A Method and System for Processing Borehole Development Maps Based on Deep Learning and Spatial Inversion

Abstract

This invention discloses a method and system for processing borehole unfolded images based on deep learning and spatial inversion, belonging to the field of image data processing technology. It includes: acquiring the original panoramic unfolded image of the borehole and its corresponding basic parameter information; extracting and identifying features from the original panoramic unfolded image based on a rock and fracture collaborative identification model to obtain rock identification results and initial fracture response results: fracture main branch response results and fracture centerline auxiliary response results; mapping the rock identification results to prior rock information; automatically extracting and selecting candidate targets for natural fractures in the original panoramic unfolded image based on the initial fracture response results and prior rock information; performing trajectory fitting and position parameter determination on the selected effective fracture candidate targets to obtain fracture trajectories; and combining borehole attitude parameters to complete fracture dip and dip angle inversion to obtain fracture parameter results. This invention improves the accuracy and stability of natural fracture identification.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, specifically to a method and system for processing borehole unfolded images based on deep learning and spatial inversion. Background Technology

[0002] In the construction of water conservancy and hydropower projects, horizontal borehole television (BMT) technology is widely used in scenarios such as tunnel surrounding rock exploration, slope internal structure detection, dam foundation rock mass investigation, underground cavern surrounding rock evaluation, and grouting effect inspection. By acquiring borehole wall images and generating panoramic unfolded images, BMT equipment can intuitively reflect the color, texture, structural characteristics, and fracture development of the surrounding rock, thus becoming an important technical means for obtaining geological information inside boreholes. Compared with traditional core sampling or localized point detection, BMT panoramic imaging provides continuous borehole wall image information, which has high application value in refined engineering geological analysis.

[0003] With the continuous improvement of the resolution of borehole television imaging equipment and the increasing demands for precision in engineering surveys, the amount of panoramic unfolded image data generated by a single horizontal borehole has increased significantly. For long-distance boreholes, multi-hole joint detection, or continuous construction monitoring scenarios, the scale of acquired image data is even larger, leading to a heavy workload for subsequent data processing. In existing technologies, the processing of borehole television panoramic unfolded images is still mostly based on manual interpretation. Technicians need to analyze the surrounding rock conditions, structural features, and fracture development in the borehole wall images segment by segment, and then combine experience to complete lithology judgment, fracture location identification, and fracture attitude estimation. This processing method is time-consuming and inefficient, and cannot meet the large-scale, rapid, and standardized data processing needs of water conservancy and hydropower projects.

[0004] Furthermore, existing manual processing methods rely heavily on professional experience, and the interpretation results are significantly influenced by the interpreter's knowledge background, engineering experience, and subjective perception. Different technicians may obtain different lithological stratification results, fracture identification results, and fracture attitude judgments for the same borehole image, leading to insufficient consistency and repeatability in data interpretation. Especially when the surrounding rock texture is complex, the image quality fluctuates significantly, or the fractures are densely developed, manual interpretation is prone to omissions, misjudgments, and result biases, which is detrimental to the objective extraction of engineering geological information and its use in subsequent decision-making.

[0005] On the other hand, the applicability of existing automated image processing methods in horizontal borehole television panoramic unfolded images remains limited. Borehole television panoramic unfolded images are not ordinary planar images, but are formed by unfolding the image of the cylindrical borehole wall, with a specific geometric mapping relationship between the image coordinates and the actual spatial position of the borehole wall. Simultaneously, due to the imaging environment, images often exhibit uneven illumination, local blurring, mud adhesion, and water stain reflections. Natural fractures often appear as thin, weak-contrast, locally interrupted linear targets interwoven with background textures, easily confused with the original rock texture, drill mark edges, or noise artifacts. For surrounding rock lithology identification, different lithologies vary complexly in color, grain structure, bedding structure, and surface texture, making it difficult to simultaneously meet the dual requirements of fine-grained lithology identification and accurate extraction of natural fractures using general image algorithms.

[0006] Existing technologies also have shortcomings in calculating fracture geometry parameters. Fractures in borehole television panoramic unfolded images typically appear as curved trajectories. Obtaining the fracture dip and inclination depends not only on the image morphology of the fracture on the unfolded image but also on the borehole axis orientation, borehole inclination, and the image unfolding method. Current processing methods often rely on empirical interpretation or simple curve fitting to estimate fracture attitude, lacking an integrated technical approach encompassing panoramic image recognition, fracture trajectory extraction, borehole wall cylindrical coordinate mapping, and spatial geometric inversion. This results in limited accuracy and stability in calculating fracture dip and inclination, making it difficult to develop a systematic and automated processing solution suitable for practical engineering applications.

[0007] Therefore, in response to the problems of large data volume, low efficiency of manual processing, reliance on experience and insufficient automation of horizontal borehole panoramic imaging data, there is an urgent need to propose a horizontal borehole panoramic imaging data processing method that can be applied to water conservancy and hydropower engineering scenarios, while taking into account automatic identification of surrounding rock lithology, automatic identification of natural fractures, automatic location of fracture borehole depth, and automatic calculation of fracture inclination and dip angle, so as to improve data processing efficiency, reduce reliance on manual labor, and enhance the objectivity and consistency of results. Summary of the Invention

[0008] The technical problem this invention aims to solve is the existing issues in processing horizontal borehole panoramic video unfolded images, including large data volume, low efficiency of manual interpretation, susceptibility of fracture identification to interference from complex surrounding rock textures, and reliance on experience in calculating fracture dip and inclination. This invention aims to provide a borehole unfolded image processing method and system based on deep learning and spatial inversion. It constructs a complete technical solution from panoramic unfolded image input, surrounding rock zoning identification, natural fracture constraint extraction, to automatic fracture attitude inversion, applicable to the automatic data processing of horizontal borehole panoramic video unfolded images in water conservancy and hydropower projects. This invention uses a complete panoramic video unfolded image corresponding to a single borehole as the processing object. Combining basic parameters such as borehole depth, diameter, borehole azimuth, and inclination angle, it achieves automated extraction of borehole wall geological information through steps such as image standardization, automatic surrounding rock lithology identification, automatic extraction of natural fractures, fracture trajectory fitting, and fracture attitude inversion. Finally, it outputs the surrounding rock lithology identification results, as well as the borehole depth location, circumferential location, dip, and inclination angle corresponding to each natural fracture. This scheme can further transform the surrounding rock zoning results into prior constraint information for fracture identification based on the automatic identification of surrounding rock types, thereby improving the accuracy and stability of natural fracture identification and solving the above technical problems.

[0009] This invention is achieved through the following technical solution:

[0010] In a first aspect, the present invention provides a borehole unfolding image processing method based on deep learning and spatial inversion, the method comprising:

[0011] Obtain the original panoramic unfolded image of the borehole and its corresponding basic parameter information, and establish the first mapping relationship between the vertical pixels of the image and the borehole depth, and the second mapping relationship between the horizontal pixels of the image and the circumferential angle of the borehole wall;

[0012] Based on a pre-constructed collaborative identification model for surrounding rock and fractures, features are extracted and identified from the original panoramic unfolded image to obtain surrounding rock identification results and initial fracture response results; the initial fracture response results include the main branch response results and the auxiliary response results of the fracture centerline.

[0013] Based on the first mapping relationship, the surrounding rock identification results are mapped to surrounding rock prior information; the surrounding rock prior information includes surrounding rock category prior and surrounding rock boundary prior.

[0014] Based on the initial response results of the fractures, natural fractures in the original panoramic unfolded image are automatically extracted and candidate targets are screened based on the prior information of the surrounding rock to obtain effective fracture candidate targets.

[0015] Based on the second and first mapping relationships, trajectory fitting and position parameter determination are performed on the effective fracture candidate targets to obtain the fracture trajectory; the fracture trajectory is mapped to the borehole wall cylindrical coordinate system, and the fracture inclination and dip angle are inverted in combination with the borehole attitude parameters to obtain the fracture parameter results.

[0016] Furthermore, the basic parameter information includes borehole depth, borehole wall circumferential angle, borehole wall radius, borehole axis azimuth angle, and borehole inclination angle.

[0017] Furthermore, the method also includes:

[0018] The original panoramic unfolded image is preprocessed and standardized to obtain a standardized input tensor.

[0019] Preprocessing and standardization include: brightness equalization, local contrast enhancement and numerical normalization, as well as circumferential position prior.

[0020] Furthermore, the surrounding rock and fracture collaborative identification model extracts surrounding rock texture features and fracture structure features within the same model framework. It performs multi-scale feature extraction on the original panoramic unfolded image through a shared coding backbone to obtain the same shared feature expression suitable for surrounding rock discrimination and fracture identification.

[0021] The expression for the collaborative identification model of surrounding rock and fractures is:

[0022] ;

[0023] in, This indicates the output of the surrounding rock identification branch, which outputs the probability of the surrounding rock category and the corresponding surrounding rock identification result in different borehole depth sections of the image;

[0024] This indicates the output of the crack segmentation branch, which outputs the initial probability response results of natural cracks in the original panoramic unfolded image;

[0025] This indicates the auxiliary branch output of the fracture centerline, which outputs the probability result of the location of the main fracture skeleton;

[0026] Table model parameter set.

[0027] Furthermore, based on the first mapping relationship, the surrounding rock identification result is mapped to surrounding rock prior information; the surrounding rock prior information includes surrounding rock category prior and surrounding rock boundary prior, including:

[0028] Based on the first mapping relationship, the surrounding rock identification results are directly mapped to the surrounding rock category results within the borehole depth range;

[0029] The surrounding rock category results of longitudinally adjacent sections are subjected to continuity constraints and smoothing to obtain a set of surrounding rock intervals arranged in order of borehole depth within the entire borehole range;

[0030] Based on the set of surrounding rock intervals, a sequence of surrounding rock categories distributed along the vertical direction of the image is constructed, and a surrounding rock category function is constructed based on the surrounding rock category sequence and used as a priori for surrounding rock categories.

[0031] Based on the boundary locations between the surrounding rock sections, a priori boundaries for the surrounding rock are constructed.

[0032] Furthermore, based on the initial response results of the fractures, natural fractures in the original panoramic unfolded image are automatically extracted and candidate targets are screened based on prior information of the surrounding rock to obtain effective fracture candidate targets, including:

[0033] Based on the prior knowledge of the surrounding rock category, an adaptive fracture extraction threshold for the surrounding rock category is constructed; and according to the relationship between the fracture extraction threshold and the fracture enhancement response map, a candidate binary map of fracture is obtained; the fracture enhancement response map is obtained by fusing the fracture main branch response results and the fracture centerline auxiliary response results.

[0034] Connected regions are extracted from the binary map of fracture candidates to form an initial set of fracture candidate targets;

[0035] The overlap ratio of the surrounding rock boundary is introduced as a constraint, and the overlap ratio is calculated based on the candidate targets in the initial fracture candidate target set and the prior knowledge of the surrounding rock boundary.

[0036] Candidate targets with an overlap ratio greater than a preset boundary overlap threshold are identified as low-confidence candidate targets, and these low-confidence candidate targets are suppressed or eliminated to obtain further crack candidate targets.

[0037] By combining the prior knowledge of the surrounding rock category, further fracture candidate targets are geometrically screened to obtain effective fracture candidate targets and their corresponding fracture centerline point sets.

[0038] Furthermore, based on the second and first mapping relationships, trajectory fitting and position parameter determination are performed on the effective fracture candidate targets to obtain the fracture trajectory, including:

[0039] Based on the set of fracture centerline points of effective fracture candidate targets, the fracture trajectory is curve-fitted to establish the fracture trajectory in the original panoramic unfolded image.

[0040] Based on the distribution of the fracture centerline point set along the vertical axis of the image, the representative hole depth of the fracture is determined; based on the representative hole depth location and the first mapping relationship, the corresponding representative hole depth of the fracture is determined.

[0041] The starting and ending depths of the fracture are determined based on the minimum and maximum longitudinal positions of the fracture centerline point set, and the corresponding depth range of the fracture is determined based on the starting and ending depths.

[0042] The circumferential angle of the borehole wall is determined based on the average lateral position of the point set of the fracture centerline and the second mapping relationship.

[0043] Furthermore, by combining the borehole attitude parameters, the fracture dip and inclination angle are inverted to obtain fracture parameter results, including:

[0044] By introducing drilling attitude parameters, a transformation relationship between the local drilling coordinate system and the engineering space coordinate system is constructed.

[0045] Based on the transformation relationship, solve for the spatial normal vector of the fracture surface;

[0046] The dip direction and dip angle of the fracture are calculated based on the spatial normal vector as fracture parameter results.

[0047] Furthermore, the method also includes:

[0048] The results of surrounding rock identification and fracture parameters are summarized and organized into a structured result set, and the structured result set is output.

[0049] Secondly, the present invention provides a borehole unfolding image processing system based on deep learning and spatial inversion, the system comprising:

[0050] The acquisition unit is used to acquire the original panoramic unfolded image of the borehole and its corresponding basic parameter information, and to establish a first mapping relationship between the vertical pixels of the image and the borehole depth, and a second mapping relationship between the horizontal pixels of the image and the circumferential angle of the borehole wall.

[0051] The collaborative identification unit is used to extract features from the original panoramic unfolded image based on a pre-constructed collaborative identification model of surrounding rock and fractures, and obtain the surrounding rock identification results and the initial response results of fractures; the initial response results of fractures include the main branch response results of fractures and the auxiliary response results of fracture centerline;

[0052] The surrounding rock prior information unit is used to map the surrounding rock identification result into surrounding rock prior information according to the first mapping relationship; the surrounding rock prior information includes surrounding rock category prior and surrounding rock boundary prior;

[0053] The fracture candidate target screening unit is used to automatically extract and screen natural fractures in the original panoramic unfolded image based on the fracture initial response result and the prior information of the surrounding rock, so as to obtain effective fracture candidate targets.

[0054] The fracture trajectory determination unit is used to perform trajectory fitting and position parameter determination on the effective fracture candidate targets according to the second mapping relationship and the first mapping relationship to obtain the fracture trajectory;

[0055] The inversion calculation unit is used to map the fracture trajectory to the borehole wall cylindrical coordinate system, and combine the borehole attitude parameters to complete the inversion of fracture inclination and dip angle, and obtain fracture parameter results.

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] 1. The present invention is a borehole unfolding map processing method and system based on deep learning and spatial inversion. It integrates the surrounding rock identification, natural fracture extraction and fracture orientation inversion of a complete panoramic unfolding map of a single borehole into the same technology chain. Furthermore, it transforms the surrounding rock identification results into prior constraint information in the fracture identification process, so that the surrounding rock identification no longer exists as an independent output result, but directly serves the natural fracture identification and subsequent parameter calculation, thereby forming an integrated processing scheme with a clear technical coupling relationship.

[0058] 2. This invention relates to a borehole unfolding image processing method and system based on deep learning and spatial inversion. It constructs a collaborative identification model for surrounding rock and fractures in panoramic unfolded borehole images. Within the same deep learning framework, it simultaneously outputs surrounding rock identification results, initial response results for fracture regions, and auxiliary response results for fracture centerlines. The fracture identification branch serves as the main branch to characterize fracture region distribution, while the fracture centerline auxiliary branch acts as an auxiliary constraint branch to enhance the continuity and clarity of the fracture framework. This structure allows surrounding rock texture information and fracture structure information to be learned together in a shared feature space, while simultaneously improving the skeleton representation capability of the fracture region results.

[0059] 3. This invention relates to a borehole unfolding map processing method and system based on deep learning and spatial inversion. The invention constructs the surrounding rock identification results into surrounding rock category priors and surrounding rock boundary priors, and introduces them into the fracture response calculation in an explicit mathematical form, allowing the surrounding rock zoning results to directly affect the fracture identification process. Specifically, the surrounding rock category prior is used to enhance or modulate the true fracture response under different surrounding rock backgrounds, while the surrounding rock boundary prior is used to suppress the pseudo-fracture response distributed along lithological boundaries. This reduces the interference of surrounding rock texture boundaries, lithological boundaries, and pseudo-edges on natural fracture identification at both the pixel-level and candidate target-level, improving the accuracy and stability of fracture identification.

[0060] 4. This invention relates to a borehole development diagram processing method and system based on deep learning and spatial inversion. After obtaining effective natural fracture candidate targets filtered by prior constraints of the surrounding rock, this invention further establishes an automated calculation link from fracture trajectory fitting and borehole wall cylindrical coordinate mapping to spatial normal vector inversion. By transforming the fracture from a linear target in a two-dimensional image into a spatial intersection line on the borehole wall cylindrical surface, and combining the borehole axis azimuth angle and borehole inclination angle to automatically calculate the normal vector of the fracture surface, the dip and dip angle of the fracture are finally obtained. This allows the calculation of fracture attitude parameters to be based on clear spatial geometric relationships, avoiding the problem of excessive reliance on manual experience and empirical formulas in existing technologies. Attached Figure Description

[0061] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is a flowchart of the borehole unfolding diagram processing method based on deep learning and spatial inversion of the present invention;

[0063] Figure 2 This is a schematic diagram of the structure of the collaborative identification model for surrounding rock and fractures of the present invention;

[0064] Figure 3 This is a block diagram of the borehole unfolding image processing system based on deep learning and spatial inversion of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0066] The purpose of this invention is to provide a method and system for processing borehole unfolded maps based on deep learning and spatial inversion. This addresses the problems in existing horizontal borehole television panoramic unfolded map processing, such as large data volume, low efficiency of manual interpretation, susceptibility to interference from complex surrounding rock textures in fracture identification, and reliance on experience in calculating fracture dip and inclination. The invention aims to provide a complete technical solution for processing borehole unfolded maps based on deep learning and spatial inversion, constructing a complete technical solution from panoramic unfolded map input, surrounding rock zoning identification, natural fracture constraint extraction to automatic fracture attitude inversion. This solution can further transform the surrounding rock zoning results into prior constraint information for fracture identification based on automatic identification of surrounding rock categories, thereby improving the accuracy and stability of natural fracture identification.

[0067] Another objective of this invention is to construct a collaborative identification model for surrounding rock and fractures based on a panoramic unfolded image of the borehole wall. This model simultaneously outputs surrounding rock identification results, initial response results for fracture regions, and auxiliary response results for fracture centerlines within the same deep learning framework. The fracture identification branch serves as the main branch to characterize the distribution of fracture regions, while the fracture centerline auxiliary branch enhances the continuity of the fracture main branch's skeleton, thereby improving the clarity and continuity of the fracture main skeleton and providing a more reliable foundation for subsequent fracture candidate target selection and trajectory fitting.

[0068] Another objective of this invention is to transform the surrounding rock identification results from mere interpretation into computable prior information in the fracture identification process. Specifically, the surrounding rock identification results are constructed into a priori maps of surrounding rock categories and surrounding rock boundaries, and these prior maps are directly incorporated into the fracture response calculation. This enhances the true fracture response under different surrounding rock categories and suppresses pseudo-fracture responses distributed along the surrounding rock boundaries. In this way, the surrounding rock identification results directly influence the fracture identification stage, avoiding the technical independence between surrounding rock identification and fracture identification, thereby improving the technical coupling and inventiveness of the entire solution.

[0069] Another objective of this invention is to establish an automated calculation link from fracture identification results to fracture spatial attitude parameters. For valid natural fracture candidates selected after prior constraints on the surrounding rock, trajectory fitting is performed to determine their borehole depth and circumferential position. Then, the fracture trajectory is mapped from a two-dimensional panoramic unfolded image to the borehole wall cylindrical coordinate system. Combined with the borehole axis azimuth and borehole inclination angle, the spatial normal vector of the fracture surface is automatically inverted, and finally, the fracture dip and dip angle are calculated. This method automates the processing from borehole wall images to fracture attitude parameters, reduces reliance on manual experience, and improves the geometric rigor and engineering applicability of the results.

[0070] To achieve the above objectives, the present invention designs the following technical solution: Obtain a complete panoramic unfolded image of a single borehole and its corresponding basic parameters such as borehole depth, borehole diameter, borehole axis azimuth, and borehole inclination angle; perform standardized preprocessing on the panoramic unfolded image to construct a standardized image representation suitable for subsequent identification model input; utilize a collaborative identification model for surrounding rock and fractures to obtain surrounding rock identification results, fracture main branch response results, and fracture centerline auxiliary response results, respectively; map the surrounding rock identification results to surrounding rock category priors and surrounding rock boundary priors; construct a fracture enhancement response map under the combined effect of surrounding rock priors and centerline auxiliary results, and complete the extraction and screening of natural fracture candidate targets; perform trajectory fitting and position parameter determination on the retained effective fracture candidate targets; map the fracture trajectory to the borehole wall cylindrical coordinate system and combine it with borehole attitude parameters to complete the fracture dip and dip angle inversion; finally output the surrounding rock identification results and the number, borehole depth location, circumferential location, dip, and dip angle of each natural fracture.

[0071] Compared with existing technologies, this invention has at least the following beneficial effects: First, by transforming the surrounding rock identification results into prior constraints for fracture identification, it can effectively reduce the interference of surrounding rock texture boundaries, lithological boundaries, and local pseudo-edges on natural fracture identification, thereby improving the accuracy and stability of natural fracture identification. Second, through the synergistic effect of the main fracture branch and the auxiliary branch of the fracture centerline, the continuity and geometric stability of the main framework of natural fractures are improved. Third, by establishing the transformation relationship between two-dimensional image coordinates, borehole wall cylindrical coordinates, and engineering space coordinates, the automatic inversion of fracture dip and dip angle is realized. Fourth, an integrated data processing chain is formed from prior constraints of surrounding rock zoning, automatic fracture identification, to fracture occurrence output, which has a strong degree of automation and engineering application value.

[0072] Example 1

[0073] like Figure 1 As shown, Figure 1 This is a flowchart of the borehole unfolding image processing method based on deep learning and spatial inversion of the present invention; the borehole unfolding image processing method based on deep learning and spatial inversion of the present invention includes:

[0074] S1, acquire the original panoramic unfolded image of the borehole and its corresponding basic parameter information, and establish the first mapping relationship between the vertical pixels of the image and the borehole depth, and the second mapping relationship between the horizontal pixels of the image and the circumferential angle of the borehole wall;

[0075] Step S1 of this invention is used to complete the acquisition of the original panoramic unfolded image of a single hole and the construction of the basic parameter relationship, providing a unified data foundation for subsequent panoramic image preprocessing, surrounding rock lithology identification, natural fracture extraction, fracture location positioning, and fracture dip and dip angle inversion.

[0076] In this embodiment, each processing object is a single horizontal hole, and the borehole television imaging result of the horizontal hole is stored as a complete panoramic unfolded image of the hole wall. First, the original panoramic unfolded image corresponding to the single hole is read, and the corresponding basic parameter information is read simultaneously. The basic parameter information includes the borehole depth, the circumferential angle of the borehole wall, the radius of the borehole wall, the azimuth angle of the borehole axis, and the inclination angle of the hole.

[0077] Let the original panoramic unfolded image corresponding to the current single aperture be... for:

[0078]

[0079] in, Indicates the number of pixels vertically in the image. Indicates the number of pixels horizontally in the image. Represents the number of image channels. Any pixel in the image. Recorded as:

[0080]

[0081] in, Represents the horizontal pixel coordinates of the image. This represents the vertical pixel coordinates of the image. Since the vertical axis of the original panoramic unfolded image corresponds to the borehole axial unfolding direction, and the horizontal axis corresponds to the borehole wall circumferential unfolding direction, it is necessary to establish a mapping relationship between the vertical pixels and the actual borehole depth (i.e., the first mapping relationship), and a mapping relationship between the horizontal pixels and the borehole wall circumferential angle (i.e., the second mapping relationship).

[0082] For the first mapping relationship, let the initial hole depth of the single hole be... The depth of the termination hole is Then the actual hole depth length corresponding to each pixel in the vertical direction of the image is:

[0083]

[0084] in, This represents the calibration coefficient from the vertical pixel to the aperture depth. Therefore, any vertical pixel position in the image... The corresponding actual hole depth can be expressed as:

[0085]

[0086] in, Indicates pixel position The corresponding actual borehole depth value. Through this relationship, the subsequently identified surrounding rock sections, fracture locations, and fracture center points can be uniformly mapped to the actual borehole depth coordinates.

[0087] For the second mapping relationship, since the image horizontally corresponds to the circumference of the hole wall, let the image width be... Then any horizontal pixel position The corresponding circumferential angle of the hole wall is:

[0088]

[0089] in, Indicates pixel position The corresponding circumferential angle of the hole wall, in radians. If expressed in degrees, it can be written as:

[0090]

[0091] This relationship allows the lateral position of cracks or local anomalies in the image to be converted into the actual circumferential position of the borehole wall. Since the left and right boundaries of the original panoramic unfolded image are joined end-to-end on the actual borehole wall, this circumferential positional relationship also has periodic continuity, providing a basis for circumferential position perception in subsequent models.

[0092] To establish the geometric parameter foundation for subsequent fracture space inversion, borehole diameter information also needs to be read. Let the borehole diameter be... Then the radius of the hole wall is:

[0093]

[0094] in, This represents the borehole wall radius. This parameter is used to subsequently map the fracture trajectory in the 2D panoramic unfolded image to the borehole wall cylindrical coordinate system.

[0095] Furthermore, the inversion of fracture inclination and dip angle also depends on the attitude information of the borehole axis in space. Therefore, in step S1, the azimuth angle of the borehole axis is further read. and the bevel angle Based on these two parameters, the unit direction vector of the borehole axis in the engineering space coordinate system can be constructed:

[0096]

[0097] in, This represents the borehole axis direction vector, with its three components representing the projections of the borehole axis onto various directions in the spatial coordinate system. This direction vector will be used as a key input parameter in the subsequent calculation of the fracture surface normal vector and fracture orientation.

[0098] After constructing the parameters described above, a unified mapping relationship between image pixels and the basic parameter set can be further established. For any pixel in the image... Its corresponding basic parameters can be expressed as:

[0099]

[0100] in, Represents pixels The corresponding set of basic parameters includes borehole depth, circumferential angle, borehole wall radius, borehole azimuth angle, and borehole inclination angle. Through this unified mapping relationship, any pixel location in the image can have a clear engineering geological spatial meaning.

[0101] As a further implementation, the method also includes the following steps between S1 and S2: preprocessing and standardizing the original panoramic unfolded image to reduce the impact of factors such as uneven lighting, local noise, moisture reflection and circumferential boundary fracture on subsequent surrounding rock identification and fracture identification during borehole television imaging, and constructing a standardized input tensor suitable for subsequent collaborative identification model input.

[0102] Specifically, in this invention, the preprocessing is not a general process for ordinary planar images, but rather a process tailored to the imaging characteristics of the original panoramic unfolded images from borehole television. Since these images are formed by unfolding the cylindrical surface of the borehole wall, their lateral direction corresponds to the circumferential position of the borehole wall, their left and right boundaries are connected end-to-end in actual space, and their longitudinal direction corresponds to the axial depth of the borehole. Therefore, when performing image enhancement and standardization, it is necessary to preserve the surrounding rock texture and fracture edge information while maintaining the continuity of the circumferential structure.

[0103] First, the original panoramic unfolded image undergoes brightness equalization processing to reduce brightness fluctuations caused by light source attenuation, borehole wall reflection, and local shadows, thereby improving the discernibility of surrounding rock texture and fracture boundaries. Let the original panoramic unfolded image obtained in step S1 be... Its local background brightness estimate is Then the image after brightness equalization can be represented as:

[0104]

[0105] in, This represents the pixel value after brightness equalization. To prevent extremely small positive numbers with a denominator of zero, this processing reduces the slowly varying lighting components in the image, resulting in a more stable representation of differences in surrounding rock texture and fracture edge features.

[0106] Secondly, based on brightness equalization, local contrast enhancement and numerical normalization are performed on the image to ensure a relatively consistent data distribution among different single-aperture images when inputting them into the model. Let the minimum and maximum values ​​of the enhanced image be respectively... and The standardized image can then be represented as:

[0107]

[0108] in, This represents the normalized image pixel values. This process maps panoramic images obtained under different imaging conditions to a uniform numerical range, reducing input offset caused by differences in image brightness and contrast.

[0109] Considering that the left and right boundaries of the original panoramic borehole image are continuous on the actual borehole wall, to avoid artificially cutting off gaps or texture structures at the boundaries during subsequent recognition, this invention introduces circumferential continuity representation in the preprocessing stage. Specifically, after image standardization, the correspondence between the lateral position and the circumferential angle of the borehole wall is preserved, and the circumferential mapping function established in step S1 is used:

[0110]

[0111] Further, a circumferential position prior for the horizontal position of the image is constructed. The circumferential position prior is preferably encoded using a sine-cosine form to avoid circumferential angles being... and The numerical jumps between these values. Correspondingly, the circumferential position encoding is defined as:

[0112]

[0113] in, Indicates the horizontal position of the image The corresponding circumferential position prior vector. In this way, the circumferential periodic features of the borehole wall can be explicitly introduced into the subsequent model input, enabling the model to better identify cracks that span the left and right boundaries of the image and the surrounding rock texture features with circumferential regular distribution.

[0114] To match the subsequent collaborative identification model for surrounding rock and fractures, after image enhancement, normalization, and circumferential prior construction, the standardized image and circumferential position prior are combined to form the model input representation. Let the final preprocessed image be... The standardized input tensor can then be expressed as:

[0115]

[0116] in, This indicates a channel-based splicing operation. This represents the normalized input tensor used by subsequent models. Through this input representation, subsequent models can utilize not only the texture and edge information of the image itself, but also the geometric prior information of the circumferential position of the hole wall.

[0117] S2, based on the pre-constructed collaborative identification model of surrounding rock and fracture, extracts and identifies features of the standardized input tensor to obtain the surrounding rock identification results and the initial response results of the fracture; the initial response results of the fracture include the main branch response results of the fracture and the auxiliary response results of the fracture centerline;

[0118] Step S2 is used to construct a collaborative identification model for surrounding rock and fractures suitable for the original panoramic unfolded image of a horizontal borehole. Based on the obtained standardized input tensor, features are extracted and identified from the original panoramic unfolded image to obtain surrounding rock identification results and initial fracture response results. This step is one of the key steps of this invention, and its purpose is to provide a unified feature basis and initial identification results for subsequent prior construction of surrounding rock zoning and identification of natural fracture constraints.

[0119] In this invention, the collaborative identification model for surrounding rock and fractures does not treat surrounding rock identification and fracture identification as two independent processing steps. Instead, it extracts surrounding rock texture features and fracture structure features within the same model framework, ensuring that surrounding rock information and fracture information originate from the same set of shared feature expressions. This improves the model's adaptability to panoramic borehole wall unfolding images and creates conditions for subsequent fracture identification based on prior constraints of surrounding rock zoning.

[0120] That is, the standardized input tensor obtained in the previous step is Therefore, the collaborative identification model of surrounding rock and fracture constructed in this invention can be expressed as:

[0121]

[0122] in, This indicates the output of the surrounding rock identification branch, which outputs the probability of the surrounding rock category and the corresponding surrounding rock identification result in different borehole depth sections of the image;

[0123] This indicates the output of the crack segmentation branch, which outputs the initial probability response results of natural cracks in the original panoramic unfolded image;

[0124] This indicates the auxiliary branch output of the fracture centerline, which outputs the probability result of the location of the main fracture skeleton;

[0125] Table model parameter set.

[0126] like Figure 2 As shown, Figure 2 This is a schematic diagram of the collaborative identification model for surrounding rock and fractures of the present invention. In terms of model structure, the input tensor is first processed through a shared encoding backbone. Multi-scale feature extraction is performed to obtain shared feature representations suitable for surrounding rock discrimination and fracture identification. Considering that surrounding rock identification relies more on color, texture, and structural features, while natural fracture identification relies more on elongated edges and local continuity features, this invention employs a multi-scale feature extraction method in the shared encoding process, enabling the network to simultaneously perceive local texture details and regional contextual information. Let the... Layer input features are Then its enhanced output features can be expressed as:

[0127]

[0128] in, Used to extract local texture and crack edge features. Used to extract large-scale surrounding rock structure information. This is used to expand the receptive field and enhance the ability to express the direction of fracture extension and regional context. This indicates a feature fusion operation. This represents convolution mapping and nonlinear activation operations.

[0129] Based on shared feature extraction, the surrounding rock identification branch outputs the probability of surrounding rock categories in different borehole depth sections of the image. Let the category probability vector output by the surrounding rock identification branch be... for:

[0130]

[0131] in, This indicates that the current image region belongs to the first... Predicted probability of surrounding rock type This represents the total number of surrounding rock categories. The corresponding surrounding rock identification result is:

[0132]

[0133] in, This label indicates the rock type of the current image area. The rock type can be set according to the application requirements of water conservancy and hydropower projects, including but not limited to granite, diorite, sandstone, siltstone, mudstone, limestone, dolomite, slate, phyllite, schist, gneiss, basalt, and tuff.

[0134] Simultaneously, the fracture segmentation branch is used to output the initial probability response results of natural fractures in the borehole wall image. Let the output of the fracture segmentation branch be... ,but Represents pixels The probability value for a natural fracture. The fracture centerline auxiliary branch outputs the probability result of the location of the main fracture framework, denoted as... ,in Represents pixels The probability value of belonging to the centerline of the fracture. By simultaneously outputting the probability of the fracture region and the probability of the centerline, the completeness and continuity of the fracture structure representation can be improved.

[0135] To enhance the synergy between the fracture region results and the fracture centerline results, this invention further fuses the fracture main branch response results and the fracture centerline auxiliary response results to obtain a fracture enhancement response map:

[0136]

[0137] in, This represents the fracture enhancement response value after fusion. This represents the fusion weighting coefficient. This fracture enhancement response map retains information about the fracture region's extent while highlighting the main fracture framework features, providing a foundation for subsequent candidate fracture target selection.

[0138] The key to step S2 is not directly outputting the final fracture result, but generating two types of basic results required for subsequent processing: first, the surrounding rock identification result, used to construct the surrounding rock zoning prior; and second, the fracture enhancement response result, used as the initial input for natural fracture candidate extraction. In other words, the surrounding rock identification result in step S2 is not an independently terminated output, but an important prerequisite result for fracture constraint identification in subsequent steps.

[0139] To achieve joint optimization of surrounding rock identification, fracture segmentation, and fracture centerline enhancement, this invention constructs a joint loss function for the collaborative identification model of surrounding rock and fractures. Let the total loss be... Then we have:

[0140]

[0141] in, Indicates the loss in the classification of surrounding rock. Indicates the fracture segmentation loss. Indicates auxiliary loss along the fracture centerline. , and This represents the weight coefficient corresponding to each loss term. By using joint loss constraints, the model can be optimized simultaneously on both rock identification and fracture identification tasks, thereby improving the overall identification performance.

[0142] S3, based on the first mapping relationship, map the surrounding rock identification result to the surrounding rock prior information; the surrounding rock prior information includes the surrounding rock category prior and the surrounding rock boundary prior;

[0143] Step S3 is used to automatically identify the surrounding rock type corresponding to the original panoramic unfolded image based on the surrounding rock identification results output in step S2, and establish a correspondence between the surrounding rock identification results and the actual borehole depth range. On this basis, a priori surrounding rock zoning for fracture identification is further constructed. The role of this step is not only to form the surrounding rock stratification results at the whole borehole scale, but also to convert the surrounding rock identification results into prior information that can be used as constraints for subsequent fracture extraction, thereby forming a clear technical coupling relationship between surrounding rock identification and fracture identification.

[0144] In this embodiment, step S3 specifically includes:

[0145] First, based on the first mapping relationship, the surrounding rock identification results are projected onto the actual borehole depth range. Let the first... The starting and ending positions of each analysis segment in the vertical direction of the image are as follows: and Its surrounding rock category label is The actual borehole depth range corresponding to this surrounding rock section is:

[0146]

[0147]

[0148] in, Indicates the first The starting hole depth corresponding to each analysis section Indicates the first The analysis section corresponds to the termination hole depth. Therefore, the surrounding rock identification results can be directly mapped to the surrounding rock category results within the hole depth range.

[0149] In practical processing, the identification results of adjacent analysis sections may fluctuate over short distances due to local texture changes, image noise, or lithological transition zones. To improve the continuity of the surrounding rock stratification results and the rationality of geological interpretation, this invention preferably applies continuity constraints and smoothing processing to the identification results of longitudinally adjacent sections. When adjacent sections have the same surrounding rock category, they can be merged into the same surrounding rock interval; when adjacent sections have different categories but small differences in category probability, and are located in lithological transition zones, boundary adjustments can be made by combining the probability distribution of adjacent sections and longitudinal continuity, thereby reducing the impact of local jumps on the overall borehole lithological interpretation results.

[0150] After the above processing, a set of surrounding rock intervals arranged in order of borehole depth within the entire borehole area can be obtained. , is represented as:

[0151]

[0152] in, Indicates the total number of surrounding rock sections. and They represent the first The starting and ending borehole depths of each surrounding rock section. This indicates the surrounding rock category label corresponding to this interval.

[0153] After obtaining the set of surrounding rock intervals, this invention goes beyond simply outputting the surrounding rock category; it further constructs a priori information on surrounding rock partitions for subsequent fracture identification and screening. Specifically, firstly, a sequence of surrounding rock categories distributed longitudinally along the image is constructed based on the set of surrounding rock intervals. Then, based on this sequence, a surrounding rock category function, i.e., the priori information on surrounding rock categories, is constructed as follows:

[0154]

[0155] in, Indicates the vertical position of the image The corresponding surrounding rock category label. This function is used to characterize the surrounding rock zoning at different longitudinal locations in the image.

[0156] Furthermore, to identify the boundary locations between surrounding rock zones, a surrounding rock boundary indicator function is constructed. Let the boundary location between surrounding rock sections be... Then the a priori representation of the surrounding rock boundary can be expressed as:

[0157]

[0158] in, Indicates vertical position Whether the location is within the boundary of the surrounding rock. Indicates the first The neighborhood range near the boundary of each surrounding rock zone. This prior knowledge of the surrounding rock boundary is used for subsequent identification of pseudo-fracture targets that coincide with the lithological boundary.

[0159] Therefore, the output of step S3 includes not only the lithological range of the surrounding rock within the entire borehole, but also two types of prior information about the surrounding rock required for fracture identification: one is the surrounding rock category function, i.e., the prior information about the surrounding rock category. The first is used to characterize the surrounding rock type corresponding to different borehole depths; the second is the a priori boundary of the surrounding rock. This is used to characterize the boundary location of the surrounding rock partition. Through this step, the surrounding rock identification result is further converted into prior constraint information that can be directly used in the fracture extraction process. Thus, in this invention, surrounding rock identification is no longer an independent result, but an important input condition for subsequent natural fracture identification.

[0160] S4. Based on the initial response results of the fractures, the natural fractures in the original panoramic unfolded image are automatically extracted and candidate targets are screened based on the prior information of the surrounding rock to obtain effective fracture candidate targets.

[0161] Step S4 is used to automatically extract and screen candidate fractures in the original panoramic unfolded image of the borehole wall based on the fracture enhancement response map output in step S2 and the surrounding rock zoning prior constructed in step S3. This step is one of the important steps that distinguishes this invention from the prior art. Its core is that instead of directly extracting fractures based on the fracture response results, it uses the surrounding rock zoning prior to constrain and identify fracture candidate targets, thereby suppressing the interference of surrounding rock boundaries, abrupt changes in lithological texture edges, and other pseudo fractures on the fracture identification results.

[0162] In this embodiment, step S4 specifically includes:

[0163] Let the crack enhancement response diagram obtained in step S2 be... Then, firstly, based on the surrounding rock category function, i.e., the prior knowledge of the surrounding rock category. An adaptive fracture extraction threshold for surrounding rock categories is constructed, and candidate binary fracture maps are obtained based on the relationship between the fracture extraction threshold and the fracture enhancement response map; let the fracture extraction threshold be... Then the candidate binary map of the fracture It can be represented as:

[0164]

[0165] in, This represents the fracture extraction threshold determined by the surrounding rock category at the current location. This method employs different fracture response criteria for different surrounding rock categories, thereby improving the adaptability of fracture extraction to various surrounding rock textures. For surrounding rock sections with complex textures and numerous false edges, the fracture extraction threshold can be increased to suppress false detections; for surrounding rock sections with weak fracture boundaries and low contrast, the threshold can be appropriately decreased to reduce missed detections.

[0166] After obtaining the binary image of the fracture candidates, the connected regions are extracted to form an initial set of fracture candidate targets. Let the extracted... The candidate fracture targets are denoted as Then the set of all initial fracture candidate targets It can be represented as:

[0167]

[0168] in, This represents the total number of initial fracture candidate targets.

[0169] To suppress false fracture targets that coincide with the surrounding rock boundary, this invention further introduces a constraint on the degree of overlap of the surrounding rock boundary. Let candidate targets be... Prior to the boundary with the surrounding rock The overlap ratio is:

[0170]

[0171] in, Indicate candidate target The total number of pixels, This indicates the degree of overlap between the candidate target and the boundary of the surrounding rock zone. When Greater than the preset boundary overlap threshold If this indicates that the candidate target is mainly distributed along the boundary of the surrounding rock zone, it is easily identified as a pseudo-fracture caused by abrupt changes in lithological boundaries or texture, and therefore it should be suppressed or eliminated. That is: if Then It was determined to be a low-confidence candidate target;

[0172] By using this boundary coincidence constraint, the probability of the surrounding rock partition boundary being misidentified as a fracture can be effectively reduced, and further fracture candidate targets can be obtained.

[0173] After suppressing boundary overlap, this invention further performs geometric screening of candidate fracture targets based on the surrounding rock type. Let the candidate targets be... The length is The area is Then its filtering criteria can be written as:

[0174]

[0175]

[0176] in, and These represent the minimum fracture length threshold and the minimum fracture area threshold, respectively, determined by the surrounding rock type. In this way, different geometric screening criteria can be applied to fracture candidate targets within different surrounding rock zones, thereby improving the accuracy of natural fracture candidate extraction.

[0177] For the valid crack candidate targets retained after screening, this invention further extracts their centerline point set to provide input for subsequent trajectory fitting. Let the... The set of centerline points corresponding to each effective fracture candidate target for:

[0178]

[0179] in, Indicates the first on the center line The coordinates of each pixel. Indicates the first The total number of centerline points of each valid fracture candidate target.

[0180] Therefore, step S4 is not essentially about directly extracting fractures from the fracture response map, but rather about automatically extracting natural fractures and screening candidate targets under the prior constraints of surrounding rock zoning. Through adaptive thresholds for surrounding rock categories, suppression of overlapping surrounding rock boundaries, and geometric screening rules related to surrounding rock categories, this invention enables the surrounding rock identification results to play a crucial role in the fracture extraction stage.

[0181] S5. Based on the second mapping relationship and the first mapping relationship, perform trajectory fitting and position parameter determination on the effective fracture candidate targets to obtain the fracture trajectory;

[0182] Step S5 is used to perform trajectory fitting on the centerline point set of each valid natural fracture candidate target obtained in step S4, and to determine the position parameters corresponding to each fracture. The purpose of this step is to further transform the fracture from a regional target in the image into a continuous geometric object that can be used for spatial inversion, while clarifying the fracture's position in the borehole depth and circumferential position of the borehole wall, providing a direct basis for subsequent cylindrical coordinate mapping and fracture attitude calculation.

[0183] In this invention, natural fractures typically appear as continuous curves or approximate curves in the panoramic unfolded view of the borehole wall. Since the set of fracture centerline points may be affected by local noise, variations in the thickness of the segmented edges, and local discontinuities, discrete points cannot be directly used as input for subsequent geometric calculations. Instead, a continuous representation of the fracture needs to be obtained through trajectory fitting.

[0184] Let the first step in step S4 be... The set of centerline points corresponding to each effective fracture candidate target for:

[0185]

[0186] in, Indicates the first on the center line of the fracture The coordinates of each pixel. This represents the total number of points along the centerline of the fracture. Based on this set of centerline points, this invention performs curve fitting on the fracture trajectory to establish a continuous trajectory representation of the fracture in the panoramic unfolded image. Let the first... The fitted trajectory of the crack is recorded for:

[0187]

[0188] in, Indicates the first The vertical position of the crack in the image The corresponding lateral position function.

[0189] Considering that natural fissures often exhibit curved trajectories in the developed cylindrical surface of the hole wall, this invention preferably obtains the parameter expression of this function through least squares fitting. Let the set of fitting parameters be... Then its solution process can be expressed as:

[0190]

[0191] in, Indicates the first The parameters to be determined for the crack trajectory function are aimed at minimizing the sum of squared deviations between the fitted curve and the centerline point set. Through this fitting process, a stable trajectory representation of each crack in the image space can be obtained.

[0192] After obtaining the fracture trajectory, the location parameters of the fracture are further determined. First, based on the distribution of the fracture centerline point set along the longitudinal direction of the image, the representative hole depth of the fracture is determined. Let the... Longitudinal average position of the set of points along the center line of the crack for:

[0193]

[0194] The representative hole depth corresponding to this crack for:

[0195]

[0196] in, This indicates that, based on the first mapping relationship established in step S1, the average longitudinal position... The actual hole depth value obtained, Indicates the first The location of the representative hole depth is indicated by the crack.

[0197] In a preferred embodiment, the initial depth of the fracture can also be determined based on the minimum and maximum longitudinal positions of the fracture centerline point set. and termination hole depth .set up:

[0198]

[0199]

[0200] Based on the initial and final hole depths, the range of hole depths corresponding to the fracture can be expressed as follows:

[0201]

[0202]

[0203] in, and They represent the first The starting and ending depths of the fracture are determined. This method can further reflect the extent of the fracture's extension within the borehole.

[0204] In addition to the hole depth, this invention further determines the circumferential location of the crack on the hole wall. Let the first... The lateral average position of the set of points along the center line of the crack for:

[0205]

[0206] The corresponding circumferential angle of the hole wall is:

[0207]

[0208] in, This indicates that, based on the second mapping relationship established in step S1, the average lateral position... The obtained circumferential angle of the hole wall, Indicates the first The crack represents the circumferential location.

[0209] Through the above processing, step S5 not only obtains the continuous trajectory representation of the fracture, but also simultaneously determines the representative depth position, depth range, and circumferential position of the fracture. Compared with the existing technology that only uses the fracture image region as the output result, this invention further represents the fracture as a geometric object with a continuous trajectory and clear positional parameters, so that the fracture identification result can directly serve the subsequent spatial coordinate mapping and fracture attitude calculation.

[0210] S6 maps the fracture trajectory to the borehole cylindrical coordinate system and combines the borehole attitude parameters to complete the fracture inclination and dip angle inversion, obtaining the fracture parameter results.

[0211] In this embodiment, step S6, mapping the fracture trajectory to the borehole wall cylindrical coordinate system, is used to transform the fracture trajectory obtained in step S5 from the two-dimensional panoramic unfolded image coordinate system to the borehole wall cylindrical coordinate system, thereby establishing the geometric representation of the fracture in the borehole wall space. The purpose of this step is to transform the fracture trajectory in the image from a "planar curve" to a "spatial intersection line on the borehole wall cylindrical surface," providing a geometric basis for subsequent inversion of fracture tendency and dip angle by combining the borehole axis azimuth angle and borehole inclination angle.

[0212] In this invention, the original panoramic unfolded image of the borehole is essentially an unfolded image of the cylindrical surface of the borehole wall. Therefore, the lateral position of any point in the image corresponds to the circumferential angle of the borehole wall, and the longitudinal position corresponds to the depth of the borehole along the borehole axis. Based on the first and second mapping relationships established in step S1, the fracture trajectory can be remapped to the local cylindrical coordinate system of the borehole.

[0213] Let the first step in step S5 be... Trajectory function of a crack for:

[0214]

[0215] in, Indicates the first The vertical position of the crack in the image The corresponding horizontal position. Based on the first mapping relationship in step S1:

[0216]

[0217] And the second mapping relationship:

[0218]

[0219] The borehole depth coordinates and circumferential angular coordinates corresponding to any point on the fracture trajectory can be obtained as follows:

[0220]

[0221]

[0222] in, Indicates the first The borehole depth coordinates of the corresponding points on the fracture trajectory. This represents the circumferential angle of the point on the cylindrical surface of the hole wall.

[0223] Further combining the borehole wall radius This allows the fracture trajectory to be mapped to the local cylindrical coordinate system of the borehole. Let the first... The spatial trajectory of the crack in the local cylindrical coordinate system is denoted as Then we have:

[0224]

[0225] in, Indicates the first The spatial coordinates of a fracture point in the borehole local coordinate system are given by the first and second components, which represent the point's position on the circumference of the borehole wall section, and the third component, which represents the point's position within the borehole along the borehole axis. Using this expression, the two-dimensional trajectory of the fracture in the original panoramic unfolded image is transformed into a three-dimensional spatial curve existing on the cylindrical surface of the borehole wall.

[0226] Therefore, the core of the above steps lies in: based on the expansion relationship of "the horizontal direction of the image corresponds to the circumferential angle, and the vertical direction of the image corresponds to the hole depth", the fracture trajectory function is... Convert to cylindrical surface parametric curve This transformation is not a simple coordinate replacement, but rather restores the image recognition results to a spatial representation consistent with the actual borehole wall geometry, thereby enabling the subsequent solution of the fracture surface to be based on the true borehole wall geometry.

[0227] For each fracture, this step ultimately yields its continuous trajectory representation in the local cylindrical coordinate system of the borehole. Compared to existing technologies that analyze fracture morphology only in a two-dimensional image plane, this invention establishes a mapping relationship between image coordinates and borehole wall cylindrical coordinates, elevating the fracture from an image object to a spatial geometric object, thus providing a clear mathematical basis for fracture attitude inversion.

[0228] In this embodiment, the fracture dip and inclination angle are inverted by combining borehole attitude parameters to obtain fracture parameter results. These results are then used to further invert the spatial attitude parameters of each natural fracture, based on the obtained cylindrical coordinate trajectory of the fracture borehole wall, and by combining the borehole axis azimuth and borehole inclination angle, to obtain the fracture dip and inclination angle. The purpose of this step is to further enhance the aforementioned image recognition results and borehole wall geometric representation into fracture attitude parameters that can be directly used in engineering geology.

[0229] In existing technologies, the acquisition of fracture orientation in borehole television images typically relies on manual interpretation or simple empirical estimation methods based on unfolded diagram morphology, which struggles to balance automation and spatial geometric rigor. This invention, building upon the automatic identification, trajectory fitting, and cylindrical coordinate mapping of fractures, further introduces borehole attitude parameters to construct a transformation relationship from the borehole local coordinate system to the engineering spatial coordinate system. Based on this, the spatial normal vector of the fracture surface is calculated, ultimately yielding the fracture's dip and dip angle.

[0230] In this embodiment, the fracture dip and inclination angle are inverted by combining the drilling attitude parameters to obtain the fracture parameter results, specifically including:

[0231] First, based on the borehole axis azimuth angle obtained in step S1 and the bevel angle Establish the unit direction vector of the borehole axis in the engineering space coordinate system:

[0232]

[0233] in, This represents the borehole axis direction vector. This vector reflects the directional relationship between the borehole local coordinate system and the engineering space coordinate system, and is the basis for subsequent spatial inversion.

[0234] For the first step in the above steps Fracture trajectory in the local cylindrical coordinate system of the borehole :

[0235]

[0236] in, This represents the circumferential angle corresponding to the point on the fracture trajectory. This represents the coordinates of the corresponding hole depth. Since this trajectory is the spatial intersection of the fracture surface and the cylindrical surface of the borehole wall, the normal vector of the fracture surface can be inverted based on the tangential characteristics of this intersection and the local geometric relationship of the borehole wall.

[0237] Specifically, regarding the fracture trajectory Taking the derivative, we can obtain the tangent vector of the crack intersection line:

[0238]

[0239] in, Indicates the first Strip cracks in parameters The tangent vector at that point. Meanwhile, the radial normal vector of the cylindrical surface of the hole wall at that point can be expressed as:

[0240]

[0241] in, This represents the local radial normal vector of the cylindrical surface of the hole wall at that point. Since the crack trajectory is the intersection of the crack surface and the cylindrical surface of the hole wall, the normal vector of the crack surface at that point can be determined by the tangent vector of the intersection line and the radial vector of the hole wall. Correspondingly, the first... Local normal vector estimate of the fracture in the borehole local coordinate system It can be written as:

[0242]

[0243] in, Represents the cross product of vectors. This represents the estimated value of the normal vector of the fracture surface in the local coordinate system of the borehole.

[0244] To reduce the impact of local noise, trajectory fitting errors, and anomalies at individual sampling points on attitude calculation, this invention preferably selects multiple sampling points along the same fracture trajectory, normalizes the local normal vectors at each sampling point, and averages them to obtain the first... The global normal vector of the fracture in the borehole local coordinate system:

[0245]

[0246] in, Indicates the number of sampling points. Indicates the first The average normal vector of the fracture in the borehole local coordinate system. To prevent extremely small positive numbers with a denominator of zero.

[0247] After obtaining the global normal vector of the fracture, it is further transformed into the engineering space coordinate system by combining the borehole attitude parameters. Let the rotation matrix from the borehole local coordinate system to the engineering space coordinate system be... Then the first The normal vector of the crack in the engineering space coordinate system can be expressed as:

[0248]

[0249] in, Indicates the first The normal vector of the crack in the engineering space coordinate system. The rotation matrix. From the borehole axis azimuth and the bevel angle The essence of construction lies in describing the directional transformation relationship between the local coordinate system of the borehole and the engineering space coordinate system.

[0250] set up:

[0251]

[0252] The dip angle and dip direction of the fracture can then be calculated based on this normal vector.

[0253] The fracture dip angle is defined as the angle between the fracture surface and the horizontal plane, therefore the first... dip angle of the crack It can be represented as:

[0254]

[0255] in, Indicates the first The dip angle of the fracture. When expressed in degrees, it can be multiplied by... This formula reflects the geometric relationship between the normal vector of the fracture surface and the vertical direction, thus determining the degree of inclination of the fracture surface relative to the horizontal plane.

[0256] The fracture dip direction is defined as the azimuth angle of the maximum downward dip direction of the fracture surface on the horizontal plane. Based on the projection direction of the normal vector on the horizontal plane, the... Tendency of cracks It can be represented as:

[0257]

[0258] in, This represents the four-quadrant arctangent function, used to ensure that the calculation result lies in the correct quadrant. For ease of engineering application, it is further converted to... From the perspective of angles within the range, we can obtain:

[0259]

[0260] in, Indicates the first The dip angle of the fracture, in degrees.

[0261] Through the above processing, this step realizes the conversion of fracture image recognition results into engineering geological attitude parameters. Compared with the existing technology that relies on manual estimation or empirical formula approximation, this invention achieves automated calculation of fracture dip and dip angle through automatic fracture identification, trajectory fitting, cylindrical coordinate mapping, normal vector inversion, and attitude calculation. Moreover, the entire process has a clear spatial geometric basis and can better meet the data processing needs of horizontal borehole television images in water conservancy and hydropower projects.

[0262] As a further implementation, the method further includes:

[0263] S7 summarizes and organizes the surrounding rock identification results and fracture parameter results into a structured result set, and outputs the structured result set.

[0264] Step S7 is used to summarize, organize, and output the surrounding rock identification results obtained in step S2 and the fracture parameter results obtained in step S6, forming a single-hole borehole panoramic imaging data processing result for engineering applications. The purpose of this step is to integrate the identification and calculation results obtained in the preceding steps into a structured, interpretable, and directly applicable final output, thereby completing the complete processing loop of the method described in this invention.

[0265] In this invention, the output of step S7 mainly includes two parts: first, the identification results of the surrounding rock lithology corresponding to each depth interval within the entire borehole range; second, the location parameters and attitude parameters corresponding to each natural fracture. The former originates from the set of surrounding rock intervals in step S2, and the latter originates from the set of fracture parameters in step 6. After unified integration, a complete expression of the geological information of the borehole wall of a single borehole can be formed.

[0266] For the surrounding rock lithology results, let the set of surrounding rock intervals output in step S2 be:

[0267]

[0268] in, and They represent the first The starting and ending borehole depths of each surrounding rock section. This indicates the surrounding rock category label corresponding to this interval. This represents the total number of surrounding rock sections. This result reflects the stratified distribution of the surrounding rock along the borehole depth throughout the entire borehole.

[0269] Regarding the fracture parameter results, let the set of fracture parameters output in step S6 be... for:

[0270]

[0271] in, Indicates the first The location of the represented hole depth by the crack. It indicates that it represents the circumferential position. Indicate their inclination, Indicates its inclination angle, This represents the total number of effective natural fractures.

[0272] To facilitate results management and engineering applications, this invention preferably assigns a unique number to each crack, forming a crack result record set. Let the first crack be... The numbering of the cracks is The crack output result can then be further expressed as:

[0273]

[0274] in, Indicates the first Each natural fracture is numbered. This method ensures that each fracture is clearly identified during subsequent results presentation, statistical analysis, and manual verification.

[0275] In terms of output format, this invention preferably generates a structured result set, unifying the surrounding rock lithology results and fracture parameter results into a single-well processing result. Correspondingly, the final structured result set of this invention can be represented as:

[0276]

[0277] in, This represents the final structured result set of single-hole borehole panoramic imaging data processing. Indicates the results of the surrounding rock lithology interval. This represents the results of natural fracture parameters.

[0278] When necessary, the output can be further presented in the form of image overlay, table summary, or whole-hole geological interpretation results map, so as to directly serve the surrounding rock evaluation, structural surface analysis, geological logging, and subsequent engineering decision-making in water conservancy and hydropower projects.

[0279] Example 2

[0280] like Figure 3 As shown, Figure 3 This is a block diagram of the borehole unfolding image processing system based on deep learning and spatial inversion of the present invention. The difference between this embodiment and Embodiment 1 is that this embodiment provides a borehole unfolding image processing system based on deep learning and spatial inversion, which corresponds one-to-one with the borehole unfolding image processing method based on deep learning and spatial inversion in Embodiment 1; the system includes:

[0281] The acquisition unit is used to acquire the original panoramic unfolded image of the borehole and its corresponding basic parameter information, and to establish a first mapping relationship between the vertical pixels of the image and the borehole depth, and a second mapping relationship between the horizontal pixels of the image and the circumferential angle of the borehole wall.

[0282] The collaborative identification unit is used to extract features from the original panoramic unfolded image based on a pre-constructed collaborative identification model of surrounding rock and fractures, and obtain the surrounding rock identification results and the initial response results of fractures; the initial response results of fractures include the main branch response results of fractures and the auxiliary response results of fracture centerline;

[0283] The surrounding rock prior information unit is used to map the surrounding rock identification result into surrounding rock prior information according to the first mapping relationship; the surrounding rock prior information includes surrounding rock category prior and surrounding rock boundary prior;

[0284] The fracture candidate target screening unit is used to automatically extract and screen natural fractures in the original panoramic unfolded image based on the fracture initial response result and the prior information of the surrounding rock, so as to obtain effective fracture candidate targets.

[0285] The fracture trajectory determination unit is used to perform trajectory fitting and position parameter determination on the effective fracture candidate targets according to the second mapping relationship and the first mapping relationship to obtain the fracture trajectory;

[0286] The inversion calculation unit is used to map the fracture trajectory to the borehole wall cylindrical coordinate system, and combine the borehole attitude parameters to complete the inversion of fracture inclination and dip angle, and obtain fracture parameter results.

[0287] The execution process of each unit can be carried out according to the process flow of the borehole unfolding diagram processing method based on deep learning and spatial inversion in Example 1, and will not be described in detail in this example.

[0288] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0289] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0290] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0291] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0292] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A borehole unfolding image processing method based on deep learning and spatial inversion, characterized in that, The method includes: Obtain the original panoramic unfolded image of the borehole and its corresponding basic parameter information, and establish the first mapping relationship between the vertical pixels of the image and the borehole depth, and the second mapping relationship between the horizontal pixels of the image and the circumferential angle of the borehole wall; Based on a pre-constructed collaborative identification model for surrounding rock and fractures, features are extracted and identified from the original panoramic unfolded image to obtain surrounding rock identification results and initial fracture response results; the initial fracture response results include fracture main branch response results and fracture centerline auxiliary response results. Based on the first mapping relationship, the surrounding rock identification result is mapped to surrounding rock prior information; the surrounding rock prior information includes surrounding rock category prior and surrounding rock boundary prior; Based on the initial response results of the fracture, the natural fractures in the original panoramic unfolded image are automatically extracted and candidate targets are screened based on the prior information of the surrounding rock to obtain effective fracture candidate targets. Based on the second mapping relationship and the first mapping relationship, trajectory fitting and position parameter determination are performed on the effective fracture candidate targets to obtain the fracture trajectory; the fracture trajectory is mapped to the borehole wall cylindrical coordinate system, and fracture inclination and dip angle inversion are completed in combination with the drilling attitude parameters to obtain fracture parameter results.

2. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, The basic parameter information includes borehole depth, borehole wall circumferential angle, borehole wall radius, borehole axis azimuth angle, and borehole inclination angle.

3. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, The method also includes: The original panoramic unfolded image is preprocessed and standardized to obtain a standardized input tensor; The preprocessing and normalization include: brightness equalization, local contrast enhancement and numerical normalization, as well as circumferential position prior.

4. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, The surrounding rock and fracture collaborative identification model extracts surrounding rock texture features and fracture structure features within the same model framework. It performs multi-scale feature extraction on the original panoramic unfolded image through a shared coding backbone to obtain the same shared feature expression suitable for surrounding rock discrimination and fracture identification. The expression for the collaborative identification model of surrounding rock and fractures is: ； in, This indicates the output of the surrounding rock identification branch, which outputs the probability of the surrounding rock category and the corresponding surrounding rock identification result in different borehole depth sections of the image; This indicates the output of the crack segmentation branch, which outputs the initial probability response results of natural cracks in the original panoramic unfolded image; This indicates the auxiliary branch output of the fracture centerline, which outputs the probability result of the location of the main fracture skeleton; Table model parameter set.

5. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, Based on the first mapping relationship, the surrounding rock identification result is mapped to the surrounding rock prior information; The prior information of the surrounding rock includes prior information on the surrounding rock category and prior information on the surrounding rock boundary, including: Based on the first mapping relationship, the surrounding rock identification result is directly mapped to the surrounding rock category result within the borehole depth range; The surrounding rock category results of longitudinally adjacent sections are subjected to continuity constraints and smoothing to obtain a set of surrounding rock intervals arranged in order of borehole depth within the entire borehole range; Based on the set of surrounding rock intervals, a sequence of surrounding rock categories distributed along the vertical direction of the image is constructed, and based on the sequence of surrounding rock categories, a surrounding rock category function is constructed and used as a priori for surrounding rock categories; Based on the boundary locations between the surrounding rock sections, a priori boundaries for the surrounding rock are constructed.

6. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, Based on the initial response results of the fractures, natural fractures in the original panoramic unfolded image are automatically extracted and candidate targets are screened based on the prior information of the surrounding rock to obtain effective fracture candidate targets, including: Based on the prior knowledge of the surrounding rock category, an adaptive fracture extraction threshold for the surrounding rock category is constructed; and according to the magnitude relationship between the fracture extraction threshold and the fracture enhancement response map, a candidate binary map of fractures is obtained; the fracture enhancement response map is obtained by fusing the fracture main branch response result and the fracture centerline auxiliary response result; The connected regions of the binary map of the fracture candidates are extracted to form an initial set of fracture candidate targets; A constraint on the overlap of the surrounding rock boundary is introduced, and the overlap ratio is calculated based on the candidate targets in the initial set of candidate fracture targets and the prior knowledge of the surrounding rock boundary. Candidate targets with an overlap ratio greater than a preset boundary overlap threshold are identified as low-confidence candidate targets, and these low-confidence candidate targets are suppressed or eliminated to obtain further crack candidate targets. Based on the prior knowledge of the surrounding rock category, the further fracture candidate targets are geometrically screened to obtain effective fracture candidate targets and their corresponding fracture centerline point sets.

7. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, Based on the second mapping relationship and the first mapping relationship, trajectory fitting and position parameter determination are performed on the effective fracture candidate targets to obtain the fracture trajectory, including: Based on the set of fracture centerline points of the effective fracture candidate targets, the fracture trajectory is curve-fitted to establish the fracture trajectory in the original panoramic unfolded image. Based on the distribution of the crack centerline point set along the vertical axis of the image, the representative hole depth position of the crack is determined; based on the representative hole depth position and the first mapping relationship, the corresponding representative hole depth of the crack is determined. The starting and ending depths of the fracture are determined based on the minimum and maximum longitudinal positions of the fracture centerline point set, and the corresponding depth range of the fracture is determined based on the starting and ending depths. The circumferential angle of the borehole wall is determined based on the average lateral position of the set of points along the center line of the fracture and the second mapping relationship.

8. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, By combining borehole attitude parameters, fracture dip and inclination inversion is performed to obtain fracture parameter results, including: By introducing drilling attitude parameters, a transformation relationship between the local drilling coordinate system and the engineering space coordinate system is constructed. Based on the transformation relationship, solve for the spatial normal vector of the fracture surface; Based on the spatial normal vector, the dip direction and dip angle of the fracture are calculated as fracture parameter results.

9. The borehole unfolding image processing method based on deep learning and spatial inversion according to claim 1, characterized in that, The method also includes: The surrounding rock identification results and the fracture parameter results are summarized and organized into a structured result set, and the structured result set is output.

10. A borehole unfolding image processing system based on deep learning and spatial inversion, characterized in that, The system includes: The acquisition unit is used to acquire the original panoramic unfolded image of the borehole and its corresponding basic parameter information, and to establish a first mapping relationship between the vertical pixels of the image and the borehole depth, and a second mapping relationship between the horizontal pixels of the image and the circumferential angle of the borehole wall. A collaborative identification unit is used to extract features from the original panoramic unfolded image based on a pre-constructed collaborative identification model of surrounding rock and fractures to obtain surrounding rock identification results and fracture initial response results; the fracture initial response results include fracture main branch response results and fracture centerline auxiliary response results; The surrounding rock prior information unit is used to map the surrounding rock identification result into surrounding rock prior information according to the first mapping relationship; the surrounding rock prior information includes surrounding rock category prior and surrounding rock boundary prior; The fracture candidate target screening unit is used to automatically extract and screen natural fractures in the original panoramic unfolded image based on the fracture initial response result and the prior information of the surrounding rock, so as to obtain effective fracture candidate targets. The fracture trajectory determination unit is used to perform trajectory fitting and position parameter determination on the effective fracture candidate targets according to the second mapping relationship and the first mapping relationship to obtain the fracture trajectory; The inversion calculation unit is used to map the fracture trajectory to the borehole wall cylindrical coordinate system, and combine the borehole attitude parameters to complete the inversion of fracture inclination and dip angle, and obtain fracture parameter results.

Images