A coal-rock identification method for an anti-bumping drilling robot during drilling and pressure relief

Abstract

A method for identifying coal and rock in the process of drilling and decompression of an anti-collision drilling robot is disclosed. The method uses a high-definition camera to collect the original drill cuttings image discharged in the process of drilling of the anti-collision drilling robot. After denoising the original drill cuttings image through an improved adaptive regional weighted guided filtering model, the denoised drill cuttings image is output. The denoised drill cuttings image is clustered and segmented through a local information weighted intuitive fuzzy C-means clustering algorithm. The pixels in the input image are divided into two categories, coal or rock. The two corresponding cluster centers are c1 or c2, respectively. At this time, the clustering ends. Then, whether it is coal or rock is obtained by comparing the cluster center c1 and c2. In actual operation, the drill cuttings discharged by the drilling robot in the process of drilling and decompression can directly reflect the coal and rock drilling state of the current drill rod. The method can realize the coal and rock identification in the process of drilling and decompression by processing the drill cuttings image, and improve the coal and rock identification accuracy and efficiency.

Description

Technical Field

[0001] The invention relates to a coal-rock identification method, in particular to a coal-rock identification method in a pressure relief process of drilling by an anti-bumping drilling robot, belonging to the technical field of coal-rock identification in the pressure relief process of drilling. Background Art

[0002] Coal-rock identification during the decompression process of the anti-bumping drilling robot is a global problem in the coal field and a key technology for robotic operations in the decompression area of ​​drilling. In the research on coal-rock identification methods, domestic and foreign scholars mainly focus on radioactive ray detection, vibration detection, drilling pressure, drilling torque, etc. However, due to the complex and changeable coal-rock characteristics in underground mine tunnels, the complex structure of the drilling rig equipment, and the high noise of the sensor signal, the actual application effect of the above methods is not ideal. Therefore, it is urgent to provide new technologies and methods to solve the problem of accurate coal-rock identification during the decompression process of drilling. Summary of the invention

[0003] The purpose of the present invention is to provide a method for identifying coal and rock during the drilling and pressure relief process of an anti-collision drilling robot. The drilling robot can directly reflect the current coal and rock drilling status of the drill rod through the drill cuttings discharged during the drilling and pressure relief process, thereby realizing accurate identification of coal and rock during the drilling and pressure relief process.

[0004] In order to achieve the above object, the present invention provides a method for identifying coal and rock during the pressure relief process of an anti-bumping drilling robot, comprising the following steps:

[0005] Step 1: Use a high-definition camera to collect the original drill cuttings images discharged during the drilling process of the anti-collision drilling robot. Aiming at the uneven noise existing in the coal mine underground images, an improved adaptive regional weighted guided filtering model is designed. The scale of Gaussian filtering is adaptively selected according to the noise intensity at different positions of the original drill cuttings image, and the relationship between noise filtering and edge protection is further balanced to achieve sufficient filtering of noise components and unimportant detail information; after denoising the original drill cuttings image, the denoised drill cuttings image is output.

[0006] Step 2: De-noised drill cuttings image Cluster segmentation is performed. After the improved adaptive regional weighted guided filtering, there is still a certain degree of residual noise in the drill cuttings image, which is easy to reduce the accuracy of the image segmentation result. Therefore, the local information weighted intuitive fuzzy C-means clustering algorithm is used to realize the cluster segmentation of the denoised image. The pixels in the input image are divided into two categories: coal or rock. The corresponding two cluster set centers are c1 or c2 respectively. At this time, the clustering ends.

[0007] Step 3: Coal and rock identification. If the cluster center c1>c, the image pixel corresponding to the cluster center c1 is rock, and the image pixel corresponding to the cluster center c2 is coal; otherwise, the image pixel corresponding to the cluster center c1 is coal, and the image pixel corresponding to the cluster center c2 is rock.

[0008] The steps of denoising the original drill cuttings image in step 1 of the present invention are as follows:

[0009] Step 1.1: Guidance image for improved adaptive region-weighted guided filtering

[0010] Where G is the input original cuttings image;

[0011] ω is the adaptive weighting factor;

[0012] is the image processed by Gaussian filter;

[0013] Define the image noise factor ξ:

[0014]

[0015] In the formula, is the grayscale variance of the original drill cuttings image G;

[0016] is the image processed by Gaussian filter Grayscale variance of ;

[0017] is a positive constant to ensure that the image noise factor ξ does not appear distorted. The value of is defined as (1 / L 2 ), L is the gray level of the original cuttings image G;

[0018] On this basis, the adaptive weighting factor ω is defined as follows:

[0019]

[0020] Where: τ is the control parameter, which is set according to actual needs;

[0021] In the area without noise or with low noise intensity, the image noise degree factor ξ tends to 1. In order to avoid the Gaussian filter from over-smoothing the original image, thereby damaging the image edge information, the adaptive weighting factor ω tends to 0, thereby avoiding the loss of key information. In the area with high noise intensity, excessive noise will cause serious damage to the image information. At this time, the adaptive weighting factor ω tends to 1, so that the image smoothed by the Gaussian filter plays a major role in the guided image, reducing the impact of noise on the output image as much as possible, and increasing the noise smoothing effect.

[0022] Step 1.2: In the improved adaptive region weighted guided filtering process, the cost function E is defined as follows:

[0023]

[0024] Where p is the pixel in the image;

[0025] p′ is a square window θ with radius r in the image r The center of (p′);

[0026] is the edge perception factor;

[0027] λ is the penalty coefficient; it can be used for larger Punish, avoid If it is too large, the output image will be infinitely close to the original image, causing the filter to fail;

[0028] By minimizing the cost function shown in equation (3), we can obtain and is the gain factor, is the bias factor, as shown in formula (4),

[0029]

[0030] In formula (4), It is the product of the pixel grayscale values ​​of the two image matrices.

[0031] for The mean grayscale value of pixels in a window with a radius of r;

[0032] for The mean grayscale value of pixels in a window with a radius of r;

[0033] μ G,r (p′) is the grayscale mean of the pixels in the window of radius r;

[0034] is the grayscale variance of the pixels in the window with radius r;

[0035] Step 1.3: In the defined dimension θ r In the image window (p′), the denoised cuttings image is output It can be obtained by using the improved adaptive region weighted guided filtering model from the guidance image, as shown in formula (5):

[0036]

[0037] Where: and is the dimension θ r (p′) in the image window and The average value of can be obtained by formula (6):

[0038]

[0039] Where: |θ r (p′)| is the size of θ r The number of pixels contained in the window of (p′).

[0040] In step 2 of the present invention, the denoised cuttings image is clustered by using the local information weighted intuitionistic fuzzy C-means clustering algorithm. The clustering segmentation steps are as follows:

[0041] Step 2.1: Introducing uncertainty into the local information weighted intuitionistic fuzzy C-means clustering algorithm This achieves a balance between image noise filtering and image detail preservation during image segmentation; at the same time, an adaptive local information weight κ is introduced ij , local information H ij , to balance the impact of local information on fuzzy segmentation and avoid over-reliance on local information H in low-noise areas ij The influence of membership degree on clustering results is ignored. On the contrary, in the part with more noise, local information H ij The core of the local information weighted intuitionistic fuzzy C-means clustering algorithm is to solve the membership of the known data individuals to the cluster set center by minimizing the objective function J(U,V), as shown in formula (7).

[0042]

[0043] In the formula, sample x j Belongs to the sample data set X = {x1, x2, ..., x N};

[0044] c is the number of cluster sets selected according to the requirements;

[0045] N is the number of individuals in the sample data set;

[0046] m is a fuzzy constant (usually 2);

[0047] For sample x j For the composite membership of the i-th cluster set;

[0048] is the cluster center and sample x of each cluster set j The distance between

[0049] is the cluster center of the ith cluster set, and the calculation update formula is shown in formula (8):

[0050]

[0051] In order to avoid the reduction of the importance of the standard membership due to the replacement of the standard fuzzy membership by the intuitive fuzzy membership, the composite membership relationship is defined as follows in the local information weighted intuitive fuzzy C-means clustering algorithm:

[0052]

[0053] In the formula, μ i,j is the membership degree of fuzzy C-means clustering considering local information;

[0054] is the membership degree of intuitionistic fuzzy C-means clustering, and the calculation method is as follows:

[0055]

[0056]

[0057] In formula (7) and formula (10), the local information H ij The relationship between image details and image noise can be balanced, and is defined as follows:

[0058]

[0059] In the formula, 1≤i≤c, 1≤j≤N, N j is the number of neighborhood pixels in a 3×3 window centered on pixel j;

[0060] k is the kth pixel in a 3×3 window centered at pixel j;

[0061] d j,k is the spatial Euclidean distance between pixel k and pixel j;

[0062] μ i,k is the sample x k For the membership degree of the i-th cluster set;

[0063] In formula (7), the local information weighting parameter κ ij The calculation is as follows:

[0064]

[0065] In the formula, σ j 2 is the variance of the sample data in the jth 3×3 window;

[0066] For all σ j 2 The mean of ij The result is distorted, and ρ = 10 is introduced -3 η to ensure κ ij The result is valid;

[0067] η is the denoised cuttings image Gray levels;

[0068] In formula (7), the uncertainty The calculation formula is as follows:

[0069]

[0070] In the formula, α is defined as the uncertainty parameter, and its value is generally between 0.8 and 0.9;

[0071] Step 2.2: Calculate the objective function according to formula (7), set the objective function accuracy ε of the algorithm iteration, and use formula (7) to calculate the objective function values ​​J(t) and J(t+1) with iteration times t and t+1 respectively. If |J(t)-J(t+1)|<ε, the iteration ends and the cluster centers c1 and c2 are output. Otherwise, jump to step 2.1 and update the membership by formula (8) and formula (9) and cluster centers Until the iteration ends;

[0072] Step 2.3: According to the obtained membership matrix, take the class corresponding to the maximum sample membership as the result of sample clustering, and divide the pixels in the input image into two categories: coal or rock. The corresponding two cluster set centers are c1 or c2 respectively, and the clustering ends at this time.

[0073] Compared with the prior art, the present invention uses a high-definition camera to collect the original drill cuttings image discharged during the drilling process of the anti-collision drilling robot, and outputs the denoised drill cuttings image after denoising the original drill cuttings image. The denoised drill cuttings image is clustered using the local information weighted intuitionistic fuzzy C-means clustering algorithm. Cluster segmentation is performed to divide the pixels in the input image into two categories, coal or rock, and the corresponding two cluster centers are c1 or c2 respectively. At this time, clustering ends, and then the cluster centers c1 and c2 are compared to determine whether it is coal or rock. Specifically, when the cluster center c1>c2, the image pixel point corresponding to the cluster center c1 is rock, and the image pixel point corresponding to the cluster center c2 is coal; otherwise, the image pixel point corresponding to the cluster center c1 is coal, and the image pixel point corresponding to the cluster center c2 is rock; in actual operation, the drill cuttings discharged by the drilling robot during the drilling and pressure relief process can directly reflect the coal and rock drilling status of the current drill rod. Therefore, the present invention can realize coal and rock recognition during the drilling and pressure relief process by processing the drill cuttings image, thereby improving the accuracy and efficiency of coal and rock recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is the flow chart of coal rock identification of the present invention. DETAILED DESCRIPTION

[0075] The present invention will be further described below in conjunction with the accompanying drawings.

[0076] like Figure 1 As shown, a method for identifying coal and rock during the pressure relief process of an anti-bumping drilling robot comprises the following steps:

[0077] Step 1: Use a high-definition camera to collect the original drill cuttings images discharged during the drilling process of the anti-collision drilling robot. Aiming at the uneven noise existing in the coal mine underground images, an improved adaptive regional weighted guided filtering model is designed. The scale of Gaussian filtering is adaptively selected according to the noise intensity at different positions of the original drill cuttings image, and the relationship between noise filtering and edge protection is further balanced to achieve sufficient filtering of noise components and unimportant detail information; after denoising the original drill cuttings image, the denoised drill cuttings image is output.

[0078] Step 2: De-noised drill cuttings image Cluster segmentation is performed. After the improved adaptive regional weighted guided filtering, the drill cuttings image still has a certain degree of residual noise, which can easily reduce the accuracy of the image segmentation result. Therefore, the local information weighted intuitive fuzzy C-means clustering algorithm is used to realize the cluster segmentation of the denoised image. The pixels in the input image are divided into two categories: coal or rock. The corresponding two cluster set centers are c1 or c2 respectively. At this time, the clustering ends;

[0079] Step 3: Coal and rock identification. If the cluster center c1>c, the image pixel corresponding to the cluster center c1 is rock, and the image pixel corresponding to the cluster center c2 is coal; otherwise, the image pixel corresponding to the cluster center c1 is coal, and the image pixel corresponding to the cluster center c2 is rock.

[0080] The steps of denoising the original drill cuttings image in step 1 are as follows:

[0081] Step 1.1: Design guidance image for improved adaptive region-weighted guided filtering

[0082] Where G is the input original cuttings image;

[0083] ω is the adaptive weighting factor;

[0084] is the image processed by Gaussian filter;

[0085] Define the image noise factor ξ:

[0086]

[0087] In the formula, is the grayscale variance of the original drill cuttings image G;

[0088] is the image processed by Gaussian filter Grayscale variance of ;

[0089] is a positive constant to ensure that the image noise factor ξ does not appear distorted. The value of is defined as (1 / L 2 ), L is the gray level of the original cuttings image G;

[0090] On this basis, the adaptive weighting factor ω is defined as follows:

[0091]

[0092] Where: τ is the control parameter, which is set according to actual needs.

[0093] In the area without noise or with low noise intensity, the image noise degree factor ξ tends to 1. In order to avoid the Gaussian filter from over-smoothing the original image, thereby damaging the image edge information, the adaptive weighting factor ω tends to 0, thereby avoiding the loss of key information. In the area with high noise intensity, excessive noise will cause serious damage to the image information. At this time, the adaptive weighting factor ω tends to 1, so that the image smoothed by the Gaussian filter plays a major role in the guided image, reducing the impact of noise on the output image as much as possible, and increasing the noise smoothing effect.

[0094] Step 1.2: In the improved adaptive region weighted guided filtering process, the cost function E is defined as follows:

[0095]

[0096] Where p is the pixel in the image;

[0097] p′ is a square window θ with radius r in the image r The center of (p′);

[0098] is the edge perception factor;

[0099] λ is the penalty coefficient, which can be used for larger Punish, avoid If it is too large, the output image will be infinitely close to the original image, causing the filter to fail;

[0100] By minimizing the cost function shown in equation (3), we can obtain and is the gain factor, is the bias factor, as shown in formula (4),

[0101]

[0102] In formula (4), It is the product of the pixel grayscale values ​​of the two image matrices.

[0103] for The mean grayscale value of pixels in a window with a radius of r;

[0104] for The mean grayscale value of pixels in a window with a radius of r;

[0105] μ G,r (p′) is the grayscale mean of the pixels in the window of radius r;

[0106] is the grayscale variance of the pixels in the window with radius r;

[0107] Step 1.3: In the defined dimension θ r In the image window (p′), the denoised cuttings image is output It can be obtained by using the improved adaptive region weighted guided filtering model from the guidance image, as shown in formula (5):

[0108]

[0109] Where: and is the dimension θ r (p′) in the image window and The average value of can be obtained by formula (6):

[0110]

[0111] Where: |θ r (p′)| is the size of θ r The number of pixels contained in the window of (p′).

[0112] In step 2, the denoised drill cuttings image is clustered using the local information weighted intuitionistic fuzzy C-means clustering algorithm. The clustering segmentation steps are as follows:

[0113] Step 2.1: Introducing uncertainty into the local information weighted intuitionistic fuzzy C-means clustering algorithm This achieves a balance between image noise filtering and image detail preservation during image segmentation; at the same time, an adaptive local information weight κ is introduced ij , local information H ij , to balance the impact of local information on fuzzy segmentation and avoid over-reliance on local information H in low-noise areas ij The influence of membership degree on clustering results is ignored. On the contrary, in the part with more noise, local information H ij The core of the local information weighted intuitionistic fuzzy C-means clustering algorithm is to solve the membership of the known data individuals to the cluster set center by minimizing the objective function J(U,V), as shown in formula (7).

[0114]

[0115] In the formula, sample x j Belongs to the sample data set X = {x1, x2, ..., x N};

[0116] c is the number of cluster sets selected according to the requirements (in the present invention, the drill cuttings image contains two categories, coal and rock, so c=2);

[0117] N is the number of individuals in the sample data set;

[0118] m is a fuzzy constant (usually 2);

[0119] For sample x j For the composite membership of the i-th cluster set;

[0120] is the cluster center and sample x of each cluster set j The distance between

[0121] is the cluster center of the ith cluster set, and the calculation update formula is shown in formula (8):

[0122]

[0123] In order to avoid the reduction of the importance of the standard membership due to the replacement of the standard fuzzy membership by the intuitive fuzzy membership, the composite membership relationship is defined as follows in the local information weighted intuitive fuzzy C-means clustering algorithm:

[0124]

[0125] In the formula, μ i,j is the membership degree of fuzzy C-means clustering considering local information;

[0126] is the membership degree of intuitionistic fuzzy C-means clustering, and the calculation method is as follows:

[0127]

[0128] In formula (7) and formula (10), the local information H ij The relationship between image details and image noise can be balanced, and is defined as follows:

[0129]

[0130] In the formula, 1≤i≤c, 1≤j≤N, N j is the number of neighborhood pixels in a 3×3 window centered on pixel j;

[0131] k is the kth pixel in a 3×3 window centered at pixel j;

[0132] d j,k is the spatial Euclidean distance between pixel k and pixel j;

[0133] μ i,k is the sample x k For the membership degree of the i-th cluster set;

[0134] In formula (7), the local information weighting parameter κ ij The calculation is as follows:

[0135]

[0136] In the formula, σ j 2 is the variance of the sample data in the jth 3×3 window;

[0137] For all σ j 2 The mean of ij The result is distorted, and ρ = 10 is introduced -3 η to ensure κ ij The result is valid;

[0138] η is the denoised cuttings image Gray levels;

[0139] In formula (7), the uncertainty The calculation formula is as follows:

[0140]

[0141] In the formula, α is defined as the uncertainty parameter, and its value is generally between 0.8 and 0.9;

[0142] Step 2.2: Calculate the objective function according to formula (7), set the objective function accuracy ε of the algorithm iteration, and use formula (7) to calculate the objective function values ​​J(t) and J(t+1) with iteration times t and t+1 respectively. If |J(t)-J(t+1)|<ε, the iteration ends and the cluster centers c1 and c2 are output. Otherwise, jump to step 2.1 and update the membership by formula (8) and formula (9) and cluster centers Until the iteration ends;

[0143] Step 2.3: According to the obtained membership matrix, take the class corresponding to the maximum sample membership as the result of sample clustering, and divide the pixels in the input image into two categories: coal or rock. The corresponding two cluster set centers are c1 or c2 respectively, and the clustering ends at this time.

Claims

1. A method for identifying coal and rock during the pressure relief process of an anti-bumping drilling robot, characterized in that: The following steps are involved: Step 1: Use a high-definition camera to collect the original drill cuttings image discharged during the drilling process of the anti-collision drilling robot. After denoising the original drill cuttings image through the improved adaptive regional weighted guided filtering model, the denoised drill cuttings image is output. Step 2: De-noised drill cuttings image Perform cluster segmentation, and realize cluster segmentation of denoised images through local information weighted intuitionistic fuzzy C-means clustering algorithm. Pixels in the input image are divided into two categories: coal or rock. The corresponding two cluster set centers are c1 or c2 respectively. At this time, clustering ends; Step 3: Coal and rock identification. If the cluster center c1>c2, the image pixel corresponding to the cluster center c1 is rock, and the image pixel corresponding to the cluster center c2 is coal; otherwise, the image pixel corresponding to the cluster center c1 is coal, and the image pixel corresponding to the cluster center c2 is rock. The steps of denoising the original drill cuttings image in step 1 are as follows: Step 1.1: Design guidance image for improved adaptive region-weighted guided filtering Where G is the input original cuttings image; ω is the adaptive weighting factor; is the image processed by Gaussian filter; Define the image noise factor ξ: In the formula, is the grayscale variance of the original cuttings image G; is the image processed by Gaussian filter Grayscale variance of ; is a normal number, The value of is defined as (1 / L 2 ), L is the gray level of the original cuttings image G; On this basis, the adaptive weighting factor ω is defined as follows: Where: τ is the control parameter; In the area without noise or with low noise intensity, the image noise degree factor ξ tends to 1 and the adaptive weighting factor ω tends to 0; in the area with high noise intensity, the adaptive weighting factor ω tends to 1; Step 1.2: In the improved adaptive region weighted guided filtering process, the cost function E is defined as follows: Where p is the pixel in the image; p′ is a square window with radius r in the image the center of is the edge perception factor; λ is the penalty coefficient; By minimizing the cost function E shown in formula (3), we can obtain and is the gain factor, is the bias factor, as shown in formula (4), In formula (4), It is the product of the pixel grayscale values ​​of the two image matrices. for The mean grayscale value of pixels in a window with a radius of r; for The mean grayscale value of pixels in a window with a radius of r; μ G,r (p′) is the grayscale mean of the pixels in the window of radius r; is the grayscale variance of the pixels in the window with radius r; Step 1.3: In the defined size Output the denoised drill cuttings image in the image window As shown in formula (5), Where: In size In the image window The average value of In size In the image window The average value of From formula (6), we can get: Where: For size The number of pixels contained in the window.

2. The method for identifying coal and rock during the pressure relief process of the anti-bumping drilling robot according to claim 1 is characterized in that: In step 2, the denoised drill cuttings image is clustered using the local information weighted intuitionistic fuzzy C-means clustering algorithm. The clustering segmentation steps are as follows: Step 2.1: Introducing uncertainty into the local information weighted intuitionistic fuzzy C-means clustering algorithm Adaptive local information weight κ ij , local information H ij The core of the local information weighted intuitionistic fuzzy C-means clustering algorithm is to solve the membership of known data individuals to the cluster set center by minimizing the objective function J(U,V), as shown in formula (7). In the formula, sample x j Belongs to the sample data set X={x1,x2,…,x N }; c is the number of cluster sets selected according to the requirements; N is the number of individuals in the sample data set; m is the fuzzy constant; For sample x j For the composite membership of the i-th cluster set; is the cluster center and sample x of each cluster set j The distance between is the cluster center of the ith cluster set, and the calculation update formula is shown in formula (8): In the local information weighted intuitionistic fuzzy C-means clustering algorithm, the composite membership relationship is defined as follows: In the formula, μ i,j is the membership degree of fuzzy C-means clustering considering local information; is the membership degree of intuitionistic fuzzy C-means clustering, and the calculation method is as follows: In formula (7) and formula (10), the local information H ij The definition is as follows: In the formula, 1≤i≤c, 1≤j≤N, N j is the number of neighborhood pixels in a 3×3 window centered on pixel j; k is the kth pixel in a 3×3 window centered at pixel j; d j,k is the spatial Euclidean distance between pixel k and pixel j; μ i,k is the sample x k For the membership degree of the i-th cluster set; In formula (7), the local information weighting parameter κ ij The calculation is as follows: In the formula, is the variance of the sample data in the jth 3×3 window; For all The mean of η is the denoised cuttings image Gray levels; In formula (7), the uncertainty The calculation formula is as follows: In the formula, α is defined as the uncertainty parameter, and its value is generally between 0.8 and 0.9; Step 2.2: Calculate the objective function according to formula (7), set the objective function accuracy ε of the algorithm iteration, and use formula (7) to calculate the objective function values ​​J(t) and J(t+1) with iteration times t and t+1 respectively. If |J(t)-J(t+1)|<ε, the iteration ends and the cluster centers c1 and c2 are output. Otherwise, jump to step 2.1 and update the membership by formula (8) and formula (9) and cluster centers Until the iteration ends; Step 2.3: According to the obtained membership matrix, take the class corresponding to the maximum sample membership as the result of sample clustering, and divide the pixels in the input image into two categories: coal or rock. The corresponding two cluster set centers are c1 or c2 respectively, and the clustering ends at this time.

Images