A rock mass quality index calculation method based on borehole images
By using a deep learning model to identify structural surfaces and fracture zones in borehole images and combining this with pixel resolution calculations, the problem of low efficiency in traditional RQD measurements is solved, enabling automated and accurate calculation of rock mass quality indicators.
Patent Information
- Application Number
- CN202610797364.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-04
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for core measurement (RQD) are inefficient, highly susceptible to human factors, and lack systematic and automated calculation schemes, resulting in insufficient accuracy and efficiency in obtaining rock mass quality indicators.
A deep learning-based structural surface recognition model is used to identify structural surfaces and fracture zones in borehole images. The pixel resolution is calculated by combining edge detection and contour analysis algorithms. The distance between adjacent structural surfaces and the length of fracture zones are calculated by subtraction method, thereby realizing the automated calculation of rock mass quality index (RQD).
It achieves efficient and accurate calculation of rock mass quality index (RQD), avoiding human interference and damage during core sampling in traditional methods, and significantly improving calculation efficiency and accuracy.
Smart Images

Figure CN122416154A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of engineering geological exploration technology, and more specifically to a method for calculating rock mass quality indicators based on borehole images. Background Technology
[0002] The Rock Quality Degree (RQD) is defined as the proportion of a core sample that is 10 cm (4 inches) or longer. It reflects the physical and mechanical properties of the rock mass, such as connectivity, permeability, and compressibility, and quantitatively reflects the integrity of the rock mass in engineering projects. As input to major rock mass classification systems, such as Barton's Q system and the Bieniawski rock mass classification system, RQD is widely used in rock engineering as a fundamental quality evaluation indicator for rock mass classification. Therefore, efficiently, accurately, and objectively obtaining RQD is of great significance for geological surveying, overall analysis, and basic design in engineering construction. Summary of the Invention
[0003] In view of the above problems, this disclosure provides a fast and efficient method for calculating rock mass quality index based on borehole images to estimate RQD.
[0004] This disclosure provides a method for calculating rock mass quality index based on borehole images, including: using a trained structural surface recognition model to identify structural surfaces and fracture zones in a target borehole image; the structural surface recognition model is configured to identify structural surfaces and fracture zones in the target borehole image based on the input target borehole image; the structural surface recognition model is trained using a training sample set of structural surfaces and fracture zones expanded by data augmentation and geometric transformation; based on the depth markers in the target borehole image, the pixel resolution is calculated using edge detection and contour analysis algorithms; based on the identification results of structural surfaces and fracture zones and the pixel resolution, the distance between adjacent structural surfaces and the length of fracture zones are calculated; based on the distance between adjacent structural surfaces and the length of fracture zones, the proportion of short rock segments with lengths less than a preset threshold and the proportion of fracture zones are statistically analyzed, and the rock mass quality index RQD is calculated using a subtraction method.
[0005] According to embodiments of this disclosure, a training method for a structural surface recognition model includes: cropping the original borehole image dataset based on edge parameters to remove interference areas such as image titles, image frames, and scales, resulting in processed borehole images; the edge parameters include at least upper edge parameters, lower edge parameters, left edge parameters, and right edge parameters; labeling the structural surfaces and fracture zones in the processed borehole images to construct an initial training sample set; expanding the initial training sample set using data augmentation methods to obtain a training sample set for structural surfaces and fracture zones; dividing the training sample set for structural surfaces and fracture zones into a training set, a validation set, and a test set; training the structural surface recognition model using the training set, adjusting the parameters of the structural surface recognition model using the validation set, evaluating the accuracy of the structural surface recognition model using the test set, and saving the structural surface recognition model whose accuracy reaches a preset standard as the trained structural surface recognition model.
[0006] According to embodiments of this disclosure, data augmentation methods are used to expand the initial training sample set to obtain a training sample set for structural surfaces and fracture zones. This includes: using the Mixup data augmentation method to generate synthetic borehole image samples containing mixed features by performing pixel-level linear combination of any two images and their labels in the initial training sample set; applying various geometric transformations to the synthetic borehole image samples to expand the number of samples, thereby constructing a training sample set for structural surfaces and fracture zones. The geometric transformations include at least one of flipping, rotation angle, and combinations of flipping and rotation angle.
[0007] According to embodiments of this disclosure, pixel resolution is calculated based on depth markers in a target borehole image using edge detection and contour analysis algorithms, including: performing grayscale conversion and Gaussian filtering preprocessing on the image region containing depth markers in the target borehole image; using edge detection and contour analysis algorithms to identify contours matching the depth markers and extracting the uppermost and lowermost pixel coordinates of the contours; and calculating the pixel resolution based on the actual physical distance corresponding to the depth markers, the uppermost pixel coordinates, and the lowermost pixel coordinates.
[0008] According to embodiments of this disclosure, the distance between adjacent structural surfaces and the length of the broken zone are calculated based on the identification results of the structural surfaces and the broken zone and the pixel resolution, including: calculating the actual physical distance between adjacent structural surfaces based on the depth difference of the points in the identified structural surface rectangle and the pixel resolution; and calculating the actual physical length of the broken zone based on the depth difference of the upper and lower edges of the identified broken zone rectangle and the pixel resolution.
[0009] According to embodiments of this disclosure, based on the distance between adjacent structural surfaces and the length of the fracture zone, the proportion of short rock segments with lengths less than a preset threshold and the proportion of fracture zones are statistically analyzed, and a deduction method is used to calculate the rock mass quality index (RQD). This includes: selecting short rock segments with lengths less than the preset threshold based on the distance between adjacent structural surfaces; statistically analyzing fracture zone segments based on their lengths; calculating the proportion of the sum of the lengths of the short rock segments to the total borehole depth to obtain the proportion of short rock segments; calculating the proportion of the sum of the lengths of the fracture zone segments to the total borehole depth to obtain the proportion of fracture zones; and subtracting the proportion of short rock segments and the proportion of fracture zones from 100% to obtain the rock mass quality index (RQD).
[0010] According to an embodiment of this disclosure, the method further includes: when calculating the proportion of the short rock segment and the proportion of the fractured zone, determining the overlap between the short rock segment region and the fractured zone region; if there is an overlap, the overlapping part is only included in the proportion of the fractured zone and not included in the proportion of the short rock segment.
[0011] The second aspect of this disclosure provides a rock mass quality index calculation system based on borehole images, which can be used to implement the aforementioned rock mass quality index calculation method based on borehole images. The system includes: a structural surface recognition module, used to identify structural surfaces and fracture zones in a target borehole image using a trained structural surface recognition model. The structural surface recognition model is configured to identify structural surfaces and fracture zones in the target borehole image based on the input target borehole image. The structural surface recognition model is trained using a training sample set of structural surfaces and fracture zones expanded by data augmentation and geometric transformation. A resolution calculation module, used to calculate pixel resolution based on depth markers in the target borehole image using edge detection and contour analysis algorithms. A distance-length calculation module, used to calculate the distance between adjacent structural surfaces and the length of fracture zones based on the identification results of structural surfaces and fracture zones and the pixel resolution. An index calculation module, used to calculate the proportion of short rock segments with lengths less than a preset threshold and the proportion of fracture zones based on the distance between adjacent structural surfaces and the length of fracture zones, and to calculate the rock mass quality index (RQD) using a subtraction method.
[0012] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the above-described method for calculating rock mass quality indicators based on borehole images.
[0013] A fourth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the above-described method for calculating rock mass quality indicators based on borehole images.
[0014] The rock mass quality index calculation method based on borehole images provided in this disclosure integrates a deep learning model for structural surface recognition, automatic pixel resolution calculation, and intelligent RQD calculation. By automating the process from image recognition to index calculation, it at least partially solves the technical problems of low efficiency, significant susceptibility to human factors and sampling variations in traditional core measurement methods, as well as the lack of a systematic automated RQD calculation scheme. This achieves the technical effect of highly efficient RQD calculation for rock mass quality indices. Attached Figure Description
[0015] Figure 1 A flowchart illustrating a method for calculating rock mass quality indicators based on borehole images according to an embodiment of the present disclosure is shown schematically.
[0016] Figure 2 A phased flowchart according to an embodiment of the present disclosure is illustrated schematically;
[0017] Figure 3 A schematic diagram illustrating borehole image cropping according to an embodiment of the present disclosure is shown.
[0018] Figure 4 A schematic diagram illustrating borehole image annotations according to an embodiment of the present disclosure is shown.
[0019] Figure 5 A schematic diagram illustrating sample expansion according to an embodiment of the present disclosure is shown.
[0020] Figure 6 A schematic diagram illustrating structural surface identification according to an embodiment of the present disclosure is shown.
[0021] Figure 7 This illustration schematically shows a diagram illustrating the calculation of the distance between adjacent structural surfaces and the length of the fracture zone according to an embodiment of the present disclosure;
[0022] Figure 8 A schematic diagram illustrating RQD calculation according to an embodiment of the present disclosure is shown.
[0023] Figure 9 A schematic diagram illustrating the comparison of RQD multi-sample calculation results according to embodiments of the present disclosure is shown.
[0024] Figure 10 A schematic diagram illustrating the comparison between calculated RQD values and longitudinal wave velocity test results according to an embodiment of the present disclosure is provided. Detailed Implementation
[0025] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.
[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.
[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.
[0028] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).
[0029] Currently, the common method for obtaining RQD (Recovery Quality Draw) is manual measurement, which involves manually measuring the length of core sample segments and then manually calculating the RQD. While manually measuring and calculating the RQD of a small number of core samples may be relatively simple for geologists, it is time-consuming and labor-intensive when collecting large amounts of core data and performing manual calculations. Furthermore, core samples may break or be lost during drilling and transportation, affecting the accuracy of the RQD results. Different personnel may also have different interpretations of non-structural fractures or mechanically damaged cracks, leading to calculation errors. Therefore, this cumbersome, labor-intensive, and highly human-dependent RQD measurement method urgently needs improvement.
[0030] With the widespread application of borehole camera technology in numerous engineering fields, the borehole wall image data acquired through digital borehole camera systems has overcome the shortcomings of non-in-situ core sampling. It comprehensively records geological information such as the structural characteristics, fracture distribution, and weak interlayers of the borehole rock mass, laying a data foundation for the objective calculation of rock mass quality indicators. However, existing technologies mostly focus on the intelligent extraction of spatial morphological parameters of fracture networks (such as dip, dip angle, and aperture), and systematic research on the automated calculation system for RQD indicators is still lacking.
[0031] Recently, data-driven algorithms, especially deep learning (DL), have been widely used in object detection, such as vehicle recognition and face recognition. Due to their powerful feature learning and representation capabilities, they hold promise for applications in geology, providing new ideas for the rapid and efficient estimation of RQD.
[0032] To address the aforementioned shortcomings and limitations, this disclosure proposes an effective framework for automatically calculating RQD based on deep learning (DL) and borehole wall images. Figure 1 A flowchart illustrating a method for calculating rock mass quality indicators based on borehole images according to an embodiment of the present disclosure is shown, such as... Figure 1 As shown, embodiments of this disclosure provide a method for calculating rock mass quality indicators based on borehole images, including: using a trained structural surface recognition model to identify structural surfaces and fracture zones in a target borehole image; the structural surface recognition model is configured to identify structural surfaces and fracture zones in the target borehole image based on the input target borehole image; the structural surface recognition model is trained using a training sample set of structural surfaces and fracture zones expanded by data augmentation and geometric transformation; calculating pixel resolution based on depth markers in the target borehole image using edge detection and contour analysis algorithms; calculating the distance between adjacent structural surfaces and the length of fracture zones based on the identification results of structural surfaces and fracture zones and the pixel resolution; and calculating the rock mass quality indicator RQD based on the distance between adjacent structural surfaces and the length of fracture zones, statistically analyzing the proportion of short rock segments with lengths less than a preset threshold and the proportion of fracture zones, and using a subtraction method.
[0033] Figure 2 A phased flowchart according to embodiments of the present disclosure is illustrated schematically, such as Figure 2As shown, this paper fully demonstrates the three stages of an intelligent calculation system for rock mass quality indicators based on borehole images. First, in the data preparation and model training stage, the system crops and annotates the acquired raw borehole images, and uses Mixup and geometric transformations for data augmentation to construct a high-quality dataset. This dataset is then divided into training, validation, and test sets, and used to train and test a YOLOv11 model. Only high-precision models with an accuracy exceeding 90% are retained for engineering applications. Next, in the actual engineering calculation stage, for new target borehole images, the system first automatically calculates the pixel resolution based on depth markers in the image, establishing a conversion relationship between pixels and physical dimensions. Then, the trained model is used to automatically identify structural surfaces in the image and calculate the actual distance between adjacent structural surfaces. Finally, in the RQD decision output stage, the system statistically analyzes the proportion of short rock segments with lengths less than a preset threshold and the proportion of fractured zones, and automatically calculates the final rock mass quality indicator (RQD) using a subtraction method.
[0034] Through the embodiments of this disclosure, downhole television camera technology is used to quickly record the integrity of the borehole wall and the distribution of cracks without the need to collect rock cores. It can also directly record the structural surface distribution information of the rock mass in its in-situ state. Combined with image analysis methods, RQD can be automatically calculated, which can avoid the destructive impact of the rock core sampling process, significantly save computing resources and workload, and improve calculation accuracy and efficiency.
[0035] Based on the above embodiments, the training method for the structural surface recognition model includes: cropping the original borehole image dataset according to edge parameters, removing interference areas such as image titles, image frames, and scales to obtain processed borehole images, where the edge parameters include at least upper edge parameters, lower edge parameters, left edge parameters, and right edge parameters; labeling the structural surfaces and fracture zones in the processed borehole images to construct an initial training sample set; expanding the initial training sample set using data augmentation methods to obtain a training sample set for structural surfaces and fracture zones; dividing the training sample set for structural surfaces and fracture zones into a training set, a validation set, and a test set; training the structural surface recognition model using the training set, adjusting the parameters of the structural surface recognition model using the validation set, evaluating the accuracy of the structural surface recognition model using the test set, and saving the structural surface recognition model whose accuracy reaches a preset standard as the trained structural surface recognition model.
[0036] In this embodiment, since engineering borehole images typically contain information such as image title, frame, scale, and borehole wall image, this information may affect the subsequent structural surface identification and positioning process. Therefore, image cropping needs to be achieved by setting certain cropping pixel parameters. The specific cropping process is as follows: Figure 3 As shown, the cutting formula is as follows:
[0037]
[0038]
[0039] In the formula: , These represent the horizontal and vertical pixel sizes of the borehole wall rock mass image in the borehole image, respectively; Width×Height represents the pixel size of the borehole image. , The left and right edge parameters that need to be cropped in the width direction of the entire borehole image; , The parameters for the top and bottom edges that need to be cropped along the length of the entire borehole image.
[0040] It should be noted that edge parameters can be preset and adjusted according to different models of borehole imaging instruments or different image output formats. For new equipment, the system automatically detects areas where color, texture, or contrast changes abruptly (such as the boundary between the ruler and the rock wall), initially suggests clipping lines, and allows the user to confirm or fine-tune them to obtain the edge parameters.
[0041] Figure 4 A schematic diagram illustrating borehole image annotations according to an embodiment of the present disclosure is shown, such as... Figure 4 As shown, the image was labeled using Labelimg software, and the labels were saved as .txt files (normalized coordinates) to construct the dataset. The quality of the model training results largely depends on the quality of the dataset. Therefore, when using bounding boxes for labeling, it is necessary to carefully identify the accurate location of the structural surfaces and minimize the size of the bounding boxes as much as possible to ensure that the labeled bounding boxes can accurately mark the location of the structural surfaces. This process was completed by the engineering implementation personnel. The labelers accurately marked the location of the rock mass structural surfaces by drawing bounding boxes and saved the labels as .txt files with normalized coordinates. The labeling quality directly determines the accuracy of the model in identifying and locating the structural surfaces.
[0042] Figure 6 A schematic diagram illustrating structural surface recognition according to an embodiment of the present disclosure is shown; the trained model is tested, and some test results are as follows. Figure 6 As shown, the rectangle represents the model's recognition result of the borehole image. The accuracy of each model is statistically analyzed, and models with an accuracy of 90% or higher are saved. Then, the learning sample images of models with an accuracy of less than 90% are reorganized. Figure 6 (a) shows the results of dense structure surface identification; Figure 6 (b) shows the identification results of the structure surface with shadow interference; Figure 6 (c) shows the underwater structure surface identification results; Figure 6 (d) shows the result of overlapping structural surface identification; Figure 6 (e) shows the results of large-area structural surface identification; Figure 6The middle (f) shows the result of the fracture zone identification. The rectangle in the figure represents the automatic identification result of the model, which intuitively reflects the detection capability of the algorithm under different imaging conditions and geological features. It verifies the accuracy and generalization ability of the model in practical engineering applications and can cope with complex situations such as shadow interference, underwater imaging and dense structural surfaces.
[0043] Through the embodiments of this disclosure, the original borehole image is cropped to remove non-hole wall interference areas such as image name, image frame, and scale. The effective borehole image area is calculated by formula, which removes potential interference in the subsequent structural surface identification and positioning process, and provides accurate and clean input data for the deep learning model.
[0044] Based on the above embodiments, the initial training sample set is expanded using data augmentation methods to obtain a training sample set for structural surfaces and fracture zones. This includes: using the Mixup data augmentation method to generate synthetic borehole image samples containing mixed features by performing pixel-level linear combination of any two images and their labels in the initial training sample set; applying various geometric transformations to the synthetic borehole image samples to expand the number of samples, thereby constructing a training sample set for structural surfaces and fracture zones. The geometric transformations include at least one of flipping, rotation angle, and combinations of flipping and rotation angle.
[0045] Currently, various data augmentation methods exist, such as rotation, flipping, Gaussian blur, random cropping, and random padding, which can be used to augment the original image, thereby improving the model's generalization ability. This disclosure utilizes a Python script to first apply the Mixup data augmentation method to the image to increase sample diversity, and then uses image geometric transformation operations to augment the dataset, such as... Figure 5 As shown. Mixup belongs to the Vicinity Risk Minimization framework. Its core idea is to introduce neighborhood smoothing between training samples, thereby constructing a continuous input-label space. The Mixup method generates a new synthetic image by performing pixel-level linear combination of any two images. Specifically, let (x... i y i ) and (x j y j ) are two samples and their labels randomly selected from the training set. The Mixup is generated as follows:
[0046]
[0047] Where λ ~ Beta(α,α), α∈(0,1) controls the interpolation intensity, and setting α=0.5 achieves a good balance. The generated virtual samples ( , During training, it is used as input to calculate the loss function:
[0048]
[0049] like Figure 5 As shown, the synthesized image simultaneously preserves the structural texture and fracture information of both types of rocks, improving the model's robustness to transitional types and mixed features without introducing additional sample acquisition costs. After augmentation with seven geometric transformations—horizontal, vertical, horizontal and vertical (hv), 90° rotation (rot90), 90° rotation and horizontal (rot90_h), 90° rotation and vertical (rot90_v), and 270° rotation (rot270)—the number of images per image (including the original image) can be increased by up to eight times. After Mixup and geometric flipping data augmentation operations, the 648 training set images can be expanded to 85,184 images. These images are then randomly divided into training, validation, and test sets in a 7:1:2 ratio to construct the final dataset.
[0050] In the embodiments of this disclosure, given that machine learning requires learning from a large number of image samples, relying solely on the acquired images is insufficient to meet the high-precision training requirements of the model. Therefore, it is necessary to expand the dataset to obtain more dataset images. By using Mixup data augmentation and various geometric transformation methods to expand the original structural surface image samples, the original training images are effectively expanded without introducing additional acquisition costs. This significantly increases the size and diversity of the dataset, thereby improving the adaptability and robustness of the deep learning model to different rock mass structures and imaging conditions.
[0051] Based on the above embodiments, the pixel resolution is calculated using edge detection and contour analysis algorithms based on the depth markers in the target borehole image. This includes: performing grayscale conversion and Gaussian filtering preprocessing on the image region containing the depth markers in the target borehole image; using edge detection and contour analysis algorithms to identify contours matching the depth markers and extracting the uppermost and lowermost pixel coordinates of the contours; and calculating the pixel resolution based on the actual physical distance corresponding to the depth markers, the uppermost pixel coordinates, and the lowermost pixel coordinates.
[0052] Pixel resolution is calculated based on borehole images. Given that the pixel information of images acquired via downhole television is known and fixed, determining the actual physical length represented by one pixel is necessary to calculate the length of each section of the borehole wall. Typically, the downhole television image before cropping includes depth markers on the left side (e.g., depth indicators). Figure 3As shown in the image, by utilizing the Canny edge detection algorithm to highlight the edge information of depth markers in the image, and combining this with image processing, the number of pixels for the depth markers can be obtained. The following methods are all implemented in Python code, and the specific implementation mainly consists of three steps:
[0053] (1) Image preprocessing: First, the image is converted to grayscale using the following formula:
[0054]
[0055] In the formula, R represents the pixel value of the red channel, G represents the pixel value of the green channel, and B represents the pixel value of the blue channel.
[0056] After converting the image to grayscale, a Gaussian filter is used to remove noise and improve the accuracy of edge detection.
[0057]
[0058] In the formula, x represents the distance of the pixel from the center point in the horizontal direction, y represents the distance of the pixel from the center point in the vertical direction, and σ represents the standard deviation of the Gaussian function, which controls the shape of the Gaussian function.
[0059] (2) Use the contour analysis algorithm to identify closed contours, and select the contours that best match the depth identifier based on the shape characteristics of the contours. Finally, extract the boundary points of the contours and find the pixel coordinates of the uppermost and lowermost points as the upper and lower endpoint coordinates of the depth identifier for subsequent pixel resolution calculation.
[0060] (3) Using the above edge detection algorithm and contour analysis algorithm, the distance between the image pixels corresponding to the actual depth marker can be obtained, and finally the required pixel resolution p can be obtained:
[0061]
[0062] In the formula, p represents pixel resolution, in centimeters per pixel; D represents actual physical distance, in centimeters; and G represents image pixel distance, in pixels.
[0063] Through the embodiments of this disclosure, by adopting a standardized image processing procedure to identify depth markers, image noise is effectively suppressed and the geometric edges of depth markers are extracted, solving the problem of inaccurate depth marker positioning caused by uneven image illumination and noise interference, which in turn affects the accuracy of pixel resolution calculation.
[0064] Based on the above embodiments, the distance between adjacent structural surfaces and the length of the broken zone are calculated according to the identification results of the structural surfaces and the broken zone and the pixel resolution. This includes: calculating the actual physical distance between adjacent structural surfaces based on the depth difference of the points in the identified structural surface rectangle and the pixel resolution; and calculating the actual physical length of the broken zone based on the depth difference of the upper and lower edges of the identified broken zone rectangle and the pixel resolution.
[0065] Figure 7 This schematically illustrates a calculation diagram of the distance between adjacent structural surfaces and the length of the fracture zone according to an embodiment of the present disclosure, such as... Figure 7 As shown in the figure, rectangles are used to mark the distance between two structural surfaces and one fracture zone. When calculating RQD, the distance between the center points of the identification boxes of adjacent structural surfaces needs to be calculated and compared with a preset threshold (usually 10cm). In the figure, D1=5.3cm, which is less than 10cm, so this segment is counted as a short rock segment in the short rock segment ratio RQD1; D2=22.3cm, which is greater than 10cm, meets the RQD statistical requirements and is not deducted as a short rock segment; the fracture zone length T1=33.5cm is counted in the fracture zone ratio RQD2.
[0066] In the embodiments of this disclosure, for structural surfaces, length refers to the distance between two adjacent structural surfaces. Therefore, the depth difference between the center points of the identification frames of two adjacent structural surfaces is multiplied by the pixel resolution to obtain the length of the rock mass fracture. For fracture zones, length refers to the width of the fracture zone itself. Therefore, the depth difference between the upper and lower edges of the same fracture zone identification frame is multiplied by the pixel resolution to obtain the actual length of the fracture zone.
[0067] Based on the above embodiments, the proportion of short rock segments with lengths less than a preset threshold and the proportion of fractured zones are statistically analyzed based on the distance between adjacent structural surfaces and the length of the fractured zone. A deduction method is then used to calculate the rock mass quality index (RQD). This includes: selecting short rock segments with lengths less than the preset threshold based on the distance between adjacent structural surfaces; statistically analyzing the fractured zone segments based on their lengths; calculating the proportion of short rock segments by the sum of their lengths relative to the total borehole depth; calculating the proportion of fractured zones by the sum of their lengths relative to the total borehole depth; and subtracting the proportion of short rock segments and the proportion of fractured zones from 100% to obtain the rock mass quality index (RQD).
[0068] Figure 8This diagram illustrates an RQD calculation according to an embodiment of the present disclosure. First, through model recognition, the program automatically locates and marks structural planes and fractured zones. Then, in the parameter calculation stage, the system converts the pixel distance in the image to the actual physical scale based on the recognition results, statistically analyzes the proportion of short rock segments with lengths less than a preset threshold and the proportion of fractured zones, and uses a subtraction method to quantify the rock mass integrity. Finally, specific analytical values are provided: given an actual borehole depth of 100 cm and a corresponding image pixel distance of 1500 pixels, the calculated pixel scale is approximately 0.0667 cm per pixel, and the final RQD of this rock mass segment is 61.25%. This intuitively presents the complete process from original image input to automatic output of key rock mass quality indicators and the final application effect.
[0069] Through the embodiments of this disclosure, all calculated spacing between adjacent structural surfaces and length of fracture zones are uniformly identified and classified for cumulative purposes, and short rock segment deduction items and fracture zone deduction items are accurately separated from continuous rock mass length data.
[0070] Based on the above embodiments, when calculating the proportion of short rock segments RQD1 and the proportion of fractured zones RQD2, the overlap between the short rock segment region and the fractured zone region is determined; if there is an overlap, the overlapping part is only included in the proportion of fractured zones RQD2 and is not included in the proportion of short rock segments RQD1 again.
[0071] In this embodiment, the distance between adjacent structural surfaces and the length of the fracture zone are calculated based on the identification results. After obtaining the length information of each segment, short rock segments and fracture zone segments with a length of less than 10 cm are statistically analyzed by setting a threshold t, and the RQD is automatically calculated using a subtraction method. The RQD visualization calculation process is as follows: Figure 8 As shown. RQD value calculation formula:
[0072]
[0073]
[0074]
[0075]
[0076]
[0077]
[0078] In the formula RQD1, the proportion of rock segments smaller than 10cm is represented; RQD2 represents the proportion of fractured zones; D i T is the distance (cm) between adjacent structural surfaces in the i-th segment. lt is the length of the l-th fracture zone (cm); t is the statistical threshold, taken as 10cm; L is the borehole depth (cm); h i h i-1 Let i be the depth of the points within the rectangular frame of the structure plane, and i-1 be the depth (cm) of the points within the rectangular frame of the structure plane; f(x) is the discriminant function; p l p l-1 The depth (cm) of the lower edge of the rectangular frame of the fracture zone in segment l and the upper edge of the rectangular frame of the fracture zone in segment l-1 are defined; n is the number of segments.
[0079] When calculating RQD1 and RQD2, the overlap between the short rock segment region and the fracture zone region is determined. If there is an overlap, the overlapping part is only included in the fracture zone ratio RQD2 and is not included in the short rock segment ratio RQD1.
[0080] In the embodiments of this disclosure, the classic definition of RQD only calculates the percentage of the cumulative length of intact rock segments with a length greater than or equal to 10 cm to the total borehole depth. This disclosure uses a deduction method to calculate RQD, dividing the total borehole depth into intact rock segments that meet the RQD statistical requirements, short rock segments that do not meet the RQD statistical requirements, and fractured zone segments. First, the proportion of short rock segments (RQD1) is calculated based on the distance between adjacent structural surfaces; then, the proportion of fractured zones (RQD2) is calculated based on the fractured zone identification results; finally, the proportions of short rock segments and fractured zones are subtracted by 100% to obtain the final RQD.
[0081] The method proposed in this disclosure involves borehole data acquisition, specifically using a borehole television imaging device to capture in-situ images of the rock mass deep underground and transmitting these images to a ground receiver via optical fiber. The main implementation process is as follows: After drilling, the borehole television imaging device is inserted into the hole to acquire the required borehole images and transmit them to the ground. A Python-based algorithm is used to crop the original borehole images to avoid interference from other information affecting subsequent deep learning. The cropped borehole images are then filtered and bounded using LabelImg software—images lacking structural planes, fracture zones, or of very poor quality are filtered out, and structural planes and fracture zones are bounded in two categories, ensuring the bounding boxes are sufficiently accurate. Since deep learning requires a large number of training samples, the filtered and bounded training samples are further augmented using Python code. The processed training set is fed into a computer for training. A trial-and-error approach is used to continuously adjust parameters such as epochs (number of training rounds), InitialLearnRate, and MiniBatchSize to find the optimal training parameters. After training, different models with different parameters are saved for testing and comparison. Each model is tested, and models with low accuracy are eliminated based on the test results. Downhole drilling images from a real-world engineering case are input into the final model. Relevant calculation formulas are inserted into the image recognition loop. After the drilling image is successfully recognized, parameters are calculated. By manually inputting the actual physical distance information of the drilling image as prompted, the pixel resolution and corresponding RQD of the measured drilling image can be calculated.
[0082] To verify the applicability and stability of the proposed RQD calculation method based on borehole images under different lithological conditions, three typical engineering cases were selected for analysis.
[0083] To investigate geological conditions and evaluate rock mass quality, photogrammetry was conducted using downhole television equipment at five boreholes across three projects to acquire borehole image data. To further verify the practicality of this method, automatic RQD analysis was performed on these five borehole image samples, with each borehole image being 1 meter long. First, to reduce the impact of noise points or lighting conditions in the original borehole wall images on image processing, the original images needed to be enhanced. The enhanced images were processed using a trained YOLOv11 algorithm to identify and locate structural surfaces in the images. Then, the length of each core segment was automatically measured, and the rock mass quality index (RQD) was intelligently calculated according to the definition of RQD. Finally, the RQDs of each segment were summed and averaged to obtain the RQD of the entire borehole. The RQD obtained through automatic analysis was named WRQD.
[0084] Error analysis was performed on the comparison results between the automatically calculated WRQD values and the manually measured RQD values obtained from field measurements (e.g.) Figure 9 (As shown in the figure). WRQD represents the RQD value calculated based on the intelligent recognition algorithm, RQD represents the result obtained through manual measurement, and Error is the relative error between the two. Overall, the trends of WRQD and manual RQD are highly consistent, both remaining between 70% and 95%, indicating that the intelligent algorithm can accurately reflect the integrity characteristics of the rock mass. Specifically, the errors of samples 1 and 2 are relatively large, at 9.51% and 9.61% respectively. This is mainly because the core samples at these two locations were severely damaged during core extraction, making it impossible to obtain the actual RQD value manually. This demonstrates the superiority of the proposed in-situ RQD calculation based on borehole images. The errors of samples 2, 3, and 4 are only 4.30%, 3.10%, and 6.95% respectively, showing good error control. This indicates that when the core surface is clear and fracture characteristics are significant, the algorithm's recognition results are highly consistent with manual judgment. Overall, the average error of the five boreholes was about 6.7%, which is within an acceptable range. This indicates that the method can avoid the influence of the drilling and coring process, obtain the RQD value more objectively, and the model has good stability and generalization ability.
[0085] Figure 9 The analysis of errors between manual measurement and intelligent calculation of RQD shows that the trends of WRQD and manual RQD are highly consistent, with the relative error controlled between 3.10% and 9.61%, and the average error is about 6.7%. This comparison result intuitively verifies that the method has good accuracy and reliability under different geological conditions and lithological backgrounds, providing objective and reliable data support for rock mass quality evaluation.
[0086] Borehole wave velocity test results are another indicator reflecting rock mass quality. Generally, the wave velocity test results and RQD variation patterns of the same borehole should be basically consistent. Higher wave velocities and RQD values usually represent better rock mass quality. This is because intact rock masses have a dense internal structure and fewer fractures, resulting in faster wave propagation speeds; conversely, the lower the RQD value, the more fractured the rock mass, and the lower the wave velocity. This is because fractured rock masses have more internal fractures and voids, leading to slower wave propagation speeds. The comparison results of the calculated RQD value and the P-wave velocity test results for Sample 1 are as follows: Figure 10 As shown.
[0087] like Figure 10 As shown, the distribution pattern of WRQD is highly consistent with that of P-wave velocity (Vp). Both show high values in intact rock mass sections and decrease synchronously in fractured zones, verifying the reliability and scientific nature of the method of the present invention in rock mass integrity evaluation, and further confirming the accuracy and engineering application value of intelligent calculation of RQD results.
[0088] from Figure 10 As can be seen from the comparison images, the borehole P-wave velocity test results are typically between 2.0 and 5.0 km / s, which is basically consistent with the WRQD distribution obtained by intelligent calculation, and the two show a good correlation. In conclusion, the RQD intelligent calculation method based on YOLOv11 and borehole images is reliable, and its analysis results can provide accurate data support for rock mass integrity assessment.
[0089] Compared to traditional RQD acquisition methods that rely on core measurements, this invention utilizes an automated computational framework based on in-situ borehole images and deep learning detection. This framework significantly improves efficiency and objectivity while maintaining accuracy, and is adaptable to various complex imaging environments and different lithological conditions. Comprehensive engineering verification demonstrates the following key implementation effects:
[0090] (1) Improve data processing efficiency and analysis accuracy: By using deep learning-based automatic identification of structural surfaces and intelligent RQD calculation, efficient and automated processing of borehole images is achieved, avoiding the tedious and repetitive labor in traditional manual measurement. The method has been verified in multiple engineering cases with an average error of approximately 6.7%, which meets the engineering accuracy requirements. At the same time, the WRQD value obtained by intelligent calculation is highly consistent with the borehole wave velocity test results, demonstrating the reliability and accuracy of this method.
[0091] (2) Provide in-situ, objective RQD assessment: The in-situ images of the borehole wall obtained by downhole television are used directly to identify the structural surface and calculate the RQD, which effectively avoids mechanical damage and sample loss caused by core collection and transportation, thus more realistically reflecting the original structural characteristics of the rock mass and reducing subjective errors of human interpretation.
[0092] (3) Enhanced model generalization ability and engineering applicability: Through data augmentation techniques such as Mixup and geometric transformation, the number and diversity of training samples were significantly expanded, and the model's adaptability to different lithologies and imaging conditions (such as shadows, underwater, fracture zones, etc.) was improved. It showed good recognition effect and computational stability in a variety of complex geological environments.
[0093] Based on the above-mentioned method for calculating rock mass quality indicators based on borehole images, this disclosure also provides a system for calculating rock mass quality indicators based on borehole images, which can be used to implement the above-mentioned method for calculating rock mass quality indicators based on borehole images. The system includes: a structural surface recognition module, used to identify structural surfaces and fracture zones in a target borehole image using a trained structural surface recognition model. The structural surface recognition model is configured to identify structural surfaces and fracture zones in the target borehole image based on the input target borehole image. The structural surface recognition model is trained using a training sample set of structural surfaces and fracture zones expanded by data augmentation and geometric transformation; a resolution calculation module, used to calculate pixel resolution based on depth markers in the target borehole image using edge detection and contour analysis algorithms; a distance and length calculation module, used to calculate the distance between adjacent structural surfaces and the length of fracture zones based on the identification results of structural surfaces and fracture zones and the pixel resolution; and an indicator calculation module, used to calculate the proportion of short rock segments with lengths less than a preset threshold and the proportion of fracture zones based on the distance between adjacent structural surfaces and the length of fracture zones, and to calculate the rock mass quality indicator RQD using a subtraction method.
[0094] Based on the same inventive concept, this disclosure also provides an electronic device for implementing the above-described method for calculating rock mass quality indicators based on borehole images. The electronic device according to embodiments of this disclosure includes a processor that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage portion into a random access memory (RAM). The processor may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor, and / or an associated chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may include a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this disclosure.
[0095] The RAM stores various programs and data required for the operation of the electronic device. The processor, ROM, and RAM are interconnected via a bus. The processor performs various operations of the method flow according to embodiments of the present disclosure by executing programs in the ROM and / or RAM. It should be noted that the programs may also be stored in one or more memories other than ROM and RAM. The processor may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.
[0096] According to embodiments of this disclosure, the electronic device may further include an input / output (I / O) interface, which is also connected to a bus. The electronic device may also include one or more of the following components connected to the I / O interface: an input section including a keyboard, mouse, etc.; an output section including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section including a hard disk, etc.; and a communication section including a network interface card such as a LAN card, modem, etc. The communication section performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. Removable media, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as needed.
[0097] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.
[0098] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this disclosure.
[0099] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include, but are not limited to, languages such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).
[0100] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0101] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.
[0102] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.
Claims
1. A method for calculating rock mass quality indicators based on borehole images, characterized in that, include: The trained structural surface recognition model is used to identify structural surfaces and fracture zones in the target borehole image. The structural surface recognition model is configured to identify structural surfaces and fracture zones in the target borehole image based on the input target borehole image. The structural surface recognition model is trained using a training sample set of structural surfaces and fracture zones expanded by data augmentation and geometric transformation. Based on the depth markers in the target borehole image, the pixel resolution is calculated using edge detection and contour analysis algorithms; Based on the identification results of structural surfaces and fracture zones and the pixel resolution, the distance between adjacent structural surfaces and the length of the fracture zone are calculated. Based on the distance between adjacent structural planes and the length of the fracture zone, the proportion of short rock segments with a length less than a preset threshold and the proportion of fracture zones are statistically analyzed, and the rock mass quality index RQD is calculated using the deduction method.
2. The method according to claim 1, wherein, The training method for the structural surface recognition model includes: Based on the edge parameters, the original borehole image dataset is cropped to remove interference areas such as image name, image frame, and scale, resulting in a processed borehole image. The edge parameters include at least the upper edge parameter, lower edge parameter, left edge parameter, and right edge parameter. The structural surfaces and fracture zones in the processed borehole images are labeled to construct an initial training sample set; The initial training sample set is expanded using data augmentation methods to obtain training sample sets for structural surfaces and fracture zones; The training sample set of the structural plane and fracture zone is divided into a training set, a validation set, and a test set; The training set is used to train the structural surface recognition model, the validation set is used to adjust the parameters of the structural surface recognition model, and the test set is used to evaluate the accuracy of the structural surface recognition model. The structural surface recognition model that achieves the preset accuracy standard is saved as the trained structural surface recognition model.
3. The method according to claim 2, wherein, The process of expanding the initial training sample set using data augmentation methods to obtain a training sample set for structural surfaces and fracture zones includes: The Mixup data augmentation method is used to generate synthetic borehole image samples containing mixed features by performing pixel-level linear combination of any two images and their labels in the initial training sample set. The synthetic borehole image samples are subjected to various geometric transformations to expand the number of samples, thereby constructing the training sample set of the structural surface and fracture zone. The geometric transformations include at least one of flipping, rotation angle, and combination of flipping and rotation angle.
4. The method according to claim 1, wherein, The step of calculating pixel resolution based on depth markers in the target borehole image using edge detection and contour analysis algorithms includes: The image regions containing depth markers in the target borehole image are preprocessed by grayscale conversion and Gaussian filtering; Using edge detection and contour analysis algorithms, contours that match depth identifiers are identified, and the uppermost and lowermost pixel coordinates of the contours are extracted. The pixel resolution is calculated based on the actual physical distance corresponding to the depth identifier, the coordinates of the uppermost pixel, and the coordinates of the lowermost pixel.
5. The method according to claim 1, wherein, The step of calculating the distance between adjacent structural surfaces and the length of the fracture zone based on the identification results of the structural surfaces and the fracture zone and the pixel resolution includes: The actual physical distance between adjacent structural surfaces is calculated based on the depth difference of the points within the identified structural surface rectangle and the pixel resolution. The actual physical length of the broken zone is calculated based on the depth difference between the upper and lower edges of the identified broken zone rectangle and the pixel resolution.
6. The method according to claim 1, wherein, The method of calculating the rock mass quality index (RQD) using the deduction method includes: Based on the distance between adjacent structural planes, short rock segments with a length less than a preset threshold are selected. Based on the length of the fracture zone, the fracture zone sections are counted; Calculate the proportion of the sum of the lengths of the short rock segments to the total borehole depth to obtain the proportion of short rock segments; Calculate the ratio of the sum of the lengths of the fractured zones to the total borehole depth to obtain the fractured zone percentage; The rock mass quality index RQD is obtained by subtracting the proportion of short rock segments and fractured zones from 100%.
7. The method according to claim 6, wherein, The method further includes: When calculating the proportion of the short rock segment and the proportion of the fractured zone, the overlap between the short rock segment region and the fractured zone region is determined. If there is overlap between the two, the overlapping part will only be included in the proportion of the fracture zone, and will not be included in the proportion of the short rock segment.
8. A rock mass quality index calculation system based on borehole images, characterized in that, Capable of implementing the method as described in any one of claims 1 to 7, comprising: The structural surface recognition module is used to identify structural surfaces and fracture zones in a target borehole image using a trained structural surface recognition model. The structural surface recognition model is configured to identify structural surfaces and fracture zones in the target borehole image based on the input target borehole image. The structural surface recognition model is trained using a training sample set of structural surfaces and fracture zones augmented with data augmentation and geometric transformation. The resolution calculation module is used to calculate the pixel resolution based on the depth markers in the target borehole image, using edge detection and contour analysis algorithms. The distance and length calculation module is used to calculate the distance between adjacent structural surfaces and the length of the fracture zone based on the identification results of the structural surfaces and the fracture zone and the pixel resolution. The index calculation module is used to calculate the proportion of short rock segments with a length less than a preset threshold and the proportion of fracture zones based on the distance between adjacent structural surfaces and the length of the fracture zone, and to calculate the rock mass quality index RQD using the deduction method.
9. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.
