Drill hole image restoration method based on generative adversarial network

Abstract

The invention relates to a borehole image restoration method and system, and provides a multi-source data collaborative restoration scheme fusing an on-site rock core image and a rock core scanning image aiming at the problems of blurring, local loss and geological feature distortion in a borehole camera image. According to the method, a deep convolutional network (CNN) generator and a PatchGAN discriminator are constructed based on a generative adversarial network (GAN), and geologic features of an on-site rock core image (a directly acquired rock core surface image) and a rock core scanning image (a high-precision digital scanning image) are matched, so that the rock core image and the rock core scanning image can be effectively distinguished even if a large-area imaging blurring or peeling area exists in a drill hole. By means of the method, the fracture structure can still be repaired through efficient fusion of multi-source core data, meanwhile, the single-hole image processing efficiency is improved, dependence on artificial experience is greatly reduced, and high-precision visual data support is provided for engineering geological analysis and resource exploration.

Description

Technical Field

[0001] The present invention belongs to the field of geological exploration, and specifically is a drilling image restoration method based on a generative adversarial network and on-site rock cores. Background Art

[0002] In geotechnical engineering, geological engineering, hydropower engineering, oil development and geological disaster prevention engineering, it is often necessary to understand the stability characteristics of rock mass structure. The structural surfaces such as joints, faults, weak surfaces and layers in the rock mass are important factors that determine the stability of the actual project. How to identify and analyze the various structural surfaces in the rock mass is an important topic.

[0003] Borehole imaging technology is a technical method that uses optical principles and related equipment to obtain images of the borehole wall in the borehole. By using borehole imaging technology, engineering geological information in the borehole can be accurately and quickly obtained. Borehole imaging equipment is usually composed of borehole camera probes, controllers and other components. The borehole camera probe includes light sources, cameras and other components. When the borehole camera equipment is working, the light source illuminates the borehole wall and the camera captures the image of the borehole wall. However, the use of drilling equipment often relies on the experience of engineering personnel and is very subjective, which leads to the following problems when workers collect borehole images:

[0004] (1) Noise interference: During the drilling process, due to factors such as mud adhesion and rock debris interference, the acquired images contain varying degrees of noise, which affects the image clarity and detail, and brings difficulties to subsequent image analysis and processing.

[0005] (2) Uneven illumination: When conducting borehole photography, the probe must always be in the center of the hole to ensure uniform illumination around the hole wall. However, due to the constraints of on-site conditions, it is often difficult to keep the probe in the center, and the illumination conditions of the hole wall are difficult to control, which can easily lead to uneven illumination, causing some areas in the image to be too bright or too dark, affecting the image contrast and visual effects, and further affecting the identification of features such as rock mass structural surfaces.

[0006] (3) Image blur: Affected by factors such as the probe lowering speed and probe eccentricity, the borehole image may be blurred to a certain extent, making the detailed features of the rock mass unclear, which is not conducive to the accurate judgment of the rock mass integrity and structural surface characteristics.

[0007] In recent years, in response to the above problems, the industry has proposed some image restoration methods based on deep learning, such as defogging algorithms, noise removal algorithms, and Filtersim algorithms. These algorithms mainly generate unknown areas based on known areas. Although they can well fill in the missing parts of the image and improve the quality of the image, they usually produce artificial discontinuities. Furthermore, the existing panoramic borehole camera technology has relatively high requirements for the conditions inside the hole. If the hole is not cleaned during the detection process, or mud appears during the detection process inside the hole, this will cause the image quality of the borehole camera to be poor, so it is impossible to judge the geological conditions of the section with poor image quality. At the same time, the quality of borehole camera imaging at this stage depends on the engineering experience and proficiency of the technicians. If the engineering personnel lack experience in use, the image quality will be poor, and it will be difficult to identify the structural surface and other features in the hole, which further leads to the difficulty of borehole image restoration.

[0008] In summary, there is a lack of research on borehole image restoration. Based on the above background, the present invention proposes a borehole image restoration method, which combines the on-site core image and the core scan image to repair the poor quality borehole image and improve the efficiency of borehole photography. Summary of the invention

[0009] The present invention relates to a method for repairing a borehole image, and the purpose of the present invention is to solve the problem of poor quality of borehole images. The present invention provides a method for repairing borehole images based on deep learning, which repairs poor quality borehole images by combining on-site core images with core scanned images, thereby improving the efficiency of borehole photography.

[0010] The present invention relates to a drilling image repair method, comprising the following steps:

[0011] (1) Core image collection: various types of core image data are collected at the drilling site based on the size, shape, color, bedding, and fracture characteristics of the core;

[0012] (2) Collection of borehole images: collecting borehole images based on the size, shape, color, bedding and crack characteristics of the borehole wall, and matching the collected borehole images with the core images in step (1);

[0013] (3) Core scanning image collection: Use the laboratory core scanning device to scan the cores collected at the drilling site to obtain core scanning images;

[0014] (4) Processing the core image, the borehole image, and the core scan image, cropping the images into a standard format of 512 pixels × 512 pixels, and performing super-resolution processing on the images with lower resolution to obtain high-resolution images;

[0015] (5) Feature extraction and matching: Using feature extraction algorithms such as SIFT, SURF, HOG, ORB, Harris corner detection or Canny edge detection, feature points are extracted from the borehole image, the core image and the core scan image; the feature points include key points, edges and texture information of the image; the extracted feature points are matched to determine the corresponding relationship between the borehole image, the core image and the core scan image; then, the improved RANSAC algorithm is used to purify the matched feature points and remove the feature points that are incorrectly matched;

[0016] (6) Image registration: Calculate the geometric transformation parameters between the borehole image, the core image and the core scan image based on the matched feature points, wherein the geometric transformation parameters include translation, rotation and scaling; then register the borehole image, the core image and the core scan image using the geometric transformation parameters so that the three are aligned in space; and obtain corresponding data set images;

[0017] (7) Image annotation: The images in the above datasets are labeled using the image annotation tool LABELME, and polygonal labels are used to mark the features of bedding, fractures, and key points;

[0018] (8) Image preprocessing: preprocessing the core image and the core scan image, including image cropping, scaling, grayscale or normalization operations to meet the requirements of subsequent processing;

[0019] (9) constructing a generative adversarial network model: the generative adversarial network includes a generator and a discriminator: the generator is configured to generate a repaired borehole image according to the core image and the core scan image, and the discriminator is configured to determine whether the image generated by the generator is real;

[0020] (10) Fix the generator and train the discriminator: The real drilling image and the repaired drilling image generated by the generator are input into the discriminator respectively. The discriminator will output a probability value, which represents the probability that the input image is real. Then, the loss function of the discriminator is calculated based on the output result of the discriminator and the real label. Then, the optimizer is used to back-propagate the loss function to update the parameters of the discriminator to improve the discriminant ability of the discriminator.

[0021] (11) Fixing the discriminator and training the generator: inputting the core image and the core scan image into the generator, and inputting the repaired borehole image generated by the generator into the discriminator. The discriminator will output a probability value, which represents the probability that the input image is real; then, according to the output result of the discriminator and the real label, the loss function of the generator is calculated; then, the optimizer is used to back-propagate the loss function and update the parameters of the generator to improve the generation ability of the generator and make the generated image more realistic;

[0022] (12) Alternating training of the discriminator and the generator: During the training process, the discriminator is trained first, then the generator is trained, and this process is repeated. In each training cycle, the discriminator is trained N times first, and then the generator is trained M times to maintain a balance between the two until the two models converge. Where N and M are both positive integers, NM ≥ 0, and M and N are positive integers greater than or equal to 1 respectively.

[0023] (13) When the loss function tends to a stable state, the generator and the discriminator are fixed in turn;

[0024] (14) The borehole image to be repaired and the core scan image corresponding to the borehole image to be repaired are respectively input into the generator to obtain the repaired borehole image and the repaired core scan image respectively. The repaired borehole image and the repaired core scan image are processed and calculated using a similarity function. When the calculated similarity is greater than a set threshold, the repaired borehole image is the final repaired borehole image.

[0025] Furthermore, during the training process, the learning rate of the discriminator was configured to be 0.0003-0.0006, and the learning rate of the generator was configured to be 0.00005-0.0003.

[0026] Furthermore, the learning rate of the discriminator is configured as 0.0004, and the learning rate of the generator is configured as 0.0001.

[0027] Furthermore, the images generated by the generator are monitored by comparing them with real images. When the discriminator loss function is close to 0, the network is adjusted, and the adjustment includes parameter adjustment, replacement of the loss function or adjustment of the network structure.

[0028] Furthermore, before the similarity function calculates the similarity of the repaired borehole image and the repaired core image, the eigenvalues ​​and eigenvectors of the repaired borehole image and the repaired core image are extracted, and the similarity of the repaired borehole image and the repaired core image is calculated based on the eigenvalues ​​and eigenvectors.

[0029] Furthermore, the step (8) further includes performing image processing by using data enhancement technology to increase the diversity of training data; the image processing includes rotation, flipping and translation.

[0030] Furthermore, the generator selects a convolutional neural network (CNN) as the basic architecture, and the discriminator selects PatchGAN.

[0031] Furthermore, according to the discriminator loss function and the generator loss function in the previous training cycle, the number of training times of the discriminator and the generator in the next training cycle is adjusted according to the adjustment rules.

[0032] Further, the adjustment rule is configured as follows: after each training cycle, a step of determining whether the discriminator loss function is greater than a preset value;

[0033] 1) When the discriminator loss function is greater than the preset value, in the next training cycle, the number of discriminator training times is set to N times, and the number of generator training times is set to M+K times;

[0034] 2) When the discriminator loss function is less than the preset value, in the next training cycle, the number of discriminator training times is set to N+K times, and the number of generator training times is set to M times;

[0035] Where K is the training balance coefficient, K = M × [N / M].

[0036] Furthermore, the initial value of the discriminator training times N is set to N=10 times, and the initial value of the generator training times M is set to M=3 times.

[0037] The technical effects that can be achieved by the present invention are:

[0038] The present invention is based on a generative adversarial network (GAN) network, using on-site core images and core scan images, matching the geological features of on-site core images (directly collected core surface images) with core scan images (high-precision digital scan images), and efficiently integrating multi-source core data to repair the fracture structure, while improving the efficiency of single-hole image processing, greatly reducing dependence on manual experience, and providing high-precision visual data support for engineering geological analysis and resource exploration. Even if part of the area in the hole cannot be seen clearly, the present invention can restore the borehole image through on-site core images and core scan images, reducing the requirements for technicians while improving the accuracy of the borehole image. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0040] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which this application can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by this application, should still fall within the scope of the technical contents disclosed in this application.

[0041] Figure 1 This is the network structure diagram of the CNN generator of the present invention.

[0042] Figure 2 This is the PatchGAN discriminator network structure diagram of the present invention.

[0043] Figure 3 It is the overall network structure diagram of the present invention.

[0044] Figure 4 This is the network structure testing process of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the embodiments in the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0046] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] The applicant of the present invention has found that in order to understand the spatial morphology and distribution of the structural surface in the underground hole, in addition to the borehole image, it can also be obtained through the on-site core map. Therefore, in order to solve the problems of blur, local missing and distortion of geological features in the borehole camera image, the present invention proposes a multi-source data collaborative repair scheme that integrates the on-site core image and the core scan image. The method constructs a deep convolutional network (CNN) generator to extract the cross-scale lithology texture and fracture characteristics of the borehole image to achieve detail reconstruction; designs a discriminator based on PatchGAN to constrain the fine-grained texture consistency of the repair area through the local image block discrimination mechanism; at the same time, establishes a feature alignment module for the on-site core image (directly collected core surface image) and the core scan image (high-precision digital scan image), encodes the geological parameters of the two into a spatial mapping relationship, and realizes the matching of geological features in the imaging missing area through adaptive weighted fusion. Based on the generative adversarial network (GAN), the present invention dynamically injects geological prior knowledge such as lithology classification and fracture density during the repair process to ensure the continuity of the repair result and the real stratigraphic characteristics. Its significant advantage is that even if there are large areas of blurred imaging or peeling areas in the borehole, the fracture structure can still be repaired through the efficient fusion of multi-source core data. At the same time, the efficiency of single-hole image processing is improved, which greatly reduces the dependence on manual experience, improves the visualization and accuracy of borehole images, and provides high-precision visual data support for engineering geological analysis and resource exploration.

[0048] Specifically, the present invention provides a method for repairing a borehole image based on a generative adversarial network (GAN) and a field core, which specifically includes the following steps: Figure 4 As shown,

[0049] (1) Core image collection: First, according to the characteristics of the core, collect as much image data of various cores as possible at the drilling site. In order to better match the core images with the borehole images, collect the borehole images and virtual core images of the corresponding projects done previously. The above images should cover various possible situations, such as the size, shape, color, bedding and cracks of the core, to ensure the diversity and comprehensiveness of the data.

[0050] (2) Borehole image collection: First, based on the characteristics of the borehole wall, collect as many borehole images as possible. The collected borehole images should correspond to the core images in (1).

[0051] (3) Core scanning image collection: Collect cores on site and bring them back to the laboratory to obtain core scanning images using a core scanning device.

[0052] (4) The images of the core dataset, borehole image dataset, and core scan dataset are processed and uniformly cropped into a standard format of 512 pixels × 512 pixels. For some images with lower resolution, super-resolution processing is required to obtain high-resolution images.

[0053] (5) Feature extraction and matching: Use feature extraction algorithms such as SIFT, SURF, HOG, ORB, Harris corner detection, Canny edge detection, etc. to extract feature points from borehole images, core images, and core scan images. These feature points can include information such as key points, edges, and textures in the image; match the extracted feature points to determine the correspondence between the borehole image, core image, and core scan image. Use the improved RANSAC (Random Sample Consensus) algorithm to purify the matched feature points, remove incorrectly matched feature points, and improve the accuracy and reliability of the matching.

[0054] (6) Image registration: Based on the matched feature points, the geometric transformation parameters between the borehole image, core image and core scan image are calculated, such as translation, rotation or scaling. Then, these parameters are used to register the borehole image, core image and core scan image so that the two are aligned in space.

[0055] (7) Image annotation: The images of the above datasets are labeled using the image annotation tool LABELME, and polygonal labels are used to mark the features of bedding, fractures, and key points.

[0056] (8) Image preprocessing: Preprocess the core images and core scan images. Preprocessing includes operations such as cropping, scaling, grayscale or normalization of the images to meet the requirements of subsequent processing. At the same time, data enhancement techniques such as rotation, flipping or translation can be used to increase the diversity of training data.

[0057] (9) Model construction: The generative adversarial network consists of a generator and a discriminator.

[0058]

[0059] Among them, x is the real sample, y is the conditional information, z is the noise data, P data (x) is the real data distribution; P z (z) is the prior noise distribution; G(*) is the generator model; D(*) is the discriminator model; E x～pdata(x) [logD(x)] represents the expectation of the real data, E z～pz(z) [log(1-D(G(z)))] represents the expectation of generated data; V(D,G) is the value function;

[0060] The task of the generator is to generate a repaired borehole image based on the core image and the core scan image, and the task of the discriminator is to determine whether the generated image is real. The generator selects a convolutional neural network (CNN) as the basic architecture. The CNN workflow diagram is as follows: Figure 1 As shown; the discriminator selects PatchGAN, and the workflow diagram of the PatchGAN discriminator is as follows Figure 2 shown.

[0061] (10) Fix the generator and train the discriminator: Figure 3 As shown in the figure, the real drilling image and the fake drilling image generated by the generator are input into the discriminator respectively, and the discriminator outputs a probability value, indicating the probability that the input image is real. Then, the loss function of the discriminator is calculated based on the output result of the discriminator and the real label. Then, the optimizer is used to back-propagate the parameters of the discriminator according to the loss function to improve the discriminant ability of the discriminator.

[0062] (11) Fix the discriminator and train the generator: Input the core image and the core scan image into the generator. The generator generates a fake drilling image and inputs it into the discriminator. The discriminator outputs a probability value, indicating the probability that the input image is real. Then, the loss function of the generator is calculated based on the output of the discriminator and the true label. Next, the optimizer is used to back-propagate the loss function to update the parameters of the generator to improve the generator's generation ability and make the generated image more realistic.

[0063] (12) The discriminator and generator are trained alternately: Figure 3 As shown in the figure, during the training process, the discriminator is usually trained first, and then the generator is trained, and this is repeated. In each training cycle, the discriminator can be trained several times first, and then the generator can be trained several times to maintain a balance between the two until the two models converge.

[0064] (13) When the loss function tends to a stable state, the generator and the discriminator are fixed respectively;

[0065] (14) The borehole image to be repaired and the core image corresponding to the borehole image to be repaired are respectively input into the generator to obtain the repaired borehole image and the repaired core image respectively. The repaired borehole image and the repaired core image are compared using a similarity function. When the similarity is greater than a set threshold, the repaired borehole image is the final repaired borehole image.

[0066] (15) Furthermore, during the training process, the learning rate can be adjusted appropriately according to the training situation to accelerate the convergence of the model; wherein the learning rate of the discriminator is preferably 0.0003-0.0006, and the learning rate of the generator is preferably 0.00005-0.0003; further, the learning rate of the discriminator is preferably 0.0004, and the learning rate of the generator is preferably 0.0001. By setting the learning efficiency of the discriminator to be greater than or equal to the learning efficiency of the generator, the learning pace of the discriminator is lower than that of the generator, which improves the stability of the model and reduces the probability of model collapse.

[0067] (16) Furthermore, the images generated by the generator are monitored by comparing them with real images. When the discriminator loss function is close to 0, the network is adjusted. The adjustment includes parameter adjustment, replacement of the loss function and adjustment of the network structure.

[0068] (17) Furthermore, noise is added to both the actual and generated images before feeding them into the discriminator to achieve better training results.

[0069] (18) Furthermore, before the similarity function calculates the similarity between the repaired borehole image and the repaired core image, the eigenvalues ​​and eigenvectors of the repaired borehole image and the repaired core image are extracted, and the similarity between the repaired borehole image and the repaired core image is calculated based on the eigenvalues ​​and eigenvectors.

[0070] (19) According to the discriminator loss function and the generator loss function in the previous training cycle, the number of training times of the discriminator and the generator in the next training cycle is adjusted according to the adjustment rule. The adjustment rule is configured as: after each training cycle, a step of judging whether the discriminator loss function is greater than a preset value; 1) when the discriminator loss function is greater than the preset value, in the next training cycle, the number of discriminator training times is set to N times, and the number of training times of the training generator is set to M+K times; 2) when the discriminator loss function is less than the preset value, in the next training cycle, the number of discriminator training times is set to N+K times, and the number of training times of the training generator is set to M times; wherein K is a training balance coefficient, K=M×[N / M], wherein [N / M] represents rounding calculation. The initial value of the number of discriminator training times N can be set to N=5, 10, 20, 50, 100 or 500 times, and the initial value of the number of generator training times M is set to M=3, 7, 15, 30, 70, 120 or 240 times, and N is greater than or equal to M. By setting the training balance coefficient, the discriminator and the generator can form an effective competition and reduce the probability of mode collapse. By dynamically adjusting the number of model training times, the learning efficiency and accuracy of the model are improved.

[0071] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0072] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

[0073] In this specification, each embodiment is described in a progressive, parallel, or progressive and parallel manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0074] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that an article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such article or device. In the absence of further restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the article or device including the above elements.

[0075] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A drilling image repair method, characterized in that The following steps are involved: (1) Core image collection: various types of core image data are collected at the drilling site based on the size, shape, color, bedding, and fracture characteristics of the core; (2) Collection of borehole images: collecting borehole images based on the size, shape, color, bedding and crack characteristics of the borehole wall, and matching the collected borehole images with the core images in step (1); (3) Core scanning image collection: Use the laboratory core scanning device to scan the cores collected at the drilling site to obtain core scanning images; (4) Processing the core image, the borehole image, and the core scan image, cropping the images into a standard format of 512 pixels × 512 pixels, and performing super-resolution processing on the images with lower resolution to obtain high-resolution images; (5) Feature extraction and matching: using feature extraction algorithms such as SIFT, SURF, HOG, ORB, Harris corner detection or Canny edge detection, extracting feature points from the borehole image, the core image and the core scan image; the feature points include key points, edges and texture information of the image; matching the extracted feature points to determine the corresponding relationship between the borehole image, the core image and the core scan image; Then, the improved RANSAC algorithm is used to purify the matched feature points and remove the incorrectly matched feature points; (6) Image registration: Calculate the geometric transformation parameters between the borehole image, the core image and the core scan image based on the matched feature points, wherein the geometric transformation parameters include translation, rotation and scaling; then register the borehole image, the core image and the core scan image using the geometric transformation parameters so that the three are aligned in space; and obtain corresponding data set images; (7) Image annotation: The images in the above datasets are labeled using the image annotation tool LABELME, and polygonal labels are used to mark the features of bedding, fractures, and key points; (8) Image preprocessing: preprocessing the core image and the core scan image, including image cropping, scaling, grayscale or normalization operations to meet the requirements of subsequent processing; (9) Construct a generative adversarial network model: The generative adversarial network includes a generator and a discriminator: The generator is configured to generate a repaired borehole image according to the core image and the core scan image, and the discriminator is configured to determine whether the image generated by the generator is authentic; (10) Fix the generator and train the discriminator: The real drilling image and the repaired drilling image generated by the generator are input into the discriminator respectively. The discriminator will output a probability value, which represents the probability that the input image is real. Then, the loss function of the discriminator is calculated based on the output result of the discriminator and the real label. Then, the optimizer is used to back-propagate the loss function to update the parameters of the discriminator to improve the discriminant ability of the discriminator. (11) Fixing the discriminator and training the generator: inputting the core image and the core scan image into the generator, and inputting the repaired borehole image generated by the generator into the discriminator. The discriminator will output a probability value, which represents the probability that the input image is real; then, according to the output result of the discriminator and the real label, the loss function of the generator is calculated; then, the optimizer is used to back-propagate the loss function and update the parameters of the generator to improve the generation ability of the generator and make the generated image more realistic; (12) Alternating training of the discriminator and the generator: During the training process, the discriminator is trained first, then the generator is trained, and this process is repeated. In each training cycle, the discriminator is trained N times first, and then the generator is trained M times to maintain a balance between the two until the two models converge. Where N and M are both positive integers, NM ≥ 0, and M and N are positive integers greater than or equal to 1 respectively. (13) When the loss function tends to a stable state, the generator and the discriminator are fixed in turn; (14) The borehole image to be repaired and the core scan image corresponding to the borehole image to be repaired are respectively input into the generator to obtain the repaired borehole image and the repaired core scan image respectively. The repaired borehole image and the repaired core scan image are processed and calculated using a similarity function. When the calculated similarity is greater than a set threshold, the repaired borehole image is the final repaired borehole image.

2. The method according to claim 1, characterized in that: During training, the learning rate of the discriminator was configured to be 0.0003-0.0006, and the learning rate of the generator was configured to be 0.00005-0.0003.

3. The method according to claim 2, characterized in that: The learning rate of the discriminator is configured as 0.0004, and the learning rate of the generator is configured as 0.0001.

4. The method according to claim 1, characterized in that: The images generated by the generator are monitored by comparing them with real images. When the discriminator loss function is close to 0, the network is adjusted, including parameter adjustment, replacement of the loss function or adjustment of the network structure.

5. The method according to claim 1, characterized in that: Before the similarity function calculates the similarity of the repaired borehole image and the repaired core image, the eigenvalues ​​and eigenvectors of the repaired borehole image and the repaired core image are extracted, and the similarity of the repaired borehole image and the repaired core image is calculated according to the eigenvalues ​​and the eigenvectors.

6. The method according to claim 1, characterized in that: The step (8) further includes performing image processing by using data enhancement technology to increase the diversity of training data; the image processing includes rotation, flipping and translation.

7. The method according to claim 1, characterized in that: The generator selects convolutional neural network (CNN) as the basic architecture, and the discriminator selects PatchGAN.

8. The method according to claim 1, characterized in that : According to the discriminator loss function and generator loss function in the previous training cycle, the training times of the discriminator and generator in the next training cycle are adjusted according to the adjustment rules.

9. The method according to claim 8, characterized in that : The adjustment rule is configured as: after each training cycle, a step of determining whether the discriminator loss function is greater than a preset value; 1) When the discriminator loss function is greater than the preset value, in the next training cycle, the number of discriminator training times is set to N times, and the number of generator training times is set to M+K times; 2) When the discriminator loss function is less than the preset value, in the next training cycle, the number of discriminator training times is set to N+K times, and the number of generator training times is set to M times; Where K is the training balance coefficient, K = M × [N / M].

10. The method according to claim 9, characterized in that: The initial value of the discriminator training times N is set to N=10 times, and the initial value of the generator training times M is set to M=3 times.

Images