Deep learning-based fracture recognition and fracture dip calculation method for core images

Abstract

The present invention discloses a method for identifying cracks and calculating inclination angles in core images based on deep learning. The collected core images are subjected to distortion correction and core cross-row segmentation to obtain core cross-row images. For the core cross-row images, two methods, direct and indirect, are proposed for extracting crack skeleton lines to complete the identification of core cracks. For the extracted core crack skeleton lines, a three-dimensional coordinate system is constructed for the core, the three-dimensional coordinates of the core crack skeleton lines are extracted, the crack plane is fitted, and the crack inclination angle is calculated. The two different methods for crack identification, namely the direct method and the indirect method, are used to improve the accuracy of crack identification and inclination angle calculation. It is proposed to use the direct method and the indirect method for cores without and with fracture zones, respectively.

Description

Technical Field

[0001] The present invention relates to the fields of underground engineering, rock engineering technology and image processing, and in particular to a method for crack recognition and dip angle calculation in exploration drilling core photographs. Background Art

[0002] Rock mass integrity is a key factor affecting rock mass quality, and core fractures provide crucial fundamental information about the integrity of the subsurface rock mass. In the early stages of engineering geological exploration, drilling is often used to obtain a large number of cores at designated locations throughout the study area. Cores provide the most intuitive means of obtaining subsurface geological information. For large non-coal mines, exploration lines are spaced 50 to 200 meters apart, typically with hundreds of geological survey boreholes deployed, totaling tens of thousands of meters. Cores are placed in core boxes every 6 to 9 meters, and thousands of these boxes are stored. By marking fracture locations in the cores, fracture size and joint density can be calculated, thereby assessing the integrity of the subsurface rock mass. However, manually cataloging fractures across thousands of meters of drill core is a laborious and tedious process, and mine personnel have limited time to process cores. Therefore, automated algorithms for identifying fractures in cores are underdeveloped and need to be implemented.

[0003] Currently, there is considerable research on the identification of core fractures. Olson et al. collected point cloud data by laser scanning core boxes and detected the location of core fractures based on changes in the values ​​perpendicular to the core box in the point cloud model. Ozturk and Saricam used three directional light sources during image acquisition to highlight fracture shadows and then used edge detection algorithms to extract fractures from core box images. Fatimah rotated the core 360° to collect unfolded images and used MaskRCNN to identify fractures in the unfolded core images. The core fracture algorithms proposed in the above studies all achieved good results, but the data used relied on expensive equipment or complex operating methods, hindering their use in the field. Furthermore, the algorithms ignored the interference of fracture zones in the core on fracture identification, resulting in reduced fracture identification accuracy. Summary of the Invention

[0004] In view of this, the present invention discloses a method for core image crack recognition and crack inclination calculation based on deep learning. First, the core image cracks are identified, specifically by obtaining the core crack skeleton line, and the core crack skeleton line is used to represent the core crack. Then, the core crack inclination is calculated based on the core crack skeleton line.

[0005] The specific steps include:

[0006] S1. Batch acquisition of core box images;

[0007] S2. Performing semantic segmentation-based core box image distortion correction on the core box image acquired in step S1 to obtain a standard core box image;

[0008] S3. Segment the core row of the standard core box image obtained in step S2 to obtain a core row image;

[0009] S4. Identify fractures in the core cross-row images using direct or indirect methods;

[0010] The direct method is to identify the fracture area in the core horizontal row image, and then perform skeleton processing on the identified fracture area to obtain a core fracture skeleton line image containing fracture skeleton lines;

[0011] The indirect method is to identify the core blocks and the core fracture zones in the core cross-row image, mark the core blocks and the core fracture zones in the core cross-row image, and then use a digital image processing method to obtain a core fracture skeleton line image containing the fracture skeleton line;

[0012] The present invention uses fracture skeleton lines to represent fractures in the core image. Fracture identification in the core image is actually to extract fracture skeleton lines representing the fractures in the core.

[0013] S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

[0014] As a supplement to the above technical solution, step S2 specifically includes:

[0015] S2.1 inputs the core box image into a core box semantic segmentation network and outputs a core box mask image, wherein the core box pixel values ​​and the background pixel values ​​in the core box mask image are different values, so as to distinguish the core box area from the background area;

[0016] S2.2 Core box distortion correction processing: Using the transmission transformation method, the core box mask image is multiplied by the transformation matrix to complete the image distortion correction and obtain the standard core box image.

[0017] As a supplement to the above technical solution, the specific steps of step S3 are: determining the coordinate information (Xmin, Ymin, Xmax, Ymax) of each core horizontal rectangular box in the core box, where (Xmin, Ymin) is the coordinate of the upper left corner of the core horizontal rectangular box, and (Xmax, Ymax) is the coordinate of the lower right corner of the lithology horizontal rectangular box; according to the coordinate information of each core horizontal rectangular box in the core box, the standard core box image is separated into core box horizontal images.

[0018] As a supplement to the above technical solution, step S4 adopts a direct core fracture identification method to identify core fractures, and the specific steps are as follows:

[0019] The core cross-row image is input into the trained core fracture semantic segmentation network model, the fracture area pixels in the core cross-row image are marked, and a core fracture mask image is output. The core fracture mask image is then skeletonized to output a core fracture skeleton line image. The skeletonization process is to reduce the fracture area mark from multiple pixels to unit pixels to obtain a fracture skeleton line.

[0020] The core fracture area refers to the gap between two adjacent core blocks formed by the fracture of the core, or the gap between adjacent core blocks and the core fracture zone formed by the fracture of the core.

[0021] As a supplement to the above technical solution, step S4 adopts an indirect identification method for core cracks to identify core cracks, and the specific steps are as follows:

[0022] The core cross-row image is input into the trained core block-fracture zone semantic segmentation network to identify the core blocks and core fracture zones in the core cross-row image, mark the pixels where each core block and core fracture zone are located, and output the core block and core fracture zone mask images, as well as the rectangular frame information around each core block;

[0023] Distinguish the contact type of the core block or the broken zone in the core block and core broken zone mask image. The contact type of the core block or broken zone includes "core block-core block" and "core block-broken zone"

[0024] For the "core block-core block" contact type, a digital image processing method is used to generate a rectangular frame around the core block based on the rectangular frame information, mark the edge line of the core block in the core block and core fracture zone mask image, define the intersection area of ​​the rectangular frames of two adjacent core blocks as the boundary area, and define the midline of the edge lines of two adjacent core blocks in the boundary area as the fracture skeleton line, and output the core fracture skeleton line image;

[0025] For the "core block-fracture zone" contact type, digital image processing methods are used to mark the edge line of the core block in the mask image of the core block and the core fracture zone. The edge line of the side of the core block close to the fracture zone is used as the fracture skeleton line, and the core fracture skeleton line image is output;

[0026] The output of the rectangular frame information around each core block is specifically as follows: the vertical coordinate of the lower end point of the core block to be marked is defined as Ymax, the vertical coordinate of the upper end point is defined as Ymin, the horizontal coordinate of the left end point is defined as Xmin, the horizontal coordinate of the right end point is defined as Xmax, the coordinates of the upper left corner of the rectangular frame are defined as (Xmin, Ymin), and the coordinates of the lower right corner of the rectangular frame are defined as (Xmax, Ymax).

[0027] As a supplement to the above technical solution, step S5 specifically includes the following steps:

[0028] S5.1 Establish a three-dimensional coordinate system for the core. First, rotate the core image 90° clockwise to adjust the core to a vertical angle. With the lower left corner of the core image as the origin, the X-axis is perpendicular to the core axis in the image and to the right, the Z-axis is parallel to the core axis in the image, and the Y-axis is perpendicular to the core axis and outward.

[0029] S5.2 extract the three-dimensional coordinates (x, y, z) of the core fracture;

[0030] S5.3 Fitting the fracture plane according to the three-dimensional coordinates of the core fracture, fitting the three-dimensional coordinates (x, y, z) of the core fracture to the three-dimensional plane equation ax+by+cz+d=0;

[0031] S5.4 calculates the angle between the fitted three-dimensional plane and a horizontal plane, where the horizontal plane is a plane formed by the X-axis and the Y-axis in the three-dimensional coordinate system of the core;

[0032] β=arccos[(a,b,c),(0,0,1)]

[0033] Where β represents the angle between the fitted three-dimensional plane and the horizontal plane.

[0034] As a supplement to the above technical solution, step S5.2 of extracting the three-dimensional coordinates (x, y, z) of the core fracture specifically includes the following steps:

[0035] S5.2.1 Extraction of (x, z) coordinates of the core fracture: First, rotate the fracture skeleton line image 90° clockwise, then flip it upside down, and use the lower left corner of the fracture skeleton line image as the origin of the coordinate system to extract the fracture skeleton line coordinates (x, z) in the fracture skeleton line image. The coordinates of the fracture skeleton line in the fracture skeleton line image are the (x, z) coordinates of the three-dimensional coordinates of the core fracture;

[0036] S5.2.2 Extraction of the core fracture (y) coordinates. The y coordinates of the left and right edges of the core in the fracture skeleton line image rotated 90° clockwise in step S5.2.1 are defined as 0. Based on the X coordinate of the core fracture, the Y coordinate of the fracture pixel point is calculated using the following formula:

[0037]

[0038]

[0039] In the formula, x and y are the coordinates of the pixel point where any crack skeleton line is located. min is the minimum x-coordinate value of the pixel point where the crack skeleton line is located, x max is the maximum x-coordinate of the pixel point where the fracture skeleton line is located, and D is the diameter of the core.

[0040] As a supplement to the above technical solution, when the core box image collected in step S1 does not contain the core fracture zone, the following steps are performed:

[0041] S2. Performing semantic segmentation-based core box image distortion correction on the core box image acquired in step S1 to obtain a standard core box image;

[0042] S3. Segment the core row of the standard core box image obtained in step S2 to obtain a core row image;

[0043] S4. using a direct method to identify cracks in the core cross-row image, and then skeletonizing the identified cracks to obtain a core crack skeleton line image containing crack skeleton lines;

[0044] S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

[0045] As a supplement to the above technical solution, when the core box image collected in step S1 contains a core fracture zone, the following steps are performed:

[0046] S2. Performing semantic segmentation-based core box image distortion correction on the core box image acquired in step S1 to obtain a standard core box image;

[0047] S3. Segment the core row of the standard core box image obtained in step S2 to obtain a core row image;

[0048] S4. using an indirect method to identify core blocks and core fracture zones in the core cross-row image, and then using a digital image processing method to obtain a core fracture skeleton line image including fracture skeleton lines;

[0049] S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

[0050] Beneficial Effects: This invention proposes a method for identifying and calculating the dip angle of core fractures in core box images captured by conventional cameras. This method overcomes the shortcomings of currently published algorithms, which rely on expensive equipment for data acquisition or are significantly affected by the core fracture zone. Fractures in core images are identified using direct or indirect methods. Experimental verification demonstrates that the direct method works well for core box images without fracture zones, while the indirect method works well for core box images with fracture zones. This improves the accuracy of fracture identification in core images and the accuracy of fracture dip angle calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the process of crack identification and crack dip calculation in core images.

[0052] Figure 2 Schematic diagram of the original image of the core box.

[0053] Figure 3 Schematic diagram of the core box image distortion correction process.

[0054] Figure 4 Schematic diagram of the horizontal segmentation process of the core box image.

[0055] Figure 5 Schematic diagram of the direct core fracture identification process.

[0056] Figure 6 Schematic diagram of the indirect core fracture identification process.

[0057] Figure 7 Schematic diagram of the calculation process of core fracture inclination.

[0058] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0059] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection, or communication; direct connection or indirect connection through an intermediate medium; internal communication between two elements or interaction between two elements, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0060] A method for crack identification and crack dip calculation in core images based on deep learning can identify cracks in core images and calculate the dip angle of core cracks. This patent is a method for calculating crack dip angle in core images based on deep learning, which specifically includes the following steps:

[0061] S1. Figure 2 As shown, there are no restrictions on core lithology, core color, degree of core crushing, core box material, core box color, number of core box rows, or whether the core boxes are placed horizontally. Core box images are collected in batches. The core box images are images of the core boxes after drilling and placing the cores in the core boxes. No less than 2,000 core box images should be collected.

[0062] Figure 2 (a) is the core box image with different number of rows. Figure 2 (b) is the core image of different lithologies. Figure 2 (c) Core box images with different core crushing degrees, Figure 2 (d) Image of the core box covered with dust and soil, Figure 2 (e) Image of a core box that is not filled with core, Figure 2 (f) The core box image is obtained after the core box is rotated.

[0063] S2. Performing semantic segmentation-based core box image distortion correction processing on the core box image of step S1 to obtain a standard core box image, specifically comprising the following steps:

[0064] S2.1 Input the core box image into the trained core box semantic segmentation network. The network segments the core box image and outputs the core box mask image. The core box mask image is as follows: Figure 3 As shown in (a), Figure 3 As shown in (b), in the core box mask image, the pixels where the core box is located are marked as 1, and the pixels in the background are marked as 0 to distinguish the core box area from the background area. The core box semantic segmentation network is the core box semantic segmentation network disclosed in application number CN202210682472.4.

[0065] The core box semantic segmentation network training is as follows: manually marking the core box boundaries, creating a data set, dividing the data set into a training set and a test set for training and validating the network, and stopping the training after the loss value approaches 0 and remains stable, and determining the network parameters;

[0066] S2.2 Core box distortion correction processing:

[0067] Distortion correction is used to transform the original core box image collected on-site into a rectangular horizontal standard core box image. Distortion correction is a three-dimensional spatial transformation that uses a projection transformation method to project the image onto a new viewing plane to achieve image distortion correction. The details are as follows:

[0068]

[0069] In the above formula, (u, v, 1) is the coordinate point of the core box mask image plane, and (x, y, z) is the coordinate point corresponding to the core box mask image plane after distortion correction. The above formula is expanded in the form of matrix multiplication to obtain the following formula:

[0070]

[0071] Since the core box mask image is a two-dimensional image, let The transformed three-dimensional image coordinates are reduced to two dimensions, and the formula is as follows:

[0072]

[0073] Where: (u, v) is the coordinate of the pixel point in the core box mask image, (x', y') is the coordinate of the pixel point in the standard core box image obtained after the distortion correction of the pixel point in the core box image, and k 11 、k 12 、k 13 、k 21 、k 22 、k 23 、k 31 、k 32 are all undetermined constants.

[0074] Substitute the pixel coordinates in the core box mask image into formula (3) to complete the distortion correction of the core box image and obtain the standard core box image. The standard core box image is as follows: Figure 3 (d) shown.

[0075] The k 11 、k 12 、k 13 、k 21 、k 22 、k 23 、k 31 、k32 The calculation method includes the following steps:

[0076] like Figure 3 As shown in (c), in the core box mask image, the horizontal and vertical lines passing through the center point of the core box divide the core box mask image into four partitions. The coordinates of the four corner points of the core box farthest from the center of the core box in each partition are defined as PointA (X1, Y1), PointB (X2, Y2), PointC (X3, Y3), and PointD (X4, Y4), respectively. The length L and width W of the core box are calculated based on the corner points of the core box.

[0077]

[0078]

[0079] The coordinates of the four corner points of the core box in the standard core box image after distortion correction are defined as Pointa(0,0), Pointb(0,L), Pointc(L,W), and Pointd(0,W), which form four sets of corresponding coordinate points with the coordinates of the four corner points of the core box in the core box mask image.

[0080] Substitute the coordinates of the four corner points of the core box in the core box mask image and the coordinates of the four corner points of the core box in the standard core box image into the transformation matrix to obtain k 11 、k 12 、k 13 、k 21 、k 22 、k 23 、k 31 、k 32

[0081] S3. Perform core row segmentation on the standard core box image, and segment the standard core box image into core row images, such as Figure 4 As shown, Figure 4 (a) is the standard core box image, Figure 4 (b) is a horizontal rectangular frame marking the core. Figure 4 (c) is the cropped cross-row image of the core.

[0082] The specific steps are as follows: input the standard core box image into the trained core row target detection network, output the coordinate information (Xmin, Ymin, Xmax, Ymax) of each core row rectangular box in the core box, where (Xmin, Ymin) is the coordinate of the upper left corner of the core row rectangular box, and (Xmax, Ymax) is the coordinate of the lower right corner of the core row rectangular box. According to the coordinate information of each core row rectangular box in the core box, the standard core box image is divided into core box row images;

[0083] The training of the core row target detection network is as follows: manually marking the boundaries of the core row rectangular boxes to create a core row dataset, dividing the dataset into a training set and a test set for training and verifying the network. When the loss value approaches 0 and remains stable, the training is stopped and the network parameters are determined.

[0084] The core horizontal rectangular frame is the boundary of the slot body for placing the cores in the core box.

[0085] S4. Processing the core cross-row image using a core fracture direct identification method or a core fracture indirect identification method to obtain a core fracture skeleton line image;

[0086] The specific steps of the direct identification method are: Figure 5 As shown, Figure 5 (a) Cross-row image of the core, Figure 5 (b) is the core fracture mask image, Figure 5 (c) is a core fracture skeleton line image. The core cross-row image is input into a trained core fracture semantic segmentation network model. The crack region pixels in the core cross-row image are labeled, and a core fracture mask image is output. The core fracture mask image is then skeletonized to output a core fracture skeleton line image. The skeletonization process in the direct method specifically reduces the crack region labels from multiple pixels to a unit pixel to obtain a fracture skeleton line, where the unit pixel is a single pixel.

[0087] The core fracture semantic segmentation network model is trained as follows: for the core horizontal image, the boundaries of the core fracture are manually marked to create a core fracture dataset, and the core fracture dataset is divided into a training dataset and a validation dataset for training and validating the core fracture semantic segmentation network model. When the loss value approaches 0 and remains stable, the training is stopped and the network parameters are determined;

[0088] The core crack refers to the gap between two adjacent core blocks formed by core fracture, or the gap between adjacent core blocks and the core fracture zone formed by core fracture. The core fracture zone refers to the part of the core that is broken into small pieces when drilling and coring encounter a strong fracture zone or a strong weathering zone underground.

[0089] The specific steps of the indirect identification method are as follows: Figure 6 As shown, Figure 6 (a) is the standard core box image, Figure 6 (b) Mask image of core block and core fracture zone. Figure 6(c) Core fracture skeleton line image. The core cross-row image is input into the trained core block-fracture zone semantic segmentation network. The core blocks and core fracture zones in the core cross-row image are identified, the pixels where the core blocks and core fracture zones are located are marked, and mask images of the core blocks and core fracture zones are output. At the same time, the rectangular box information around each core block is output.

[0090] Then, the contact types of the core blocks or the broken zones in the core block and the core broken zone mask images are distinguished. The contact types of the core blocks or the broken zones include "core block-core block" and "core block-broken zone".

[0091] The "core block-core block" contact type is specifically that the fracture of the core is located between two complete core blocks, and there are complete core blocks on both sides of the crack; the "core block-fracture zone" contact type is specifically that the fracture of the core is located between the core block and the fracture zone, and there is a core block on one side of the crack and a fracture zone on the other side.

[0092] For the "core block-core block" contact type, a digital image processing method is used to generate a rectangular frame around the core block based on the rectangular frame information, mark the edge line of the core block in the core block and core fracture zone mask image, define the intersection area of ​​the rectangular frames of two adjacent core blocks as the boundary area, and define the midline of the edge lines of two adjacent core blocks in the boundary area as the fracture skeleton line, and output the core fracture skeleton line image;

[0093] For the "core block-fracture zone" contact type, digital image processing methods are used to mark the edge line of the core block in the mask image of the core block and the core fracture zone. The edge line of the side of the core block close to the fracture zone is used as the fracture skeleton line, and the core fracture skeleton line image is output;

[0094] The digital image processing method may use edge detection algorithms, morphological algorithms, etc. to perform recognition processing on the image.

[0095] The output of the rectangular frame information around each core block is specifically as follows: the vertical coordinate of the lower end point of the core block to be marked is defined as Ymax, the vertical coordinate of the upper end point is defined as Ymin, the horizontal coordinate of the left end point is defined as Xmin, the horizontal coordinate of the right end point is defined as Xmax, the coordinates of the upper left corner of the rectangular frame are defined as (Xmin, Ymin), and the coordinates of the lower right corner of the rectangular frame are defined as (Xmax, Ymax).

[0096] The core block-fracture zone semantic segmentation network is trained by manually marking the boundaries of the core blocks and the core fracture zones in the core horizontal row images to create a core block-fracture zone dataset. The core block-fracture zone dataset is divided into a training dataset and a validation dataset for training the core block-fracture zone deep learning network model. When the loss value approaches 0 and remains stable, the training is stopped and the network parameters are determined.

[0097] S5. Fit the fracture three-dimensional plane according to the core fracture skeleton line image obtained in step S4, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle, such as Figure 7 As shown, the specific steps include:

[0098] S5.1 Establish a three-dimensional coordinate system for the core. First, rotate the core horizontal image 90° clockwise to adjust the core to a vertical angle. With the lower left corner of the core horizontal image as the origin, the X-axis is perpendicular to the rightward direction of the core axis in the core horizontal image, the Z-axis is parallel to the core axis in the core horizontal image, and the Y-axis is perpendicular to the outward direction of the core horizontal image. The outward direction perpendicular to the core horizontal image is perpendicular to the plane formed by the X-axis and Z-axis. Construct the three-dimensional coordinate system for the core.

[0099] S5.2 extracts the three-dimensional coordinates (x, y, z) of the core fracture. The specific steps include:

[0100] S5.2.1 Extract the (x, z) coordinates of the core fracture. First, rotate the fracture skeleton line image 90° clockwise, then flip it upside down. With the lower left corner of the fracture skeleton line image as the origin of the coordinate system, extract the fracture skeleton line coordinates (x, z) in the fracture skeleton line image. The coordinates of the fracture skeleton line in the fracture skeleton line image are the (x, z) coordinates of the core fracture in the three-dimensional coordinates.

[0101] For the fracture skeleton line images extracted using the direct and indirect methods, points with a median value of 1 are displayed in white, representing the fracture skeleton line. Pixel positions are expressed in a pixel coordinate system. Since the pixel coordinate system originates in the upper left corner of the image and is measured in pixels, with positive abscissas pointing rightward and positive ordinates pointing downward, the fracture skeleton line image is flipped upside down to align the pixel coordinate system with the three-dimensional core coordinate system.

[0102] S5.2.2 Extract the (y) coordinates of the core fractures. Determine the Y coordinate of the fracture based on the core x-coordinate and the core diameter D. In step S5.2.1, the y coordinates of the left and right edges of the core in the fracture skeleton image rotated 90° clockwise are defined as 0. The Y coordinate of the fracture pixel in the middle of the core reaches its maximum, which is L / 2. Calculate the Y coordinate of the fracture pixel based on the X coordinate of the core fracture. The specific calculation formula is as follows:

[0103]

[0104]

[0105] In the formula, x and y are the coordinates of the pixel point where any crack skeleton line is located. minis the minimum x-coordinate value of the pixel point where the crack skeleton line is located, x max is the maximum value of the x-coordinate of the pixel point where the fracture skeleton line is located, and D represents the diameter of the core;

[0106] S5.3 Fit the crack plane, fitting the three-dimensional coordinates (x, y, z) to the three-dimensional plane equation ax+by+cz+d=0;

[0107] S5.4 calculates the angle between the fitted three-dimensional plane and a horizontal plane, where the horizontal plane is a plane formed by the X-axis and the Y-axis in the three-dimensional coordinate system of the core;

[0108] β=arccos[(a,b,c),(0,0,1)] (8)

[0109] Where β represents the angle between the fitted three-dimensional plane and the horizontal plane.

[0110] The present invention also evaluates the effects of the direct and indirect crack identification methods in the above embodiments, including the following steps:

[0111] Preparation of S1 validation set. Separate core cross-section images, not repeated in the previous example, were prepared. Two types of core cross-section images were selected from the core cross-section images: (1) core cross-section images without fracture zones; and (2) core cross-section images with fracture zones. Fracture boundaries were manually marked on the core cross-section images without fracture zones, defining this as validation set 1. Fracture boundaries were manually marked on the core cross-section images with fracture zones, defining this as validation set 2.

[0112] Comparison of S2 Accuracy and Recall. Accuracy is the ratio of correctly classified samples to the total number of samples. Recall is the ratio of correctly predicted samples to the total number of actual samples in that category. Direct and indirect methods were used to identify fractures in cross-row images of cores from two validation sets, and the accuracy and recall were calculated (Table 1). The results show that for validation set 1, the direct method has higher accuracy and recall than the indirect method. For validation set 2, the direct method has lower accuracy and recall than the indirect method.

[0113] Table 1

[0114]

[0115] S3 Absolute error in fracture dip calculation. For validation sets 1 and 2, we manually annotated fracture traces and defined the calculated fracture dip as the accurate dip. We then defined the fracture dips calculated from the direct and indirect methods as the calculated dips. The calculated dips were compared with the accurate dip values ​​to calculate the absolute error (Table 2).

[0116] Table 2

[0117] method Absolute error of validation set 1 Absolute error of validation set 2 Direct method 5.2 4.5 Indirect method 4.2 4.7

[0118] Algorithm selection and recommendation are performed based on the above evaluation steps.

[0119] Since the absolute errors of fracture calculation between the two methods are similar, the fracture identification method is selected based on accuracy and recall. When the core cross-row image contains a core fracture zone, the indirect method is used to identify core fractures. When the core image does not contain a core fracture zone, the direct method is used to identify fractures.

[0120] To this end, in the core image recognition process, step S1 is first performed. The core box images are batch collected, and then the core box images are judged by the naked eye to determine whether the core crushing zone is included. If the core box images contain the core crushing zone, the following steps are performed:

[0121] S2. Correcting the distortion of the core box image to obtain a standard core box image;

[0122] S3. Perform core cross-row segmentation on the standard core box image obtained in step S2 to obtain multiple core cross-row images. The number of core cross-row images is the same as the number of slots in the core box for placing cores. The core cross-row images obtained after segmentation may contain a core fracture zone or may not contain a fracture zone. Regardless of whether the core cross-row images contain a core fracture zone, step S4 is performed.

[0123] S4. Use the indirect method to identify the core blocks and core fracture zones in the core cross-row image, and then use the digital image processing method to obtain the core fracture skeleton line image containing the fracture skeleton line.

[0124] S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

[0125] When the core box image is judged by the naked eye to contain no core fracture zone in step S1, the following steps are performed:

[0126] S2. Correcting the distortion of the core box image to obtain a standard core box image;

[0127] S3. Perform core row segmentation on the standard core box image obtained in step S2 to obtain multiple core row images.

[0128] S4. Use a direct method to identify cracks in the core horizontal row image, and then perform skeleton processing on the identified cracks to obtain a core crack skeleton line image containing crack skeleton lines.

[0129] S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

[0130] The above description is only a preferred specific implementation method of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for crack identification and crack dip calculation in core images based on deep learning, characterized in that: The following steps are involved: S1. Batch acquisition of core box images; S2. Performing semantic segmentation-based core box image distortion correction on the core box image acquired in step S1 to obtain a standard core box image; S3. Segment the core row of the standard core box image obtained in step S2 to obtain a core row image; S4. Identify fractures in the core cross-row images using direct or indirect methods; The direct method is to identify the fracture area in the core horizontal row image, and then perform skeleton processing on the identified fracture area to obtain a core fracture skeleton line image containing fracture skeleton lines; The indirect method is to identify the core blocks and the core fracture zones in the core cross-row image, mark the core blocks and the core fracture zones in the core cross-row image, and then use a digital image processing method to obtain a core fracture skeleton line image containing the fracture skeleton line; S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

2. The method for crack identification and crack dip calculation based on deep learning in core images according to claim 1, characterized in that: The step S2 specifically includes: S2.1 inputs the core box image into a core box semantic segmentation network and outputs a core box mask image, wherein the core box pixel values ​​and the background pixel values ​​in the core box mask image are different values, so as to distinguish the core box area from the background area; S2.2 Core box distortion correction processing: Using the transmission transformation method, the core box mask image is multiplied by the transformation matrix to complete the image distortion correction and obtain the standard core box image.

3. The method for crack identification and crack dip calculation based on deep learning in core images according to claim 1, characterized in that: The specific steps of step S3 are: determining the coordinate information (Xmin, Ymin, Xmax, Ymax) of each core horizontal row rectangular box in the core box, where (Xmin, Ymin) is the coordinate of the upper left corner of the core horizontal row rectangular box, and (Xmax, Ymax) is the coordinate of the lower right corner of the lithology horizontal row rectangular box; and dividing the standard core box image into core box horizontal row images according to the coordinate information of each core horizontal row rectangular box in the core box.

4. The method for core image crack identification and crack dip calculation based on deep learning according to claim 1, characterized in that: The step S4 uses a direct core crack identification method to identify core cracks, and the specific steps are as follows: The core cross-row image is input into the trained core fracture semantic segmentation network model, the fracture area pixels in the core cross-row image are marked, and a core fracture mask image is output. The core fracture mask image is then skeletonized to output a core fracture skeleton line image. The skeletonization process is to reduce the fracture area mark from multiple pixels to unit pixels to obtain a fracture skeleton line. The core fracture area refers to the gap between two adjacent core blocks formed by the fracture of the core, or the gap between adjacent core blocks and the core fracture zone formed by the fracture of the core.

5. The method for core image crack identification and crack dip calculation based on deep learning according to claim 1, characterized in that: The step S4 uses an indirect identification method for core cracks to identify core cracks, and the specific steps are as follows: The core cross-row image is input into the trained core block-fracture zone semantic segmentation network to identify the core blocks and core fracture zones in the core cross-row image, mark the pixels where each core block and core fracture zone are located, and output the core block and core fracture zone mask images, as well as the rectangular frame information around each core block; Distinguish the contact type of the core block or the broken zone in the core block and core broken zone mask image. The contact types of the core block or broken zone include "core block-core block" and "core block-broken zone". For the "core block-core block" contact type, a digital image processing method is used to generate a rectangular frame around the core block based on the rectangular frame information. The edge lines of the core block are marked in the core block and core fracture zone mask image. The intersection area of ​​the rectangular frames of two adjacent core blocks is defined as the boundary area. The midline of the edge lines of two adjacent core blocks within the boundary area is defined as the fracture skeleton line. The core fracture skeleton line image is then output. For the "core block-fracture zone" contact type, digital image processing methods are used to mark the edge lines of the core block in the mask images of the core block and the core fracture zone. The edge line of the side of the core block close to the fracture zone is used as the fracture skeleton line, and the core fracture skeleton line image is output; The output of the rectangular frame information around each core block is specifically as follows: the vertical coordinate of the lower end point of the core block to be marked is defined as Ymax, the vertical coordinate of the upper end point is defined as Ymin, the horizontal coordinate of the left end point is defined as Xmin, the horizontal coordinate of the right end point is defined as Xmax, the coordinates of the upper left corner of the rectangular frame are defined as (Xmin, Ymin), and the coordinates of the lower right corner of the rectangular frame are defined as (Xmax, Ymax).

6. The method for core image crack identification and crack dip calculation based on deep learning according to claim 1, characterized in that: The step S5 specifically includes the following steps: S5.1 Establish a three-dimensional coordinate system for the core. First, rotate the core transverse image 90° clockwise and adjust the core to a vertical angle. With the lower left corner of the core transverse image as the origin, the X-axis is perpendicular to the core axis in the transverse image and pointing rightward, the Z-axis is parallel to the core axis in the transverse image, and the Y-axis is perpendicular to the core axis and pointing outward. S5.2 extract the three-dimensional coordinates (x, y, z) of the core fracture; S5.3 Fitting the fracture plane according to the three-dimensional coordinates of the core fracture, fitting the three-dimensional coordinates (x, y, z) of the core fracture to the three-dimensional plane equation ax+by+cz+d=0; S5.4 calculates the angle between the fitted three-dimensional plane and a horizontal plane, where the horizontal plane is a plane formed by the X-axis and the Y-axis in the three-dimensional coordinate system of the core; β=arccos[(a,b,c),(0,0,1)] Where β represents the angle between the fitted three-dimensional plane and the horizontal plane.

7. The method for core image crack identification and crack dip calculation based on deep learning according to claim 5, characterized in that: The step S5.2 of extracting the three-dimensional coordinates (x, y, z) of the core fracture specifically includes the following steps: S5.2.1 Extract the (x, z) coordinates of the core fracture. First, rotate the fracture skeleton line image 90° clockwise, then flip it upside down. With the lower left corner of the fracture skeleton line image as the origin of the coordinate system, extract the fracture skeleton line coordinates (x, z) in the fracture skeleton line image. The coordinates of the fracture skeleton line in the fracture skeleton line image are the (x, z) coordinates of the core fracture in the three-dimensional coordinates. S5.2.2 Extraction of the core fracture (y) coordinates. The y coordinates of the left and right edges of the core in the fracture skeleton line image rotated 90° clockwise in step S5.2.1 are defined as 0. Based on the X coordinate of the core fracture, the Y coordinate of the fracture pixel point is calculated using the following formula: In the formula, x and y are the coordinates of the pixel point where any crack skeleton line is located. min is the minimum x-coordinate value of the pixel point where the crack skeleton line is located, x max is the maximum x-coordinate of the pixel point where the fracture skeleton line is located, and D is the diameter of the core.

8. The method for core image crack identification and crack dip calculation based on deep learning according to claim 1, characterized in that: When the core box image collected in step S1 does not contain a core fracture zone, the following steps are performed: S2. Performing semantic segmentation-based core box image distortion correction on the core box image acquired in step S1 to obtain a standard core box image; S3. Segment the core row of the standard core box image obtained in step S2 to obtain a core row image; S4. using a direct method to identify cracks in the core cross-row image, and then skeletonizing the identified cracks to obtain a core crack skeleton line image containing crack skeleton lines; S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

9. The method for core image crack identification and crack dip calculation based on deep learning according to claim 1, characterized in that: When the core box image collected in step S1 contains a core fracture zone, the following steps are performed: S2. Performing semantic segmentation-based core box image distortion correction on the core box image acquired in step S1 to obtain a standard core box image; S3. Segment the core row of the standard core box image obtained in step S2 to obtain a core row image; S4. using an indirect method to identify core blocks and core fracture zones in the core cross-row image, and then using a digital image processing method to obtain a core fracture skeleton line image including fracture skeleton lines; S5. Extract the three-dimensional coordinates of the fracture skeleton line in the core fracture skeleton line image obtained in step S4, fit the fracture three-dimensional plane, and calculate the angle between the fracture three-dimensional plane and the horizontal plane. The angle between the fracture three-dimensional plane and the horizontal plane is the fracture inclination angle.

Images