built a large number of important underground projects, involving transportation tunnels, underground mining, etc. With the increasing construction of underground projects, theories and construction methods, equipment and equipment are also developing rapidly. However, under complex geological conditions, the blasting process needs to be adjusted in time according to actual conditions. At present, there is still a strong uncertainty in the construction process, and it is a complex and high-risk system project.
[0003] As a common means of excavation in engineering, blasting technology is widely used in tunnel, mining and other projects. While blasting brings efficient engineering economic benefits, it also causes a series of problems such as rock damage. Therefore, the blasting effect can be improved by charging structure. However, there are many problems with traditional charging methods, which lead to high construction costs and safety hazards. Among them, the distribution characteristics of rock cracks can reflect the development of cracks inside the rock mass. When there are many cracks, manual measurement often takes a lot of manpower and time. Relying on digital extraction of cracks can greatly improve the speed and efficiency of obtaining crack information. The gradual shift from manual to digital methods is of great significance for the intelligent extraction of crack information and the intelligent adjustment of charging structure in blasting engineering. Charging structure is an important factor in adjusting the energy effect of explosives in blasting. With the development of blasting technology, different types of explosives continue to appear, and new charging structures using different explosives are also constantly produced, but they still cause a lot of blasting explosive energy loss.
[0004] Considering the influence of crack factors on blasting effect, quantitative analysis of rock cracks is realized based on deep learning and image processing technology, and the charging structure is further explored. A rock crack detection system and a charging intelligent matching design system are designed, and an in-hole camera equipment is built to realize the supporting development of software and hardware, so as to optimize and adjust the blasting design scheme, reduce the impact of rock damage caused by blasting, and improve the energy utilization rate of explosive blasting, which has certain engineering significance for the construction of tunnels and underground projects. Summary of the invention
[0005] The purpose of the embodiment of the present invention is to provide a method for designing intelligent matching of cracks in underground engineering blasting holes and charge, so as to alleviate the technical problems in the prior art of large damage to rock mass and low utilization rate of explosives during blasting.
[0006] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:
[0007] The first aspect of the present invention provides a method for identifying cracks in underground engineering blasting holes and intelligent matching design of charges, comprising the following steps:
[0008] S10. Obtain rock mass crack images and preprocess the rock mass crack images; establish a data set in a deep learning network model;
[0009] S20. Build a deep learning network model; use the improved network model to perform image recognition on the processed image;
[0010] S30. According to the rock mass fracture identification result of step S20, the fracture is segmented from the background and quantitatively expressed by extracting the identified fracture feature area, extracting and splicing the fracture skeleton and obtaining fracture set information;
[0011] S40. According to the recognition result of step S30, the charge structure in the blasthole is analyzed, and an intelligent charge matching design system based on the recognition of cracks in the blasthole is established;
[0012] S50. Establish an in-hole camera system.
[0013] In some modified implementations of the first aspect of the present invention, step S30 includes the following steps:
[0014] S31. Using fuzzy C-means clustering algorithm to perform image segmentation operation on rock mass crack image;
[0015] S32. Using the connected domain of the least squares method to remove tiny noise points from the image;
[0016] S33. Using Zhang-Suen thinning algorithm to extract crack skeleton lines and thin the crack image;
[0017] S34. Use the SURF algorithm to detect and match feature points to stitch the cracks, including feature point detection, descriptor extraction, feature point matching, geometric transformation and image fusion; use the weighted average fusion algorithm to stitch and fuse the two images into a complete image by superposition or mixing;
[0018] S35. Use Canny algorithm to extract edge features of rock crack images;
[0019] The crack width is calculated using the orthogonal skeleton algorithm;
[0020] The crack length is calculated using the neighborhood scanning method based on the skeleton curve;
[0021] The steps to calculate the crack area are:
[0022] S351. calibrate the image according to the scale mark or other known size information in the image;
[0023] S352. Use edge detection algorithm to extract crack contours and mark crack areas;
[0024] S353. Use the pixel counting function in the image processing software or programming language to count the number of pixels inside the crack region. This will give the pixel area of the crack region.
[0025] S354. Multiply the pixel area by the actual area of each pixel to obtain the actual area of the crack.
[0026] In some modified implementations of the first aspect of the present invention, step S31 includes the following steps:
[0027] S311. First, input the rock mass crack image data into the algorithm, set the number of clustered rock mass crack categories, fuzzy index p and initialize cluster centers, etc.;
[0028] S312. Set the threshold for stopping iteration and the maximum number of iterations;
[0029] S313. Then update the membership degree and calculate the cluster center;
[0030] S314. Determine whether the conditions of the standardized measurement function are met. If so, stop the algorithm and output the membership degree and cluster center;
[0031] S315. Divide the image pixels according to the maximum membership principle to achieve image segmentation of rock mass fissures;
[0032] Its expression is: Given a data set containing n pixels: x = {x1, x2, ..., xn}, the objective function of the fuzzy C-means (FCM) clustering algorithm is defined as follows:
[0033]
[0034]
[0035] Where: u mn is pixel x n Relative to the mth cluster center z m The fuzzy membership degree, p is the fuzzy index, d mn For x n and cluster center z m The grayscale distance between
[0036] Find the minimum value of the objective function J, and use the Lagrange multiplier method on the objective function while satisfying the constraints, so that J is m and umn The partial derivative of is 0, so we can get the membership matrix u mn , cluster center v m , the specific expression is as follows:
[0037]
[0038]
[0039] In some modified implementations of the first aspect of the present invention, step 32 includes the following steps:
[0040] S321. Find all connected domains in the rock mass fracture image and obtain the number of pixels contained in each of their connected domains;
[0041] S322. Determine the threshold range of the algorithm by using the least squares method;
[0042] S323. Determine the threshold range of the algorithm according to the middle value of the threshold;
[0043] S324. Processing the connected domain of the rock mass fracture image according to the threshold range;
[0044] The expression of the least squares method is:
[0045]
[0046] Where: a i is the coordinate information of the pixel point;
[0047] Step S33 includes the following steps:
[0048] S331: Scanning pixel by pixel, by scanning the image pixel by pixel, to find candidate pixels to be deleted; judging conditions include the black and white mode around the pixel and the connectivity of the pixel, and for the foreground color, checking whether the following conditions are met:
[0049] The current pixel is white, there is at least one black pixel in the 8-neighborhood of the current pixel P, and the number of white pixels in the 8-neighborhood of the current pixel is between 2 and 6. If all the above conditions are met, then the current pixel is marked as black;
[0050] S332: The algorithm is used to traverse each pixel of the image again, and for the foreground color, it is checked whether it meets the following conditions:
[0051] The current pixel is white, and at least one of the 8 neighborhoods of the current pixel is black. The number of white pixels in the 8 neighborhoods of the current pixel is between 2 and 6, and the black pixels in the 8 neighborhoods of the current pixel are not connected, that is, not a connected component. If all the above conditions are met, then the current pixel is marked as black. After repeated iterations, based on the first step, the rules are continuously applied iteratively until no pixels can be deleted. After the iteration, the image after refinement is obtained.
[0052] In some modified implementations of the first aspect of the present invention, step S34 includes the following steps:
[0053] S341. Determine the key point position by detecting the local extreme point in the image. If (x, y) is a pixel point in the image, the image of the point can be defined as the sum of the values of all pixels in the rectangular area surrounded by the image origin and the point. The mathematical expression is as follows:
[0054]
[0055] S342. Construct the Hessian matrix to generate stable mutation points in the image. The Hessian matrix can be expressed as:
[0056]
[0057] Where: σ is the Gaussian scale; (x, y) is the position of the image pixel; L xx (x, y, σ) represents the convolution of the second-order Gaussian derivative at the pixel (x, y) of image I, and the same is true for the other L;
[0058] S343. Generate a descriptor for each key point to describe the gradient information and distribution of the area around the key point; match the key points in the image and use a similarity measure based on the descriptor to find the corresponding key points in the image; through the matched key points, different transformation models can be used to estimate the geometric transformation relationship between the images;
[0059] S344. According to the estimated geometric transformation relationship, the two images are spliced and fused into a complete image by superposition or mixing. A weighted average fusion algorithm is adopted, and the weighting factor is u. The mathematical expression of the weighted fusion of the pixel values of the images is as follows:
[0060]
[0061] Where: u is the weighting factor; x, y are coordinates; f1, f2 represent images;
[0062] In some modified implementations of the first aspect of the present invention, the mathematical expression of the two-dimensional Gaussian function used by the Gaussian filter in the Canny algorithm is specifically as follows:
[0063]
[0064] S=G×I
[0065] Where: G represents the Gaussian filter obtained by operating the two-dimensional Gaussian function; I represents the image to be detected; S represents the image after smoothing;
[0066] The magnitude and direction of the gradient of each pixel in the smoothed image are calculated. The components of the gradient of the pixel on the edge in the vertical and horizontal directions (x, y) are as follows:
[0067]
[0068] Pixel gradient size g θ and direction θ are:
[0069]
[0070] According to the given high and low thresholds, the strong and weak edges are divided, and the isolated weak edges are eliminated to obtain the final detection result; two high and low thresholds T1 and T2 are set; the point with a gradient value larger than T1 is a strong edge point, and vice versa is a weak edge point;
[0071] The orthogonal skeleton method uses the KD tree algorithm, and the expression is as follows:
[0072]
[0073] Where: K is the dimension; d is the distance between data points, a i and i is a data point;
[0074] After using the KD tree algorithm to find the given k nearest neighbor points, the normal vector of the skeleton line is calculated by singular value decomposition, and then the two crack edge line points closest to the skeleton line normal vector are found, and the intersection of the line segment formed by these two points and the skeleton line normal vector is calculated; the expression of singular value decomposition is as follows:
[0075] Suppose the bilinear function is: f(x,y)=x T Ay
[0076] Where x, y belong to R n×1 , A belongs to R n×n , by introducing linear changes, the bilinear function can be expressed as: f(x,y)=(Uξ) T A(Vη)=ξ T U TAVη
[0077] Next, let S = U T AV
[0078] Then the bilinear function can be written as: f(x,y)=ξ T Sη
[0079] If U and V are both orthogonal matrices, S can be made into a diagonal matrix by using appropriate U and V. Specifically, it can be expressed as: S = diag(σ1,σ2,...,σ n )
[0080] Multiplying the left and right sides by U and VT can be expressed as: A = USV T ;
[0081] The crack length is calculated using the neighborhood scanning method based on the skeleton curve; the steps are as follows:
[0082] S351. First, the crack is skeletonized, and then the burr part on the crack skeleton line is processed by morphological operation;
[0083] S352. Scan the global image to find the end point of the crack, and use the end point as the starting point to search for crack pixels according to the 8-neighborhood method and count the number;
[0084] S353. Multiplying the obtained number of crack pixels by the resolution to obtain the actual length of the crack;
[0085] The calculation of the rock mass fracture direction is based on the extraction of the fracture skeleton line. By marking each point on the fracture surface, calculating the skeleton tangent direction vector, and then taking the average of all these skeleton tangent direction vectors, the average vector representing the overall skeleton tangent direction can be obtained. The average tangent direction vector is used to determine the inclination angle through the given x and y components of the average tangent direction vector. The mathematical expression is as follows:
[0086]
[0087]
[0088] In the formula: a is a vector; g1, g2, f1, f2 are coordinate points; n is the direction vector of all tangents.
[0089] After solving the direction of the average tangent, the angle θ of the crack can be obtained through inverse trigonometric functions. The specific mathematical expression is as follows:
[0090]
[0091] In some modified implementations of the first aspect of the present invention, step S10 includes the following steps:
[0092] S11. Use the weighted average method to grayscale the obtained rock mass crack color image, and perform weighted average distribution on the values of R, G and B. The expression is as follows:
[0093] R=G=B=W R R+W G G+W B B;
[0094] Where: R, G and B are the three component values of the image, and W R , W G and W B is the weight value of R, G and B. When W R =0.3,W G =0.59,W B =0.11, the required grayscale image can be obtained;
[0095] S12. The CA attention mechanism is used to obtain the attention of the image width and height. The input image is globally pooled in the horizontal and vertical directions respectively, and then the feature maps in the horizontal and vertical directions are obtained in turn. The calculation formula is as follows:
[0096]
[0097]
[0098] Where: x c is the input feature vector; h is the image height; w is the image width.
[0099] S13. The horizontal and vertical feature maps are spliced together in the spatial direction, and then a 1×1 convolution operation is performed to reduce the dimension. The feature map after the BN layer processing is sent to the Sigmoid activation function to obtain the feature map f. The calculation formula is as follows:
[0100] f=σ(F1([z h ,z w ]))
[0101] Where: F1 is the result of BN layer operation;
[0102] After a 1×1 convolution kernel, we get a feature map with the same number of channels as the original one, and decompose f into feature maps f in the spatial dimension. h and f w , and finally the attention weights in the horizontal and vertical directions are obtained through the Sigmoid activation function. The calculation formula is as follows:
[0103] g h =σ(F h (f h ))
[0104] g w =σ(F w (f w ));
[0105] Where: F h and F w f h and f w Transform into a transformation function of the same dimension as the output;
[0106] S14. Through the multiplication weighted operation on the original feature map, the final feature map with attention weights in both the horizontal and vertical directions will be obtained. The calculation formula is as follows:
[0107] y c =x c ×g h ×g w ;
[0108] Where: x c is the input feature map dimension.
[0109] In some modified implementations of the first aspect of the present invention, step S40 includes the following steps:
[0110] S41. Calculate the borehole pressure; the expression is:
[0111]
[0112] Where: P is the average pressure acting on the blast hole; P e is the initial pressure; γ is the gas expansion index; l e is the length of the charging section; l a is the length of the air column;
[0113] S42. Calculate the charge amount, the expression is:
[0114] Q = q × S × L;
[0115] Where: Q is the total charge for one blast; q is the unit consumption of blasting explosives; S is the excavation cross-sectional area; L is the average depth of the blasthole;
[0116] S43. For blastholes with discontinuous charge, the charge quantity Q can be calculated according to the following formula:
[0117]
[0118] Where: r is the radius of the blast hole; q is the unit consumption of blasting explosives; l e is the length of the charging section; K d is the radial uncoupling coefficient, K d =rb / r c , r b 、r c is the diameter of the blast hole and the diameter of the charge roll;
[0119] S44. When calculating the coupled charge, the impact load acting on the rock mass is expressed as follows:
[0120]
[0121]
[0122] We can get:
[0123]
[0124] Where: P is the maximum pressure; P H is the detonation pressure of explosive; D e is the detonation velocity of explosive; e , m is the density of explosive and rock; C p is the rock mass wave velocity;
[0125] When radial air is not coupled with charge, the peak value of blasting load pressure of charge section is as follows:
[0126]
[0127] Where: P0 is the maximum pressure peak of the charging section; K d is the uncoupling coefficient; n is the pressure increase factor, n = 8; D e is the detonation velocity of explosive; e is the density of explosive;
[0128] Since the time of air expansion in the hole is very short, it can be determined that the pressure on the hole wall after the air fills the entire blasthole is the peak pressure of the air section. The formula is as follows:
[0129]
[0130] Where: P1 is the maximum pressure peak of the air segment; γ is the gas expansion index; l e is the length of the charging section; l is the depth of the blasthole;
[0131] S45. Calculate K t1 The surface crack rate K t2 , the expression is
[0132]
[0133]
[0134] Where: A i , li are the area and trace length of the crack of the i-th structural plane respectively; V is the volume of the rock mass; A is the area of the rock mass section; n is the number of structural planes;
[0135] S46. Calculation of rock mass quality index and rock mass volume joint number J v , the expression is as follows:
[0136]
[0137] Where: L i is the length of the ith core greater than or equal to 10 cm in the drilling; L is the total length of the drilling;
[0138] The joint direction influence coefficient fi is introduced, and the mathematical expression is as follows:
[0139]
[0140] Where: TL is the total length of drilling; CW is the length of the scouring core section; Fr is the length of the core less than 5 cm; Cr is the length of the broken section core; K is the length of the karst area core;
[0141] Considering the influence of the above indicators, the new indicator LRQD based on the RQD method is as follows:
[0142]
[0143] Where: TL is the total length of drilling; Cr is the length of the core in the crushing section; Fr is the length of the core less than 5 cm; K t1 is the body crack rate; K t2 is the surface crack ratio; μ is the correction coefficient of the rock mass crack ratio; it is preliminarily determined that when the rock mass crack ratio is greater than 25, the correction coefficient μ is used to slow down the increase rate of the rock mass crack ratio; when the rock mass crack ratio is less than 25, the correction coefficient μ is 1;
[0144] S47. By comparing the new index LRQD and RQD, the body crack rate K t1 , surface crack rate K t2 , fracture area Fr and crushing area Cr, the original RQD method was modified and adjusted, and the mathematical expression of the correction coefficient α was proposed as follows:
[0145]
[0146] According to the distribution of rock mass cracks, the impact pressure is corrected to achieve the ideal effect. The formula is as follows:
[0147] P d =αP
[0148] Where: α is the correction coefficient; P is the pressure.
[0149] In some modified implementations of the first aspect of the present invention, in the in-hole camera system, the connecting step is:
[0150] S51. Connect the cable to the camera in the hole, select appropriate accessories to install on the camera in the hole, and connect the cable and power supply through the connecting wire;
[0151] S52. Connect the power supply and the industrial integrated machine to power the hole camera and the industrial integrated machine, and connect the connecting line, the converter and the video capture card in sequence to convert the analog signal of the hole camera into a digital signal that can be processed by the industrial integrated machine;
[0152] S53. Connect the converter and the video capture card to the industrial integrated machine through a suitable connecting cable. At this time, the in-hole camera device is connected and the designed program can be used on the industrial integrated machine for research activities; when in use, the cable and the in-hole camera are waterproof equipment;
[0153] S54. After use, turn off the industrial integrated machine and disconnect all components, clean them briefly and put them in the equipment box. BRIEF DESCRIPTION OF THE DRAWINGS
[0154] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required to be used in the embodiments are briefly introduced below.
[0155] Figure 1 The overall flow chart of the design method for crack identification and intelligent charge matching in underground engineering blasting holes;
[0156] Figure 2 The functions and flow chart of the rock crack detection system based on deep learning in the design method of crack identification and intelligent charge matching in underground engineering blasting holes;
[0157] Figure 3 Schematic diagram of rock crack detection system based on deep learning in the design method of crack identification and charge intelligent matching in underground engineering blasting holes;
[0158] Figure 4 The function and flow chart of charge matching design in the method of crack identification and charge intelligent matching design in underground engineering blasting holes;
[0159] Figure 5 A schematic diagram of the system for charge matching design in the method for identifying cracks in underground engineering blasting holes and intelligent charge matching design;
[0160] Figure 6This is a schematic diagram of the equipment connection of the in-hole camera system in the design method of crack identification and charge intelligent matching in underground engineering blasting holes.
[0161] In the figure: 1-industrial all-in-one machine; 2-in-hole camera; 3-cable; 4-first connecting line; 5-power supply; 6-second connecting line; 7-converter; 8-video acquisition card. DETAILED DESCRIPTION
[0162] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0163] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0164] Embodiment 1
[0165] The present invention relates to a method for designing crack identification and intelligent charge matching in underground engineering blasting blastholes, including a rock crack detection system module based on deep learning, a charge intelligent matching design system module based on crack identification in blastholes, and development of in-hole camera equipment.
[0166] like Figure 1 As shown in FIG. 1 , the design method for intelligent matching of crack identification and charge in underground engineering blasting blastholes includes the following steps:
[0167] Step (I): Obtain rock mass crack images and create a data set used in deep learning. The establishment of a data set generally includes three parts: image acquisition, image preprocessing, and image annotation. Generally, the image is grayed out and the weighted average method is used. The specific expression is as follows:
[0168] R=G=B=W R R+W G G+W B B
[0169] Where: R, G and B are the three component values of the image, and W R , W G and W B is the weight value of R, G and B.
[0170] A deep learning network model is established. The present invention uses the YOLOv5 network model. The backbone feature extraction network of YOLOv5 is Backbone. An important feature of Backbone is the use of a residual network Residual, which is easy to optimize and can improve accuracy by increasing depth. Image annotation uses a professional semantic segmentation dataset annotation software Labelimg to annotate the dataset images, mark the cracks in the image with a rectangular frame, and then explain the category. The CA attention mechanism is to obtain attention in terms of image width and height, encode more accurate data, and globally pool the input image in the horizontal and vertical directions, and then obtain the feature maps in the horizontal and vertical directions in turn. The calculation formula is as follows:
[0171]
[0172]
[0173] Where: xc is the input feature vector.
[0174] Then comes the attention generation operation, which splices the horizontal and vertical feature maps together in the spatial direction, and then performs a 1×1 convolution operation to reduce the dimension. The feature map processed by the BN layer is sent to the Sigmoid activation function to obtain the feature map f. The calculation formula is as follows:
[0175] f=σ(F1([z h ,z w ]))
[0176] Where: F1 is the result of BN layer operation.
[0177] After a 1×1 convolution kernel, we get a feature map with the same number of channels as the original one, and decompose f into feature maps f in the spatial dimension. h and f w , and finally the attention weights in the horizontal and vertical directions are obtained through the Sigmoid activation function. The calculation formula is as follows:
[0178] g h =σ(F h (f h ))
[0179] g w =σ(F w (f w ))
[0180] Where: F h and F w f h and f wTransform into a transformation function of the same dimension as the output.
[0181] The last step is the feature map correction operation. By multiplying the original feature map with weights, we can get the feature map with attention weights in both the horizontal and vertical directions. The calculation formula is as follows:
[0182] y c =x c ×g h ×g w
[0183] Where: x c is the input feature map dimension.
[0184] In order to improve the feature extraction capability of YOLOv5, the CA attention mechanism is added to the model. In order to verify whether the addition of the CA attention mechanism improves the recognition results, the recognition accuracy of rock fracture images is compared, and the test result data are specifically: Precision, Recall, Map, F1, etc.
[0185] Step (2): Use PyQt5 for interface design, with Python as programming language, Pytorch as deep learning framework, Intel(R) Core(TM) i5-10300H as processor (CPU), NVIDIA GeForce GTX 1650 as graphics processor (GPU); and Windows 10 as operating system. Qt Designer is the development interface of PyQt5, and the interface is mainly divided into five functional areas. The first part is the "Widget Box"; the second part is the "Dialog"; the third part is the object viewer; the fourth part is the property editor; and the fifth part is the signal editor. Figure 2 As shown in the figure, it is a diagram of the main functional structure of the system. The main functions of the rock crack detection system module based on deep learning are divided into: image selection, image detection, camera detection, photo taking and exit interface functions. The packaging tool used is the Pyinstaller library. The system is packaged using the Pyinstaller library to obtain the "Rock Crack Detection System Based on Deep Learning" application. Figure 3 Shown is a schematic diagram of a rock crack detection system based on deep learning.
[0186] Step (three): Based on image processing and image recognition, based on digital image technology, combined with the results of rock crack identification, select a suitable rock crack image feature extraction method to segment the cracks from the background and express them quantitatively. First, perform image segmentation on the rock crack image, using the Fuzzy C-Means (FCM) clustering algorithm. The Fuzzy C-Means clustering algorithm is also an algorithm based on the objective function. Assume that there is a rock image with cracks with a total of n pixels, that is, given a data set containing n pixels: x = {x1, x2, ..., xn}, the objective function of the Fuzzy C-Means clustering algorithm is defined as follows:
[0187]
[0188]
[0189] Where: u mn is pixel x n Relative to the mth cluster center z m The fuzzy membership degree, p is the fuzzy index, d mn For x n and cluster center z m The grayscale distance between .
[0190] To find the minimum value of the objective function J, the Lagrange multiplier method is used on the objective function while satisfying the constraints, and J is set to m and u mn The partial derivative of is 0, so we can get the membership matrix u mn , cluster center v m , the specific formula is as follows:
[0191]
[0192]
[0193] The specific steps of the Fuzzy C-means (FCM) clustering algorithm are as follows:
[0194] Step 1: First, input the data of rock mass crack image into the algorithm, set the number of rock mass crack categories to be clustered, fuzzy index p and initialize cluster center, etc.;
[0195] Step 2: Set the threshold for stopping iteration and the maximum number of iterations;
[0196] Step 3: Then update the membership degree and calculate the cluster center;
[0197] Step 4: Determine whether the conditions of the standardized measurement function are met. If so, stop the algorithm and output the membership degree and cluster center;
[0198] Step 5: Divide the image pixels according to the maximum membership principle to achieve image segmentation of rock fractures.
[0199] After the fuzzy C-means (FCM) clustering algorithm is used to segment, there are usually tiny noise points in the rock mass fracture image, so the image denoising operation is used to remove the tiny noise points. The image denoising algorithm used is a connected domain denoising method based on the least squares method, which is a commonly used image processing method. When there is not much noise in the image, and the size of the connected domain where the noise is located is significantly different from the size of the connected domain where the target is located, the least squares method can be used to obtain more accurate results. The expression of the least squares method is:
[0200]
[0201] After image denoising, crack skeleton line extraction can be performed, using the Zhang-Suen thinning algorithm. The purpose of the thinning algorithm is to refine the rough features in the image into finer lines or curves for subsequent image analysis and recognition. This algorithm can effectively refine the features in the image while maintaining the topological structure and shape characteristics of the image.
[0202] Step 1: Scan pixel by pixel. By scanning the image pixel by pixel, find the candidate pixels to be deleted. The judgment conditions include the black and white mode around the pixel and the connectivity of the pixel. For the foreground color, it will detect whether it meets the following conditions:
[0203] The current pixel is white, there is at least one black pixel in the 8-neighborhood of the current pixel P, above and below, left and right, upper left, lower left, upper right, and lower right, and the number of white pixels in the 8-neighborhood of the current pixel is between 2 and 6. If all the above conditions are met, then the current pixel is marked as black.
[0204] Step 2: Use the algorithm to traverse each pixel of the image again. For the foreground color, it will check whether it meets the following conditions:
[0205] The current pixel is white, and at least one of the 8 neighborhoods of the current pixel is black. The number of white pixels in the 8 neighborhoods of the current pixel is between 2 and 6, and the black pixels in the 8 neighborhoods of the current pixel are not connected, that is, not a connected component. If all the above conditions are met, then the current pixel is marked as black. After repeated iterations, based on the first step, the rules are continuously applied iteratively until no pixels can be deleted. After the iteration, the image after refinement is obtained.
[0206] After the rock mass crack image is refined, cracks may be broken. Therefore, it is necessary to connect the broken cracks to make them continuous and complete. The SURF feature point detection and matching method is used to stitch the cracks, including feature point detection, descriptor extraction, feature point matching, geometric transformation and image fusion. The specific steps are as follows:
[0207] The SURF algorithm determines the location of key points by detecting local extreme points in the image. If (x, y) is a pixel in the image, the image of the point can be defined as the sum of the values of all pixels in the rectangular area surrounded by the image origin and the point. The mathematical expression is as follows:
[0208]
[0209] Scale space theory is a method for detecting feature points in the SURF algorithm. It regards the determinant in the Hessian matrix as a discriminant to find the local maximum value. The Hessian matrix is constructed to generate stable mutation points in the image. The Hessian matrix can be expressed as:
[0210]
[0211] Where: σ is the Gaussian scale; (x, y) is the position of the image pixel; L xx (x, y, σ) is expressed as the convolution of the second-order Gaussian differential at the pixel point (x, y) of image I, and the same is true for the other L.
[0212] After determining the key points, the SURF algorithm generates descriptors for each key point, which can describe the gradient information and distribution of the area around the key point. Next, the key points in the image are matched, and a descriptor-based similarity measure is usually used to find the corresponding key points in the image. Through the matched key points, different transformation models can be used to estimate the geometric transformation relationship between images. Finally, based on the estimated geometric transformation relationship, the two images are spliced and fused into a complete image by superposition or mixing. The weighted average fusion algorithm is adopted, and the weighting factor is u. The mathematical expression of the weighted fusion of the pixel values of the image is as follows:
[0213]
[0214] The edge extraction of rock mass cracks is done by the Canny algorithm. In the Canny algorithm, a Gaussian filter is used. It is a linear filter obtained by sampling and normalizing a two-dimensional Gaussian function. The mathematical expression of the two-dimensional Gaussian function used by the Gaussian filter in the Canny algorithm is as follows:
[0215]
[0216] S=G×I
[0217] Wherein: G represents the Gaussian filter obtained by operating the two-dimensional Gaussian function; I represents the image to be detected; S represents the image after smoothing.
[0218] The magnitude and direction of the gradient of each pixel in the smoothed image are calculated. The vertical and horizontal components of the gradient of the pixel on the edge are as follows:
[0219]
[0220] Pixel gradient size g θ and direction θ are:
[0221]
[0222] The strong and weak edges are divided according to the given high and low thresholds, and the isolated weak edges are eliminated to obtain the final detection result. In the Canny algorithm, two high and low thresholds T1 and T2 are set. Points with gradient values greater than T1 are strong edge points, and vice versa are weak edge points.
[0223] The calculation of rock crack width is based on crack skeleton extraction and edge detection, and the crack width is calculated by the orthogonal skeleton line method. The orthogonal skeleton line method in this paper uses the K nearest neighbor method of a given point, and specifically the KD tree algorithm can be used. The KD tree algorithm is a tree data structure that stores points in a K-dimensional space for fast retrieval. The process of constructing a KD tree is to continuously divide the k-dimensional space with a hyperplane perpendicular to the coordinate axis to form a series of k-dimensional hyperrectangular regions. Among them, the method for calculating the distance between data points in the KD tree algorithm is the Euclidean distance, and its mathematical expression is as follows:
[0224]
[0225] Where: K is the dimension; d is the distance between data points, a i and i is the data point.
[0226] The general steps for finding the k nearest neighbors of a given point using the KD tree algorithm are:
[0227] First, you need to construct a KD tree for the data set. The process of constructing a KD tree involves selecting an axis and then dividing the data set into two subsets based on the axis. The left subtree contains data points less than the axis value, and the right subtree contains data points greater than the axis value. The subsets are then recursively divided until each subset contains only one data point. Starting from the root node, search recursively down the tree, comparing the coordinate value of the given point with the axis value of the current node to determine whether to search the left subtree or the right subtree. During the search process, you need to keep track of the k nearest neighbors. Once the k nearest neighbors of the target point are found, you need to backtrack to the parent node to check if there are closer points. If so, update the nearest neighbor set. When you backtrack to the root node, you can get the k nearest neighbors of the given point.
[0228] After the KD tree algorithm, the normal vector of the skeleton line is calculated by singular value decomposition (SVD), and then the two crack edge line points closest to the skeleton line normal vector are found, and the intersection of the line segment formed by these two points and the skeleton line normal vector is calculated. The definition of singular value decomposition (SVD) is as follows:
[0229] Suppose the bilinear function is: f(x,y)=x T Ay
[0230] Where x, y belong to R n×1 , A belongs to R n×n By introducing a linear change, the above formula can be expressed as: f(x,y)=(Uξ) T A(Vη)=ξ T U T AVη
[0231] Next, let S = U T AV, then the formula can be written as: f(x,y)=ξ T Sη
[0232] If U and V are both orthogonal matrices, S can be made into a diagonal matrix by using appropriate U and V. Specifically, it can be expressed as: S = diag(σ1,σ2,...,σ n )
[0233] Multiplying the left and right sides by U and VT can be expressed as: A = USV T
[0234] As mentioned above, this is the definition of singular value decomposition (SVD) given by Beltrami.
[0235] First, the geometric representation of the skeleton line is converted into a matrix form, for example, the endpoint coordinates of the skeleton line are represented as a matrix. Each row represents the coordinates of an endpoint, and each column represents a different dimension. The matrix representing the skeleton line is centered, and the coordinates of each dimension are subtracted from the mean of that dimension, so that the center of the data is located at the origin. The centered matrix is subjected to singular value decomposition, and the normal vector can usually be obtained from the result of singular value decomposition (SVD). The right singular vector corresponding to the minimum singular value can be used as the normal vector of the skeleton line representing the main direction.
[0236] After the cracks are skeletonized and spliced, the length of the crack skeleton line is calculated to be the crack width. The main methods for calculating the crack length are the neighborhood scanning method based on the skeleton curve and the approximate calculation method based on the first and last endpoints. This paper uses the neighborhood scanning method based on the skeleton curve. First, the cracks are skeletonized, and then the burrs on the crack skeleton line are processed by morphological operations; the global image is scanned to find the endpoints of the cracks, and the crack pixels are searched and counted according to the 8-neighborhood method with the endpoints as the starting point; the number of crack pixels is multiplied by the resolution to obtain the true length of the crack. The results obtained by this method are more accurate, but the processing process is more demanding. In the morphological processing process, the burr removal effect of the crack skeleton line directly affects the calculation results.
[0237] The calculation of the direction of rock fractures is based on the extraction of fracture skeleton lines. By marking each point on the fracture surface, calculating the tangent direction vector of its skeleton line, and then averaging all these skeleton line tangent direction vectors, the average vector representing the overall skeleton line tangent direction can be obtained. Using the average tangent direction vector, inverse trigonometric functions can be used to calculate the inclination of the fracture. Specifically, the arctan function (or other equivalent functions) can be used to determine the inclination by the x and y components of the given average tangent direction vector. This inclination is usually an azimuth angle relative to the horizontal plane, and the specific mathematical expression is as follows:
[0238]
[0239]
[0240] After solving the direction of the average tangent, the angle θ of the crack can be obtained through inverse trigonometric functions. The specific mathematical expression is as follows:
[0241]
[0242] The calculation of rock crack area is based on the effect diagram after image denoising operation and combined with the crack edge detection algorithm. First, the boundary of the crack is identified, and the pixel area of the crack is calculated using the image processing algorithm. Finally, the real area of the crack is calculated through the proportional relationship between the pixel area and the actual size. This crack area calculation method based on pixel area is suitable for cracks of various irregular shapes and can be achieved through digital image processing technology.
[0243] Step (IV): The development of rock mass structural surface is an important factor in judging the quality of engineering rock mass. The distribution of cracks in rock mass is one of the important factors affecting the blasting effect. Therefore, the rock mass cracks are taken as the research object, and the rock mass cracks in the blasting hole are analyzed. After obtaining the crack information in the hole, the charge structure can be designed according to the distribution and geometric information of the cracks. In the blasting process of air-spaced charge, the hole pressure can be expressed by the following formula:
[0244]
[0245] Where: P is the average pressure acting on the blast hole; Pe is the initial pressure; γ is the gas expansion index; le is the length of the charging section; la is the length of the air column.
[0246] The calculation of charge amount in blastholes has a great influence on blasting effect and blasting cost. In the case of continuous charge, the commonly used mathematical expression for calculating charge amount is calculated according to the following empirical formula:
[0247] Q=q×S×L
[0248] In the formula: Q is the total charge for one blast; q is the unit consumption of blasting explosives; S is the excavation cross-sectional area; L is the average depth of the blasthole.
[0249] For blastholes with discontinuous charge, the charge quantity Q can be calculated according to the following formula:
[0250]
[0251] Where: r is the radius of the blast hole; q is the unit consumption of blasting explosives; l e is the length of the charging section; K d is the radial uncoupling coefficient, K d =r b / r c , r b 、r c is the diameter of the blasthole and the diameter of the cartridge. The unit consumption of explosives q varies according to the different properties of the rock.
[0252] The borehole coupled charge refers to the radial uncoupled charge coefficient K d =1, K d =r b / r c , r b 、r c is the blasthole diameter and the blasthole charge diameter. When coupled charging, the impact load acting on the rock mass is as follows:
[0253]
[0254]
[0255] From the above two formulas, we can get:
[0256]
[0257] Where: P is the maximum pressure; P H is the detonation pressure of explosive; D e is the detonation velocity of explosive; e , m are the density of explosive and rock; Cp is the wave velocity of rock mass.
[0258] When radial air is not coupled with charge, the peak value of blasting load pressure of charge section is as follows:
[0259]
[0260] Where: P0 is the maximum pressure peak of the charging section; K d is the uncoupling coefficient; n is the pressure increase factor, n = 8; D e is the detonation velocity of explosive; e is the density of explosive.
[0261] Since the time of air expansion in the hole is very short, it can be determined that the pressure on the hole wall after the air fills the entire blasthole is the peak pressure of the air section. The formula is as follows:
[0262]
[0263] Where: P1 is the maximum pressure peak of the air segment; γ is the gas expansion index; l e is the length of the charging section; l is the depth of the blasthole.
[0264] The rock mass fracture ratio can be divided into body fracture ratio Kt1 and surface fracture ratio Kt2. Body fracture ratio Kt1 is the total fracture area contained in the unit volume of rock mass, and surface fracture ratio Kt2 is the total fracture length contained in the unit area of rock mass. The rock mass fracture ratio comprehensively reflects the development of the rock mass structural surface. The specific mathematical expression is as follows:
[0265]
[0266]
[0267] Where: Ai , l i are the area and trace length of the crack on the i-th structural plane respectively; V is the volume of the rock mass; A is the area of the rock mass section; n is the number of structural planes.
[0268] Rock mass integrity is an important indicator for evaluating rock mass quality. The rock mass quality index RQD is a commonly used method to indicate rock mass integrity. RQD is defined as the ratio of the cumulative length of the fractured core greater than or equal to 10 cm during drilling to the total length of drilling footage. The specific mathematical expression is as follows:
[0269]
[0270] Where: L i is the length of the ith core greater than or equal to 10 cm during drilling; L is the total length of drilling.
[0271] A new indicator RQD proposed by Azimian et al. in 2016 I Different types of rock masses are considered, including the scour area CW, the fracture area Fr, the crushing area Cr and the karst area K. The joint direction influence coefficient fi is introduced into the index, and the mathematical expression is as follows:
[0272]
[0273] Where: TL is the total length of drilling; CW is the length of the flushing core section; Fr is the length of the core less than 5 cm; Cr is the length of the broken section core; K is the length of the karst area core.
[0274] On the basis of RQD, considering the influence of cracks on rock mass quality evaluation, the rock mass crack ratio K was selected. t1 The surface crack rate K t2 Two indicators, refer to the new indicator RQD I , retaining the fracture area Fr and the crushing area Cr. These two indicators can obtain similar data by observing the distribution of cracks in the hole wall. Therefore, considering the influence of the above indicators, the new indicator LRQD based on the RQD method is as follows:
[0275]
[0276] Where: TL is the total length of drilling; Cr is the length of the core in the crushing section; Fr is the length of the core less than 5 cm; K t1 is the body crack rate; K t2 is the surface crack ratio; μ is the correction coefficient of the rock crack ratio; the preliminary judgment of the correction coefficient is: when the rock crack ratio is greater than 25, the correction coefficient μ is used to slow down the rate of increase of the rock crack ratio; when the rock crack ratio is less than 25, the correction coefficient μ is 1.
[0277] By comparing the new index LRQD and RQD for comparative analysis, the body crack rate K t1 , surface crack rate K t2 , fracture area Fr and crushing area Cr, the original RQD method was modified and adjusted, and the mathematical expression of the correction coefficient α was proposed as follows:
[0278]
[0279] The correction coefficient α is mainly used to correct the peak pressure P that the blasthole can actually withstand. By calculating the rock mass crack ratio and combining it with the on-site data for analysis, it is determined whether the charge is continuously charged and the ratio of the length of the air column in the blasthole to the charge length, so as to realize the charge structure design. According to the level type of the rock mass, the impact pressure of the rock mass of different levels is calculated, and then according to the distribution of rock mass cracks, the impact pressure is corrected to achieve the ideal effect. The formula is as follows:
[0280] P d =αP
[0281] Where: α is the correction coefficient; P is the pressure.
[0282] Step (IV): Use Matlab software to design a charge automatic matching system based on crack identification in the blasthole. Matlab provides a powerful GUI development toolbox that can quickly create a customized graphical interface. Users do not need to write tedious codes, but can build a GUI through simple drag-and-drop and configuration operations, such as Figure 4 As shown in the figure, it is the functional structure diagram of the system. Figure 5 The figure shows the schematic diagram of the system interface. In Matlab software, packaging executable programs can be done by using built-in functions.
[0283] Step (V): Establish an in-hole camera device. The hardware part of the in-hole camera device consists of an in-hole camera, in-hole camera accessories, an industrial all-in-one machine, cables, power supplies and chargers, power cables, converters and adapters, video capture cards, and equipment boxes. The software part of the in-hole camera device consists of a rock crack detection system based on deep learning and a charge intelligent matching design system based on crack identification in blastholes. Figure 6As shown, the connection relationship of the equipment is indicated by red lines and indicated by serial numbers. The parts of the red dotted line frame are the local diagram of the hole camera, the hole camera accessories and the equipment box from top to bottom. The specific connection relationship is: first connect the cable 3 to the hole camera 2, select appropriate accessories to install on the hole camera 2, connect the cable 3 and the power supply 5 through the first connecting line 4, and then connect the power supply 5 to the industrial integrated machine 1 to power the hole camera 2 and the industrial integrated machine 1. Connect the second connecting line 6, the converter 7 and the video capture card 8 in sequence to convert the analog signal of the hole camera 2 into a digital signal that can be processed by the industrial integrated machine 1. Finally, connect the converter 7 and the video capture card 8 to the industrial integrated machine 1 through appropriate connecting lines. At this time, the hole camera equipment is connected, and the designed program can be used on the industrial integrated machine 1 for research activities. When using, the cable 3 and the hole camera 2 are both waterproof equipment, and the waterproof ability of other parts is weak, so special attention should be paid. After use, turn off the industrial integrated machine 1 and disconnect the various components, and put them into the equipment box after simple cleaning.
[0284] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. Design method for crack identification and intelligent charge matching in underground engineering blasting holes, It is characterized in that The method comprises the following steps: S10. Obtain rock mass crack images and preprocess the rock mass crack images; establish a data set in a deep learning network model; S20. Build a deep learning network model; use the improved network model to perform image recognition on the processed image; S30. According to the rock mass fracture identification result of step S20, the fracture is segmented from the background and quantitatively expressed by extracting the identified fracture feature area, extracting and splicing the fracture skeleton and obtaining fracture set information; S40. According to the recognition result of step S30, the charge structure in the blasthole is analyzed, and an intelligent charge matching design system based on the recognition of cracks in the blasthole is established; S50. Establish an in-hole camera system.
2. The method for designing crack identification and intelligent charge matching in underground engineering blasting holes according to claim 1, It is characterized in that Step S30 includes the following steps: S31. Using fuzzy C-means clustering algorithm to perform image segmentation operation on rock mass crack image; S32. Using the connected domain of the least squares method to remove tiny noise points from the image; S33. Using Zhang-Suen thinning algorithm to extract crack skeleton lines and thin the crack image; S34. Use the SURF algorithm to detect and match feature points to stitch the cracks, including feature point detection, descriptor extraction, feature point matching, geometric transformation and image fusion; use the weighted average fusion algorithm to stitch and fuse the two images into a complete image by superposition or mixing; S35. Use Canny algorithm to extract edge features of rock crack images; The crack width is calculated using the orthogonal skeleton algorithm; The crack length is calculated using the neighborhood scanning method based on the skeleton curve; The steps to calculate the crack area are: S351. calibrate the image according to the scale mark or other known size information in the image; S352. Use edge detection algorithm to extract crack contours and mark crack areas; S353. Use the pixel counting function in the image processing software or programming language to calculate the number of pixels inside the crack region; that is, the pixel area of the crack region; S354. Multiply the pixel area by the actual area of each pixel to obtain the actual area of the crack.
3. The method for designing crack identification and charge intelligent matching in underground engineering blasting holes according to claim 2, It is characterized in that Step S31 includes the following steps: S311. First, input the rock mass crack image data into the algorithm, set the number of clustered rock mass crack categories, fuzzy index p and initialize cluster centers, etc.; S312. Set the threshold for stopping iteration and the maximum number of iterations; S313. Then update the membership degree and calculate the cluster center; S314. Determine whether the conditions of the standardized measurement function are met. If so, stop the algorithm and output the membership degree and cluster center; S315. Divide the image pixels according to the maximum membership principle to achieve image segmentation of rock mass fissures; Its expression is: Given a data set containing n pixels: x = {x1, x2, ..., xn}, the objective function of the fuzzy C-means (FCM) clustering algorithm is defined as follows: Where: u mn is pixel x n Relative to the mth cluster center z m The fuzzy membership degree, p is the fuzzy index, d mn For x n and cluster center z m The grayscale distance between Find the minimum value of the objective function J, and use the Lagrange multiplier method on the objective function while satisfying the constraints, so that J is m and u mn The partial derivative of is 0, so we can get the membership matrix u mn , cluster center v m , the specific expression is as follows:
4. The method for designing crack identification and charge intelligent matching in underground engineering blasting holes according to claim 2, It is characterized in that Step 32 includes the following steps: S321. Find all connected domains in the rock mass fracture image and obtain the number of pixels contained in each of their connected domains; S322. Determine the threshold range of the algorithm by using the least squares method; S323. Determine the threshold range of the algorithm according to the middle value of the threshold; S324. Processing the connected domain of the rock mass fracture image according to the threshold range; The expression of the least squares method is: Where: a i is the coordinate information of the pixel point; Step S33 includes the following steps: S331: Scanning pixel by pixel, by scanning the image pixel by pixel, to find candidate pixels to be deleted; judging conditions include the black and white mode around the pixel and the connectivity of the pixel, and for the foreground color, checking whether the following conditions are met: The current pixel is white, there is at least one black pixel in the 8-neighborhood of the current pixel P, and the number of white pixels in the 8-neighborhood of the current pixel is between 2 and 6. If all the above conditions are met, then the current pixel is marked as black; S332: The algorithm is used to traverse each pixel of the image again, and for the foreground color, it is checked whether it meets the following conditions: The current pixel is white, and in the 8-neighborhood of the current pixel, at least one is black, and in the 8-neighborhood of the current pixel, the number of white pixels is between 2 and 6, and the black pixels in the 8-neighborhood of the current pixel are not connected, that is, it is not a connected component; if all the above conditions are met, then the current pixel is marked as black; after repeated iterations, based on the first step, the rules are continuously applied iteratively until no pixels can be deleted; after the iteration, the image after refinement is obtained.
5. The method for designing crack identification and charge intelligent matching in underground engineering blasting holes according to claim 2, It is characterized in that Step S34 includes the following steps: S341. Determine the key point position by detecting the local extreme point in the image. If (x, y) is a pixel point in the image, the image of the point can be defined as the sum of the values of all pixels in the rectangular area surrounded by the image origin and the point. The mathematical expression is as follows: S342. Construct the Hessian matrix to generate stable mutation points in the image. The Hessian matrix can be expressed as: Where: σ is the Gaussian scale; (x, y) is the position of the image pixel; L xx (x, y, σ) represents the convolution of the second-order Gaussian derivative at the pixel (x, y) of image I, and the same is true for the other L; S343. Generate a descriptor for each key point to describe the gradient information and distribution of the area around the key point; match the key points in the image and use a similarity metric based on the descriptor to find the corresponding key points in the image; Through the matched key points, different transformation models can be used to estimate the geometric transformation relationship between images; S344. According to the estimated geometric transformation relationship, the two images are spliced and fused into a complete image by superposition or mixing. A weighted average fusion algorithm is adopted, and the weighting factor is u. The mathematical expression of the weighted fusion of the pixel values of the images is as follows: Where: u is the weighting factor; x and y are coordinates; f 1 、f 2 Represents an image.
6. The method for identifying cracks in underground engineering blasting holes and intelligent matching of charges according to claim 2, It is characterized in that The mathematical expression of the two-dimensional Gaussian function used by the Gaussian filter in the Canny algorithm is as follows: S=G×I Where: G represents the Gaussian filter obtained by operating the two-dimensional Gaussian function; I represents the image to be detected; S represents the image after smoothing; The magnitude and direction of the gradient of each pixel in the smoothed image are calculated. The components of the gradient of the pixel on the edge in the vertical and horizontal directions (x, y) are as follows: Pixel gradient size g θ and direction θ are: According to the given high and low thresholds, the strong and weak edges are divided, and the isolated weak edges are eliminated to obtain the final detection result; set the high and low thresholds T 1 , T 2 ; Gradient value ratio T 1 Large points are strong edge points, and smaller points are weak edge points. The orthogonal skeleton method uses the KD tree algorithm, and the expression is as follows: Where: K is the dimension; d is the distance between data points, a i and i is a data point; After using the KD tree algorithm to find the given k nearest neighbor points, the normal vector of the skeleton line is calculated by singular value decomposition, and then the two crack edge line points closest to the skeleton line normal vector are found, and the intersection of the line segment formed by these two points and the skeleton line normal vector is calculated; the expression of singular value decomposition is as follows: Suppose the bilinear function is: f(x,y)=x T Oh Where x, y belong to R n×1 , A belongs to R n×n , by introducing linear changes, the bilinear function can be expressed as: f(x,y)=(Uξ) T A(Vη)=ξ T U T AVh Next command: S=U T OF Then the bilinear function can be written as: f(x,y)=ξ T S If U and V are both orthogonal matrices, S can be made into a diagonal matrix by using appropriate U and V, which can be specifically expressed as: S=diag(σ 1 ,s 2 ,...,s n ) Multiplying the left and right sides by U and VT can be expressed as: A=USV T ; The crack length is calculated using the neighborhood scanning method based on the skeleton curve; the steps are as follows: S351. First, the crack is skeletonized, and then the burr part on the crack skeleton line is processed by morphological operation; S352. Scan the global image to find the end point of the crack, and use the end point as the starting point to search for crack pixels according to the 8-neighborhood method and count the number; S353. Multiplying the obtained number of crack pixels by the resolution to obtain the actual length of the crack; The calculation of the rock mass fracture direction is based on the extraction of the fracture skeleton line. By marking each point on the fracture surface, calculating the skeleton tangent direction vector, and then taking the average of all these skeleton tangent direction vectors, the average vector representing the overall skeleton tangent direction can be obtained. The average tangent direction vector is used to determine the inclination angle through the given x and y components of the average tangent direction vector. The mathematical expression is as follows: Where: a is a vector; g 1 , g 2 、f 1 、f 2 is the coordinate point; n is the total tangent direction vector; After solving the direction of the average tangent, the angle θ of the crack can be obtained through inverse trigonometric functions. The specific mathematical expression is as follows:
7. The method for intelligent charge matching based on crack identification in underground engineering blasting holes according to claim 1, It is characterized in that Step S10 includes the following steps: S11. Use the weighted average method to grayscale the obtained rock mass crack color image, and perform weighted average distribution on the values of R, G and B. The expression is as follows: R=G=B=W R R+W G G+W B B; Where: R, G and B are the three component values of the image, and W R , W G and W B is the weight value of R, G and B. When W R =0.3,W G =0.59,W B =0.11, the required grayscale image can be obtained; S12. The CA attention mechanism is used to obtain the attention of the image width and height. The input image is globally pooled in the horizontal and vertical directions respectively, and then the feature maps in the horizontal and vertical directions are obtained in turn. The calculation formula is as follows: Where: x c is the input feature vector; h is the image height; w is the image width; S13. The horizontal and vertical feature maps are spliced together in the spatial direction, and then a 1×1 convolution operation is performed to reduce the dimension. The feature map after the BN layer processing is sent to the Sigmoid activation function to obtain the feature map f. The calculation formula is as follows: f=σ(F 1 ([With h ,With w ])) Where: F 1 is the result of BN layer operation; After a 1×1 convolution kernel, we get a feature map with the same number of channels as the original one, and decompose f into feature maps f in the spatial dimension. h and f w , and finally the attention weights in the horizontal and vertical directions are obtained through the Sigmoid activation function. The calculation formula is as follows: g h =σ(F h (f h )) g w =σ(F w (f w )); Where: F h and F w f h and f w Transform into a transformation function of the same dimension as the output; S14. Through the multiplication weighted operation on the original feature map, the final feature map with attention weights in both the horizontal and vertical directions will be obtained. The calculation formula is as follows: and c =x c ×g h ×g w ; Where: x c is the input feature map dimension.
8. The method for designing crack identification and intelligent charge matching in underground engineering blasting holes according to claim 1, It is characterized in that Step S40 includes the following steps: S41. Calculate the borehole pressure; the expression is: Where: P is the average pressure acting on the blast hole; P e is the initial pressure; γ is the gas expansion index; l e is the length of the charging section; l a is the length of the air column; S42. Calculate the charge amount, the expression is: Q = q × S × L; Where: Q is the total charge for one blast; q is the unit consumption of blasting explosives; S is the excavation cross-sectional area; L is the average depth of the blasthole; S43. For blastholes with discontinuous charge, the charge quantity Q can be calculated according to the following formula: Where: r is the radius of the blast hole; q is the unit consumption of blasting explosives; l e is the length of the charging section; K d is the radial uncoupling coefficient, K d =r b / r c , r b 、r c is the diameter of the blast hole and the diameter of the charge roll; S44. When calculating the coupled charge, the impact load acting on the rock mass is expressed as follows: We can get: Where: P is the maximum pressure; P H is the detonation pressure of explosive; D e is the detonation velocity of explosive; e , m is the density of explosive and rock; C p is the rock mass wave velocity; When radial air is not coupled with charging, the peak value of the blasting load pressure of the charging section is as follows: Where: P 0 K is the maximum pressure peak of the charging section; d is the uncoupling coefficient; n is the pressure increase factor, n = 8; D e is the detonation velocity of explosive; e is the density of explosive; Since the time of air expansion in the hole is very short, it can be determined that the pressure on the hole wall after the air fills the entire blasthole is the peak pressure of the air section. The formula is as follows: Where: P 1 is the maximum pressure peak of the air segment; γ is the gas expansion index; l e is the length of the charging section; l is the depth of the blasthole; S45. Calculate K t1 The surface crack ratio K t2 , the expression is: Where: A i , l i are the area and trace length of the crack of the i-th structural plane respectively; V is the volume of the rock mass; A is the area of the rock mass section; n is the number of structural planes; S46. Calculation of rock mass quality index and rock mass volume joint number J v , the expression is as follows: Where: L i is the length of the ith core greater than or equal to 10 cm in the drilling; L is the total length of the drilling; The joint direction influence coefficient fi is introduced, and the mathematical expression is as follows: Where: TL is the total length of drilling; CW is the length of the scouring core section; Fr is the length of the core less than 5 cm; Cr is the length of the broken section core; K is the length of the karst area core; Considering the influence of the above indicators, the new indicator LRQD based on the RQD method is as follows: Where: TL is the total length of drilling; Cr is the length of the core in the crushing section; Fr is the length of the core less than 5 cm; K t1 is the body crack rate; K t2 is the surface crack ratio; μ is the correction coefficient of the rock mass crack ratio; it is preliminarily determined that when the rock mass crack ratio is greater than 25, the correction coefficient μ is used to slow down the increase rate of the rock mass crack ratio; when the rock mass crack ratio is less than 25, the correction coefficient μ is 1; S47. By comparing the new index LRQD and RQD, the body crack rate K t1 , surface crack rate K t2 , fracture area Fr and crushing area Cr, the original RQD method was modified and adjusted, and the mathematical expression of the correction coefficient α was proposed as follows: According to the distribution of rock mass cracks, the impact pressure is corrected to achieve the ideal effect. The formula is as follows: P d =αP Where: α is the correction coefficient; P is the pressure.
9. The method for designing crack identification and intelligent charge matching in underground engineering blasting holes according to claim 1, It is characterized in that In the hole camera system, the connection steps are: S51. Connect the cable (3) to the camera in the hole (2), select suitable accessories to install the camera in the hole (2), and connect the cable (3) and the power supply (5) through the first connecting line (4); S52. Connect the power supply (5) and the industrial integrated machine (1) to power the hole camera (2) and the industrial integrated machine (1), and connect the second connecting line (6), the converter (7) and the video acquisition card (8) in sequence to convert the analog signal of the hole camera (2) into a digital signal that can be processed by the industrial integrated machine (1); S53. Connect the converter (7) and the video capture card (8) to the industrial integrated machine (1) through a suitable connecting cable. At this time, the hole camera device is connected and the designed program can be used on the industrial integrated machine (1) for research activities; when in use, the cable (3) and the hole camera (2) are both waterproof devices; S54. After use, turn off the industrial integrated machine (1) and disconnect all components, clean it briefly and put it in the equipment box.
Citation Information
Patent Citations
Deeply-buried tunnel local area difference directional fracture control blasting method
CN114547721A
Deep tunnel surrounding rock blasting damage test system and test method
CN115791803A
Analysis method of rock mass fracture state and training method of rock mass fracture recognition model
CN116630773A
Tunnel blasting method based on intelligent identification of surface fracture and rock mass internal lithology
CN116912300A
Intelligent grading method for tunnel surrounding rock based on image information
CN117853323A
Cited By
Method and system for quantizing integrity of rasterized rock mass based on excavation and drilling images
CN119991592A
Underwater blasting construction method and system based on real-time video image processing
CN120689734A
Surface mine blasting parameter optimization method and system based on karst fissure recognition
CN120970411A
Karst fissure identification-based open-pit mine blasting parameter optimization method and system
CN120970411B
Tunnel blast hole residual length measuring method and system based on key point detection
CN121685613A
