A deep learning-based drilling image intelligent picking method and system

Abstract

The application discloses a kind of based on deep learning's borehole image automatic pickup algorithm model BIAR (Borehole Image Auto-Recognition).First, the borehole data collected by borehole probe is used to make the data set of borehole image, and the instance segmentation module in the deep learning network model YOLOv8 is used to construct the multi-classification task to automatically pick up the different categories of structure surface of dike filling and fissure on the borehole wall image, then the mask coordinates obtained after instance segmentation are fitted into the sinusoidal curve of structure surface according to the polynomial fitting formula, and the extreme points of the curve after fitting are calculated through mathematical transformation, space transformation and other ways, so as to obtain the occurrence of the corresponding dip and dip angle of structure surface, finally output the borehole image after automatic analysis and drawing of algorithm model.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of artificial intelligence and geophysical exploration, specifically to a method and system for intelligent acquisition of borehole images based on deep learning. Background Technology

[0002] Rock mass structural planes are an important component of rock masses. Their distribution and combination significantly affect the strength, deformation, and seepage characteristics of the rock mass, playing a crucial role in geological exploration, engineering, and resource development. In practical engineering, borehole camera technology is commonly used to analyze information from borehole wall images and digital core images to interpret fractures or discontinuities. These interpretations include the depth, dip, dip angle, width, surface roughness, and infill properties of the structural planes. The distribution of fractures obtained from borehole images provides data processing personnel with intuitive data for assessing stratigraphic characteristics, rock types, and rock formation stability. However, with the continuous development of borehole television technology and the increasing depth of boreholes, the amount of image data that needs to be processed is also increasing. Furthermore, since the identification of fractures and discontinuities in borehole wall images mostly relies on the experience of data processing personnel, the efficiency and accuracy of manual borehole wall image acquisition become particularly prominent issues. How to achieve automated identification of structural surfaces in borehole images through various algorithms has gradually become a research hotspot for solving this problem.

[0003] The earliest automatic detection methods for borehole images used a series of feature extraction algorithms based on the Hough transform. These algorithms could automatically identify and extract various line features from borehole wall images and obtain accurate structural surface information using methods such as polynomial fitting. Based on this idea, many researchers used edge detection algorithms such as the Canny operator or the Sobel operator to further extract structural surface information from borehole images, while improving these algorithms to better meet the requirements of feature curve extraction from borehole images. However, edge detection algorithms suffer from problems such as high noise levels and unclear edge features when extracting edges from borehole images.

[0004] With the rapid development of deep learning in image processing, more and more scholars are choosing to use algorithms such as CNN (Convolutional Neural Networks) to perform object detection and image segmentation on borehole images to analyze and calculate relevant parameters of rock mass structural surfaces within the images. However, using CNN models for object detection in borehole images has a serious problem: object detection algorithms can only identify the location of structural surfaces, but cannot identify their specific morphology like edge detection algorithms. If the specific morphology of the structural surfaces cannot be identified, parameters such as their dip and dip angle cannot be calculated. However, CNN-based image segmentation tasks can effectively identify the specific location and morphology of structural surfaces, thus providing a foundation for the subsequent calculation of the attitude of the structural surfaces.

[0005] However, a common problem in the application of both edge detection and object detection algorithms to borehole images is the lack of classification of structural surfaces. In actual data processing, researchers need to identify not only the location and orientation of structural surfaces, but also whether they are joints, fissures, or dike infills, and whether their morphology is open or closed. In most cases, closed joints and dike infills do not significantly affect the properties of the rock mass, and researchers typically do not select these stable structural surfaces when capturing images. However, edge detection and image segmentation algorithms identify all structural surfaces, both those that should and should not be selected. This approach can lead to misjudgments in subsequent engineering design and construction.

[0006] Deep learning methods utilize artificial neural networks to form instance segmentation networks. By extracting structural features such as vein filling and fractures from borehole wall images, they automate the borehole image picking process. Compared to traditional manual picking, deep learning-based automated borehole image picking algorithms completely eliminate manual intervention in the processing of borehole exploration data, achieving a high degree of automation and significantly improved picking accuracy. Furthermore, trained on large datasets of borehole images, these algorithms can adapt to work area scenarios of varying complexity, demonstrating strong generalization ability and robustness. Summary of the Invention

[0007] The purpose of this invention is to provide a deep learning-based intelligent borehole image acquisition method and system. Multi-class acquisition is performed using the YOLOv8_seg instance segmentation model to extract features and obtain mask coordinates for structural surfaces such as dike infill and fractures in the borehole wall image. Subsequently, the coordinates are input into a polynomial fitting algorithm to calculate the curves and extreme points of the structural surfaces. Finally, based on the fitting results, the dip and dip angle of the structural surfaces are automatically calculated, and the analyzed and plotted borehole image is output.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A deep learning-based intelligent borehole image acquisition method includes the following steps:

[0010] Step 1: Collect borehole data to obtain a dataset of borehole images, and divide the collected dataset into training set, validation set and test set;

[0011] Step 2: Using the YOLOv8_seg instance segmentation model, a multi-class picking task is constructed to extract features from different types of structural surfaces such as vein filling and fractures on the borehole wall image, and the corresponding mask coordinates of each structural surface are obtained as the basis for the next step of the algorithm model calculation.

[0012] Step 3: Input the mask coordinates obtained in Step 2 into the polynomial fitting formula, and calculate the structure curve of the structure surface in the image based on the mask coordinates obtained from the model, thereby obtaining the extreme points of the structure surface fitting curve;

[0013] Step 4: Then, based on the extreme values ​​of the curves fitted in Step 3 and the corresponding parameters, calculate the dip and dip angle of the structural surface, and finally output the borehole image after automatic analysis and drawing by the algorithm model.

[0014] As a further aspect of the present invention: in step one, the dataset is divided into a training set, a validation set, and a test set in an 8:1:1 ratio.

[0015] As a further aspect of the present invention: in step two, the instance segmentation model YOLOv8_seg picks up the unconformities of different shapes in the hole wall image;

[0016] The labels for structural surfaces include Close, Open, and Dike Filling.

[0017] As a further aspect of the present invention: the picking rule for Close in structural surface picking is the structural surface region in the image that is in a closed state;

[0018] In structural surface picking, the picking rule for Open is to select structural surface regions that are in an open state;

[0019] In the structural surface picking, the picking rule for Dike Filling is to select structural surface areas that are clearly distinguishable from the rock mass in color, such as those containing calcite or mica.

[0020] As a further aspect of the present invention: In step three, when the BIAR automatic drilling image picking algorithm model is training the dataset, the input dataset format is COCO format, the backbone network uses the backbone network built into YOLOv8, the optimizer is AdamW, the learning rate is designed to be 1e-4, and the GPU is used to accelerate the training process without freezing the training.

[0021] As a further aspect of the present invention: the process of instance segmentation using the YOLOv8_seg instance segmentation model is as follows:

[0022] The input module preprocesses the input image (drill hole image), including scaling and normalization, to ensure the image fits the model's input size and format. It then extracts target features from the input image using the Backbone, transforming the image into a feature representation with rich semantic information. Next, the FPN (Feature Pyramid Networks) module performs multi-scale feature fusion to construct multi-scale feature maps, enhancing the model's ability to detect targets of different sizes. PANet (Path Aggregation Network) further integrates feature information from different levels, improving the expressive power of the feature maps. The Detection Head then performs target classification and bounding box regression, generating class predictions and bounding box regression values ​​(Bbox) for each candidate region. After target detection, the Instance Segmentation Head performs pixel-level segmentation of the image and outputs the precise contour of each target. During model training, the Loss Function module calculates the difference between the model output and the ground truth annotations and adjusts the model parameters through backpropagation to minimize the loss. The post-processing module uses NMS (Non-Maximum Suppression) to remove redundant overlapping boxes, filtering out overlapping detection boxes to obtain accurate target locations. Finally, the algorithm outputs the detection and segmentation results, including the target category, confidence score, bounding box coordinates, and segmentation mask.

[0023] As a further aspect of this invention: the structural surface mask obtained in the previous image segmentation step is fitted using polynomial fitting to obtain the corresponding curve parameters. Before fitting the data using a polynomial function, it is first necessary to define a polynomial function f(x) = a based on the fitting requirements. n x n +a n-1 x n-1The form is +...+a1x+a0. Then, using optimization algorithms such as least squares, the formula for minimizing the sum of squared residuals is used. The polynomial coefficients are solved by using the sum of squares of the differences between the actual observed values ​​and the predicted values ​​from the fitted curve. Finally, once the polynomial coefficients are determined, the fitted polynomial function can be used for further data analysis of the mask coordinates.

[0024] Where, in the formula, a n a n-1 ..., a1, a0 are the coefficients to be found, n is the degree of the polynomial, and y... i Let f(x) be the i-th observed value of the dependent variable, i.e., the actual value recorded in the experiment or data. i () is based on the model and the independent variable x i The predicted value is calculated from the value.

[0025] As a further aspect of the present invention: the extreme points A(MinX,MinY) and B(MaxX,MaxY) of the structure surface calculated in the fitting formula are connected by the formula. The absolute value of the difference between the coordinates of the extreme points is calculated by taking the difference in their y-axis coordinates. val and Y val And calculate the difference according to the formula. Calculate the dip angle of each structural plane.

[0026] Where, Y in the formula cal C is the actual diameter of the borehole. degrees This is the calculated angle of inclination.

[0027] As a further aspect of the invention: the orientation of the lowest point of the structural surface curve on the y-axis is mapped between 0° and 360°. If the coordinates of the maximum point x... i From 0 to x mid (The minimum values ​​of the borehole image on the x-axis) can be mapped to an average value between 180° and 360°; if the coordinates of the maximum point x i In x mid To x max If the borehole image's tendency is between its maximum value on the x-axis, then its tendency can be mapped to an average value between 0° and 180°. The mapping rule formula is as follows:

[0028]

[0029] Where, α in the formula Angle The structure surface inclination is obtained after calculating the piecewise function.

[0030] A deep learning-based intelligent borehole image acquisition system, characterized in that it includes:

[0031] The data acquisition module is used to acquire borehole data to obtain a dataset of borehole images, and divides the acquired dataset into a training set, a validation set, and a test set.

[0032] The instance segmentation module uses the YOLOv8_seg instance segmentation model to construct a multi-class picking task, extracts features from different types of structural surfaces such as dike filling and fractures in the borehole image, and obtains the corresponding mask coordinates of each structural surface as the basis for the next calculation of the algorithm model.

[0033] The attitude calculation module is used to input the initially obtained structure surface mask coordinates into a polynomial function to fit a structure surface curve that approximates a sine function. The dip and dip angle of the structure surface are then calculated using the extreme values ​​of the fitted structure surface curve.

[0034] The automatic borehole image picking module uses a trained model to automatically and quickly pick up structural surfaces such as fractures and discontinuities in the borehole wall image amidst the interference of complex information, calculate their orientation, and finally output the borehole image automatically picked by the model.

[0035] The beneficial effects of this invention are as follows: This invention not only completes the entire process from image recognition to parameter calculation, but also achieves intuitive visualization of the results (automatically labeling and plotting the calculated occurrence information on the borehole image). Simultaneously, it provides geological engineers with an intuitive and convenient analysis tool, which can greatly improve the efficiency and accuracy of geological exploration and engineering assessment, and can provide new ideas and solutions for the development of borehole image analysis towards intelligence and automation. Attached Figure Description

[0036] The invention will now be further described with reference to the accompanying drawings.

[0037] Figure 1 The diagram shows the overall structural framework of the BIAR algorithm model.

[0038] Figure 2 Here is a diagram of the YOLOv8_seg model structure;

[0039] Figure 3 This is the result of a polynomial fitting;

[0040] Figure 4 The structural diagram shows the inclination angle of the structural surface in space.

[0041] Figure 5 A structural diagram showing the inclination of the structural planes within space;

[0042] Figure 6This represents the automated picking results of the BIAR algorithm model. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Example 1

[0045] Please see Figure 1 As shown, this invention is a method and system for intelligent borehole image acquisition based on deep learning, comprising the following steps:

[0046] Step 1: Building the dataset

[0047] To verify the model's ability to automatically pick up various complex structural surfaces in borehole images and to improve the model's accuracy and generalization as much as possible during training;

[0048] In this study, data from six wells located in different areas in southern China were selected. These wells together contain over 1,000 meters of borehole image data.

[0049] The strata in the area where the borehole data is located are mostly composed of dense igneous and metamorphic rocks, and the structural planes in the borehole images are relatively clear and regular.

[0050] The borehole data was processed according to the rule of one image for every 50 centimeters of actual depth, resulting in a dataset containing 2,000 borehole images.

[0051] The obtained borehole image dataset was divided into training, validation and test sets in an 8:1:1 ratio, and the velocity spectrum images were labeled using the Labelimg software.

[0052] Step 2: Instance Segmentation

[0053] During instance segmentation, structural plane picking includes, but is not limited to, Close, Open, and Dike Filling labels, thus transforming a single single-classification task into a multi-classification task;

[0054] Among them, the picking rule for Close in structural surface picking is the structural surface region in the image that is in a closed state;

[0055] In structural surface picking, the picking rule for Open is to select structural surface regions that are in an open state;

[0056] In the structural surface picking, the picking rule for Dike Filling is to select structural surface areas that are clearly distinguishable from the rock mass in color, such as those containing calcite or mica.

[0057] Step 3: Polynomial Fitting

[0058] To accurately depict the structural surface masks identified in the previous image segmentation step, a polynomial fitting method was employed. This process begins by defining the basic form of the polynomial function based on the specific requirements of the fitting.

[0059] Then, optimization techniques such as least squares are used to minimize the sum of squares of the deviations between the actual observed values ​​and the polynomial predicted values ​​by adjusting the coefficients of the polynomial, that is, to find the best fitting curve.

[0060] When these polynomial coefficients are solved precisely, the optimized polynomial function can be used to perform more in-depth data analysis on the mask coordinates, thereby achieving an accurate fit to the characteristics of the structural surface.

[0061] Step 4: Automatic acquisition of borehole images

[0062] After training, the BIAR (Borehole Image Automatic Picking) algorithm model can recognize various complex structural surface information. When a borehole image is input, the BIAR algorithm model automatically picks it and calculates the attitude of the corresponding structural surface.

[0063] The BIAR (Biologically Analyzed Arithmetic Image Acquisition) algorithm model constructed in this embodiment can accurately identify a large number of borehole images and output instance segmentation images in a short time. It can accurately identify relevant information of structural surfaces such as dike infill and fractures in the borehole images, and can accurately and quickly calculate the corresponding attitude information of each structural surface. Compared with the traditional manual acquisition method, it greatly improves the efficiency and accuracy of velocity spectrum acquisition, and does not require manual intervention.

[0064] Example 2

[0065] like Figure 2As shown, the instance segmentation algorithm model YOLOv8_seg can complete the instance segmentation task of borehole images through a series of modules and finally output the structural surface picking results. First, the input module preprocesses the input image (borehole image), including scaling and normalization, to ensure that the image is adapted to the input size and format of the model. Next, the backbone extracts target features from the input image and transforms the image into a feature representation with rich semantic information. Then, FPN (Feature Pyramid Networks) constructs multi-scale feature maps through multi-scale feature fusion, enhancing the model's ability to detect targets of different sizes. PANet (Path Aggregation Network) further integrates feature information from different levels to improve the expressive power of the feature maps. Next, the Detection Head is responsible for target classification and bounding box regression, generating the category prediction and bounding box regression value (Bbox) for each candidate region. After the target is detected, the Instance Segmentation Head performs pixel-level segmentation of the image and outputs the precise contour of each target. During the model training process, the LossFunction module calculates the difference between the model output and the ground truth annotation and adjusts the model parameters through backpropagation to minimize the loss value. The post-processing module uses Non-Maximum Suppression (NMS) to remove redundant overlapping bounding boxes and filter out overlapping detection boxes, thereby obtaining accurate target locations. Finally, the algorithm outputs the detection and segmentation results, including the target category, confidence score, bounding box coordinates, and segmentation mask. Through the collaborative work of these modules, YOLOv8 can efficiently and accurately perform instance segmentation of structural surfaces in borehole images.

[0066] Example 3

[0067] The overall process of polynomial fitting is as follows: First, the form of the polynomial function is defined according to the fitting requirements. Then, optimization algorithms such as the least squares method are used to solve for the polynomial coefficients by minimizing the sum of squared residuals (i.e., the sum of squares of the differences between the actual observed values ​​and the predicted values ​​of the fitted curve). Finally, once the polynomial coefficients are determined, the fitted polynomial function can be used for further data analysis of the mask coordinates.

[0068] like Figure 3 As shown in Figure a, the image represents a drill hole with an actual length of 80 cm and a diameter of 15 cm. The blue area represents the mask area obtained in the previous image segmentation process, which is surrounded by hundreds of mask coordinates. Figure 3b is an image in which the coordinates of these hundreds of masks are plotted in a rectangular coordinate system. The coordinates of the magnified points in the image are the coordinates of the mask in this coordinate system. The coordinate system is established with the upper left corner of each drilling image as the origin, the horizontal axis as the x-axis, and the vertical axis as the y-axis. At the same time, the coordinates of each picking point are set as the coordinates of the pixel points on the image (without units). Figure 3 c is a comparison between the fitted curve obtained by fitting using the polynomial fitting formula and the structural surface in the original figure. It can be seen that the curve retains the sinusoidal characteristics of the rock mass structural surface as much as possible.

[0069] Specifically:

[0070] f(x) = a n x n +a n-1 x n-1 +...+a1x+a0

[0071]

[0072] Among them, a n a n-1 a1, a0 are the coefficients to be determined, and n is the degree of the polynomial. i Let f(x) be the i-th observed value of the dependent variable, i.e., the actual value recorded in the experiment or data. i () is based on the model and the independent variable x i The predicted value is calculated from the value.

[0073] Example 4

[0074] To further calculate the attitude of structural planes in borehole images, the intersection of the stratum bedding plane and any imaginary horizontal plane is called the strike line, which is a straight line connecting two points of equal elevation on the same bedding plane. The directions indicated by its two ends are the strike of the stratum, which can be represented by two azimuth angles 180° apart. It represents the horizontal extension direction of the stratum in space. The straight line on the stratum bedding plane that is perpendicular to the strike line and extends downward along the slope is called the dip line. The direction indicated by the projection of the dip line onto the horizontal plane is called the dip direction of the stratum. The dip direction indicates the direction in which the stratum dips. The angle between the dip line and its projection onto the horizontal plane is called the dip angle. It represents the angle between the stratum bedding plane and the horizontal plane and reflects the degree of dip of the stratum.

[0075] like Figure 4 As shown in Figure a, point A is the highest point where the structural plane intersects with the core, point B is the lowest point where the structural plane intersects with the core, point C is the intersection of the downward extension of point A and the horizontal plane where point B is located, straight line AB is the dip line, the direction pointed to by straight line CB is the dip direction of the rock layer, and the angle α between straight line AB and straight line CB is the dip angle of the rock layer.

[0076] Meanwhile, after calculating the fitting formula in the previous step, we can obtain the coordinates of the extreme points of the fitting curve of each structural surface in the coordinate system: “MaxX”, “MinX”, “MaxY”, and “MinY” (the coordinates are the positions of the pixels where the extreme points are located in the coordinate system).

[0077] Given two legs of a right triangle, the angle between one leg and the hypotenuse can be calculated using the arctangent function. Therefore, to know the dip angle α of a structural plane, it is only necessary to know the distance AC between the highest point A where the structural plane intersects the core and the intersection point C on the horizontal plane containing the lowest point B where the structural plane intersects the core, and the borehole diameter BC. For example... Figure 4 As shown in b, since the borehole diameter BC is known in actual work, if we want to calculate the dip angle of the structural surface from the coordinates of the extreme points in the borehole image, we only need to know the difference in coordinates on the y-axis between the extreme points A(MinX,MinY) and B(MaxX,MaxY) of the structural surface, and convert the difference into the actual length of segment AC in the core according to the standard of 1500 pixels per meter in the borehole image. Then we can calculate the dip angle of the structural surface based on the arctangent function.

[0078] like Figure 5 As shown in Figure a, if the x-coordinate of the maximum point A is XEast (one-quarter of the way along the x-axis in the borehole image), then its corresponding azimuth is due east. Simultaneously, according to the definition of a sine function in the two-dimensional unfolded diagram of the structural surface, the azimuth corresponding to its minimum point B is due west. Therefore, the dip direction indicated by the dip line of this structural surface is 270°. Figure 5 As shown in b, if the coordinate of the maximum point A in the x-direction is XWest (three-quarters of the way along the x-axis in the borehole image), then its corresponding azimuth is due west, and the azimuth of its minimum point B is due east. Therefore, the dip line of this structural surface indicates a dip of 90°. Figure 5 As shown in Figure c, if the coordinate of the maximum point A in the x-direction is XNE (one-eighth of the borehole image on the x-axis), then its corresponding orientation is northeast, and the orientation of its minimum point B is southwest. Therefore, the dip line of this structural surface indicates a dip of 225°. Similarly, if the coordinate xi of the maximum point is between 0 and xmid, then its dip can be mapped to an average of 180°-360°; if the coordinate xi of the maximum point is between xmid and xmax (the maximum value of the borehole image on the x-axis), then its dip can be mapped to an average of 0°-180°.

[0079] like Figure 6 The image shown is the result of automatic borehole image acquisition. Figure 6 'a' represents two overlapping and intersecting structural surfaces. As can be seen, even when the two structural surfaces are mixed together, the algorithm can still distinguish the differences between the different structural surfaces very well. Figure 6 b and c are the results of picking structural surfaces with different opening degrees; Figure 6 d represents two structural surfaces of different categories that are very close to each other; Figure 6 e represents multiple structural surfaces of the same category that are very close to each other; Figure 6 f represents an irregular structural surface with an irregular surface curve. Through Figure 6 The BIAR algorithm model can be observed to automatically and accurately pick up structural surfaces and calculate their attitude in various complex borehole images.

[0080] Specifically:

[0081]

[0082] Among them, Y cal C is the actual diameter of the borehole. degrees α is the calculated inclination angle. Angle The structure surface inclination is obtained after calculating the piecewise function.

[0083] Example 5

[0084] A deep learning-based intelligent borehole image acquisition system includes:

[0085] The data acquisition module is used to acquire borehole data to obtain a dataset of borehole images, and divides the acquired dataset into a training set, a validation set, and a test set.

[0086] The instance segmentation module and the target picking module use the YOLOv8_seg instance segmentation model to construct a multi-class picking task to extract features from different types of structural surfaces such as dike filling and fractures in the borehole image, and obtain the corresponding mask coordinates of each structural surface as the basis for the next calculation of the algorithm model.

[0087] The attitude calculation module is used to input the initially obtained structure surface mask coordinates into a polynomial function to fit a structure surface curve that approximates a sine function. The dip and dip angle of the structure surface are then calculated using the extreme values ​​of the fitted structure surface curve.

[0088] The automatic borehole image picking module uses a trained model to automatically and quickly pick up structural surfaces such as fractures and discontinuities in the borehole wall image amidst the interference of complex information, calculate their orientation, and finally output the borehole image automatically picked by the model.

[0089] This invention not only completes the entire process from image recognition to parameter calculation, but also achieves intuitive visualization of the results (automatically labeling and plotting the calculated occurrence information on the borehole image). Simultaneously, it provides geological engineers with an intuitive and convenient analysis tool, which can greatly improve the efficiency and accuracy of geological exploration and engineering assessment, and can offer new ideas and solutions for the development of borehole image analysis towards intelligence and automation.

[0090] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for intelligent acquisition of borehole images based on deep learning, characterized in that, Includes the following steps: Step 1: Collect borehole data to obtain a dataset of borehole images, and divide the collected dataset into a training set, a validation set, and a test set; Step 2: Using the YOLOv8_seg instance segmentation model, a multi-class picking task is constructed to extract features from different categories of structural surfaces of dike filling and fractures on the borehole wall image, and obtain the corresponding mask coordinates of each structural surface as the basis for the next calculation of the algorithm model; wherein, the labels of the structural surfaces include three categories: Close, Open, and Dike Filling; the picking rule for Close is the structural surface area in the image that is in a closed state, the picking rule for Open is the structural surface area in an open state, and the picking rule for Dike Filling is the structural surface area that contains calcite or mica, which are clearly distinguishable from the rock mass in color; Step 3: Input the mask coordinates obtained in Step 2 into the polynomial fitting formula, and calculate the structure curve of the structure surface in the image based on the mask coordinates obtained from the model, thereby obtaining the extreme points of the structure surface fitting curve; Step 4: Then, based on the extreme values ​​of the fitted curves and corresponding parameters obtained in Step 3, calculate the dip and dip angle of the structural surface. Finally, output the borehole image automatically analyzed and drawn by the algorithm model. The calculation process for the dip angle is as follows: connect the extreme points A(MinX, MinY) and B(MaxX, MaxY) of the structural surface calculated in the fitting formula using the formula... The absolute value of the difference between the coordinates of the extreme points is calculated by taking the difference in their y-axis coordinates. val and Y val And calculate the difference according to the formula. Calculate the dip angle of each structural plane; where Y in the formula... cal C is the actual diameter of the borehole. degrees The calculated dip angle is the dip angle measure; the calculation process for the dip orientation is as follows: the mapping of the lowest point of the structural surface curve on the y-axis to the azimuth between 0° and 360°, if the coordinates of the maximum point x i From 0 to x mid Between these ranges, its tendency is to be mapped to an average of 180°~360°; if the coordinates of the maximum point x i In x mid To x max If the value is between 0° and 180°, then the average value tends to be mapped to the range of 0° to 180°. The mapping rule formula is as follows: ； Where α~Angle~ is the structural surface inclination obtained after piecewise function calculation.

2. The method for intelligent borehole image acquisition based on deep learning according to claim 1, characterized in that, In step one, the dataset is divided into training set, validation set and test set in an 8:1:1 ratio.

3. The intelligent borehole image acquisition method based on deep learning according to claim 1, characterized in that, In step three, when training the BIAR automatic borehole image picking algorithm model on the dataset, the input dataset format is COCO, the backbone network uses the backbone network built into YOLOv8, the optimizer is AdamW, the learning rate is designed to be 1e-4, and the training process is accelerated using GPU without freezing the training.

4. The method for intelligent borehole image acquisition based on deep learning according to claim 1, characterized in that, The process of instance segmentation using the YOLOv8_seg instance segmentation model is as follows: The input module preprocesses the input image, which is a borehole image, including scaling and normalization to ensure the image fits the model's input size and format. It then extracts target features from the input image using the Backbone module, transforming the image into a feature representation with rich semantic information. Next, the FPN module performs multi-scale feature fusion to construct multi-scale feature maps, enhancing the model's ability to detect targets of different sizes. PANet further integrates feature information from different levels to improve the expressive power of the feature maps. Finally, the Detection Head is responsible for target classification and bounding box regression, generating class predictions and bounding box regression values ​​for each candidate region. After detecting a target, Instance SegmentationHead performs pixel-level segmentation of the image and outputs the precise contour of each target. During model training, the LossFunction module calculates the difference between the model output and the ground truth annotations and adjusts the model parameters through backpropagation to minimize the loss value. The post-processing module removes redundant overlapping boxes and filters out overlapping detection boxes through Non-Maximum Segmentation (NMS) to obtain the accurate target location. Finally, the algorithm outputs the detection and segmentation results, including the target category, confidence score, bounding box coordinates, and segmentation mask.

5. The method for intelligent borehole image acquisition based on deep learning according to claim 1, characterized in that, The process of fitting the structural surface curve is as follows: The structure surface mask obtained in the previous image segmentation step is fitted using a polynomial fitting method to obtain the corresponding curve parameters. Before fitting the data using the polynomial function, it is first necessary to define the polynomial function according to the fitting requirements. The form is then used; then, optimization algorithms such as least squares are employed to minimize the formula for the sum of squared residuals. The polynomial coefficients are then determined; finally, once the polynomial coefficients are determined, the fitted polynomial function is used to perform further data analysis on the mask coordinates. Where, in the formula, a n a n-1 ..., a1, a0 are the coefficients to be found, n is the degree of the polynomial, and y... i Let f(x) be the i-th observed value of the dependent variable, i.e., the actual value recorded in the experiment or data. i () is based on the model and the independent variable x i The predicted value is calculated from the value.

6. A deep learning-based intelligent borehole image acquisition system, used to implement the method as described in claim 1, characterized in that, include: The data acquisition module is used to acquire borehole data to obtain a dataset of borehole images, and divides the acquired dataset into a training set, a validation set, and a test set. The instance segmentation module uses the YOLOv8_seg instance segmentation model to construct a multi-class picking task, extracts features from different types of structural surfaces such as dike filling and fractures in the borehole image, and obtains the corresponding mask coordinates of each structural surface as the basis for the next calculation of the algorithm model. The attitude calculation module is used to input the initially obtained structure surface mask coordinates into a polynomial function to fit a structure surface curve that approximates a sine function, and to calculate the dip and dip angle of the structure surface through the extreme values ​​of the fitted structure surface curve. The automatic borehole image picking module uses a trained model to automatically and quickly pick up fractures and discontinuous surface structures amidst the interference of complex information on the borehole wall image, calculate their orientation, and finally output the borehole image automatically picked by the model.

Images