Digital core image expansion generation method and device

Abstract

The invention relates to the technical field of digital cores, and particularly discloses a digital core image expansion generation method and device, and the method comprises the steps: obtaining the initial core image data and expansion direction information of a target reservoir; performing expansion generation on the initial core image data along a specified expansion direction represented by the expansion direction information by using a target expansion model to obtain corresponding target expansion image data; the target expansion model comprises a target stable diffusion model, the target stable diffusion model is obtained by adding a low-rank matrix into a related layer of an original stable diffusion model by using a low-rank adapter algorithm and performing fine tuning, and parameters of the low-rank adapter algorithm are obtained by training a core image data set; and generating target rock core image data according to the initial rock core image data, the extension direction information and the target extension image data. According to the scheme, expansion of the core image can be realized, so that the core image with larger size and more information can be generated.

Description

Technical Field

[0001] This specification relates to the technical field of digital cores, and particularly to a method and device for generating extended digital core images. Background Art

[0002] At present, unconventional oil and gas energy is of great significance to energy development. The reservoirs of unconventional oil and gas reservoirs usually have the characteristics of being dense, low-porosity and low-permeability. Therefore, the analysis of the microscopic seepage mechanism of reservoir rocks is crucial. Digital core technology solves problems such as complex samples and short working cycles existing in real cores by establishing three-dimensional digital core images reflecting the physical properties of cores. Existing digital core technology uses advanced image scanning methods to obtain core images, and specific algorithms, programs or professional software can be used to process and analyze core samples. However, due to the high cost of current scanning means and the influence of other external environmental factors, researchers can often only obtain partial core images, and the unknown areas have a certain degree of spatial randomness.

[0003] In response to the above problems, no effective solution has been proposed yet. Summary of the Invention

[0004] Embodiments of this specification provide a method and device for generating extended digital core images to solve the problem in the prior art that only partial core images can be obtained due to the high cost of current scanning means and the influence of external environmental factors.

[0005] Embodiments of this specification provide a method for generating extended digital core images, including:

[0006] Obtaining initial core image data of a target reservoir and extension direction information; the extension direction information is used to characterize a specified extension direction in which the initial core image data is to be extended;

[0007] Using a target extension model to perform extension generation on the initial core image data along the specified extension direction to obtain corresponding target extension image data; the target extension model includes a target stable diffusion model, and the target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of an original stable diffusion model and fine-tuning it using a low-rank adapter algorithm, and the parameters of the low-rank adapter algorithm are obtained by training using a core image data set;

[0008] Generating target core image data according to the initial core image data, the extension direction information and the target extension image data.

[0009] In one embodiment, the target expansion model further includes a tiled diffusion variational autoencoder, which is used to extract features from the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship;

[0010] Correspondingly, the input of the target stable diffusion model includes the latent vector and the expansion direction information, and the target stable diffusion model is used to expand and generate the latent vector along the specified expansion direction to obtain preliminary expanded image data.

[0011] In one embodiment, the target expansion model further includes a local redrawing model; the target stable diffusion model is used to expand and generate the initial core image data along the specified expansion direction to obtain preliminary expanded image data; the input of the local redrawing model includes the preliminary expanded image data; the local redrawing model is used to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain the target expanded image data.

[0012] In one embodiment, the target expansion model further includes: a tiled diffusion variational autoencoder and a local redrawing model;

[0013] Using the target expansion model to expand and generate the initial core image data along the specified expansion direction to obtain the corresponding target expanded image data includes:

[0014] Using the tiled diffusion variational autoencoder to extract features from the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship;

[0015] Using the target stable diffusion model to expand and generate the initial core image data along the specified expansion direction based on the latent vector to obtain preliminary expanded image data;

[0016] Using the local redrawing model to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain the target expanded image data.

[0017] In one embodiment, the expansion direction information includes multiple specified expansion directions;

[0018] Correspondingly, using the target expansion model, the initial core image data is expanded along the specified expansion direction to generate corresponding target expansion image data, including:

[0019] Using the target expansion model, the initial core image data is expanded along each of the specified expansion directions among the multiple specified expansion directions to generate the target expansion image data corresponding to each specified expansion direction;

[0020] Correspondingly, according to the initial core image data, the expansion direction information, and the target expansion image data, target core image data is generated, including:

[0021] According to the initial core image data, the expansion direction information, and the target expansion image data corresponding to each specified expansion direction, target core image data is generated.

[0022] In one embodiment, according to the initial core image data, the expansion direction information, and the target expansion image data, generating target core image data includes:

[0023] According to the expansion direction information, determine the splicing position for splicing the target expansion image and the initial core image data;

[0024] Perform splicing processing and fusion processing on the initial core image data and the target expansion image data according to the splicing position to obtain the target core image data.

[0025] The embodiments of this specification also provide a digital core image expansion generation device, including:

[0026] An acquisition module, configured to acquire the initial core image data and expansion direction information of the target reservoir; the expansion direction information is used to characterize the specified expansion direction in which the initial core image data is to be expanded;

[0027] An expansion module, configured to use the target expansion model to expand and generate the initial core image data along the specified expansion direction to obtain corresponding target expansion image data; the target expansion model includes a target stable diffusion model, and the target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model for fine-tuning using the low-rank adapter algorithm, and the parameters of the low-rank adapter algorithm are obtained by training using a core image dataset;

[0028] A generation module, configured to generate target core image data according to the initial core image data, the expansion direction information, and the target expansion image data.

[0029] In one embodiment, the target expansion model further includes: a tiled diffusion variational autoencoder and a local redrawing model;

[0030] Correspondingly, the expansion module is specifically configured to: use the tiled diffusion variational autoencoder to extract features from the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship; use the target stable diffusion model to perform expansion generation on the initial core image data along the specified expansion direction based on the latent vector to obtain preliminary expanded image data; use the local redrawing model to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain target expanded image data.

[0031] An embodiment of this specification also provides a computer device, including a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, the steps of the digital core image expansion generation method described in any of the above embodiments are implemented.

[0032] An embodiment of this specification also provides a computer-readable storage medium, on which computer instructions are stored. When the instructions are executed, the steps of the digital core image expansion generation method described in any of the above embodiments are implemented.

[0033] In the embodiments of this specification, a method for generating extended digital core images is provided. Initial core image data of a target reservoir and extended direction information can be obtained, and using a target extension model, the initial core image data is extended along a specified extension direction represented in the extended direction information to generate corresponding target extended image data. The target extension model includes a target stable diffusion model, which is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model and fine-tuning it using the low-rank adapter algorithm. The parameters of the low-rank adapter algorithm are obtained by training using a core image dataset. According to the initial core image data, extended direction information, and target extended image data, target core image data is generated. In the above solution, through the image extension generation technology in the target stable diffusion model, the dependence of digital cores on high-cost scanning is reduced, and the cost of analyzing rock characteristics using digital cores is significantly reduced. This not only enables more research projects to be carried out within a limited budget but also provides economic feasibility for large-scale research on unconventional oil and gas reservoir reservoirs, promoting the development of this field. Further, the original stable diffusion model is fine-tuned using the low-rank adapter algorithm, enabling the obtained target stable diffusion model to quickly adapt to the characteristics of core image data. Without changing most of the parameters of the original model, the unique features and patterns of core images are effectively learned through the core image dataset to obtain the parameters of the low-rank adapter algorithm. This method improves the adaptability and generalization ability of the target stable diffusion model, which is not only applicable to current core image data but also has good processing effects on core images from different sources and of different types, providing possibilities for more diverse core research in the future. In addition, the technical achievements of this solution not only have important applications in the research of unconventional oil and gas reservoir reservoirs but can also be extended to other fields. For example, in geological exploration, more detailed underground rock structure images can be generated to help more accurately judge the formation structure and mineral distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The drawings described herein are used to provide a further understanding of this specification, form a part of this specification, and do not limit this specification. In the drawings:

[0035] Figure 1 The flowchart of the method for generating extended digital core images in an embodiment of this specification is shown;

[0036] Figure 2 The original core image obtained by a high-precision scanning instrument in an embodiment of this specification is shown;

[0037] Figure 3 The core image to be extended in an embodiment of this specification is shown;

[0038] Figure 4Shows the flowchart of picture expansion in an embodiment of this specification;

[0039] Figure 5 Shows the schematic diagram of the final image expansion result in an embodiment of this specification;

[0040] Figure 6 Shows the schematic diagram of the digital core image expansion generation device in an embodiment of this specification;

[0041] Figure 7 Shows the schematic diagram of the computer device in an embodiment of this specification. Detailed implementation manners

[0042] Hereinafter, the principles and spirit of this specification will be described with reference to several exemplary implementation manners. It should be understood that these implementation manners are provided only to enable those skilled in the art to better understand and then implement this specification, rather than limiting the scope of this specification in any way. On the contrary, these implementation manners are provided to make the disclosure of this specification more thorough and complete, and to be able to convey the scope of this disclosure fully to those skilled in the art.

[0043] Those skilled in the art know that the implementation manners of this specification can be implemented as a system, a device, a method, or a computer program product. Therefore, the disclosure of this specification can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0044] The embodiment of this specification provides a digital core image expansion generation method. Figure 1 Shows the flowchart of the digital core image expansion generation method in an embodiment of this specification. Although this specification provides the method operation steps or device structures as shown in the following embodiments or drawings, based on routine or non-creative labor, more or fewer operation steps or module units may be included in the method or device. In the steps or structures where there is no necessary causal relationship logically, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure described in the embodiments of this specification and shown in the drawings. When the method or module structure is applied to the actual device or terminal product, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or drawings (for example, in a parallel processor or multi-threaded processing environment, or even a distributed processing environment).

[0045] Specifically, as Figure 1 shown, the digital core image expansion generation method provided by an embodiment of this specification may include the following steps.

[0046] Step S101: Obtain the initial core image data of the target reservoir and the extension direction information; the extension direction information is used to characterize the specified extension direction in which the initial core image data is to be extended.

[0047] The method in this embodiment can be applied to a computer device. The computer device can be a server or a client device with processing and computing capabilities. The initial core image data of the target reservoir can be obtained. The initial core image data is the core image data to be extended. In one embodiment, the initial core image data of the target reservoir can be obtained by a high-precision scanning instrument. The extension direction information can be the extension direction specified by the user. In one embodiment, the image shape corresponding to the initial core image data can be square or rectangular. Correspondingly, the specified extension direction can include at least one of the following: left, right, up, down, upper left, lower left, upper right, lower right. In another embodiment, the image shape corresponding to the initial core image data can be circular. Correspondingly, the specified extension direction can be radial.

[0048] Step S102: Use the target extension model to perform extension generation on the initial core image data along the specified extension direction to obtain the corresponding target extended image data; the target extension model includes a target stable diffusion model, and the target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model and performing fine-tuning using the low-rank adapter algorithm, and the parameters of the low-rank adapter algorithm are obtained by training using a core image dataset.

[0049] After obtaining the initial core image data and the extension direction information, the target extension model can be used to perform extension generation on the initial core image data along the specified extension direction to obtain the corresponding target extended image data. The target extension model includes a target stable diffusion model. The target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model and performing fine-tuning using the low-rank adapter algorithm. The Stable Diffusion (SD) model is a generative artificial intelligence model based on the diffusion model. The core idea of the diffusion model is to gradually add noise to the data (such as pictures) and then learn how to recover the original data from the noise. In the image generation task, it starts from a pure noise image and gradually generates a realistic target image through multiple iterative denoising processes. Its training process is relatively stable and can generate high-quality images.

[0050] The Low-Rank Adaptation (LoRA) algorithm is used for efficient fine-tuning on large-scale pre-trained models (e.g., the original Stable Diffusion model in this embodiment). A large amount of core image data can be used to perform LoRA training on the original Stable Diffusion model. LoRA enables the model to quickly adapt to the characteristics of core image data without changing most of the parameters of the original model by adding low-rank matrices to certain layers of the model. During image extension generation, the model fine-tuned by LoRA can better generate content that conforms to the characteristics of core images, such as generating an extended part similar to the initial core image in terms of texture, structure, etc., thereby achieving effective extension of core images. The introduction of low-rank matrices reduces the number of parameters to be trained, thus reducing the computational cost and training time. By training LoRA, the resulting target Stable Diffusion model can learn specific features and patterns of core image data, making the generated core images more accurately reflect the properties of real cores.

[0051] Step S103: Generate target core image data according to the initial core image data, the extension direction information, and the target extended image data.

[0052] After obtaining the target extended image data, target core image data can be generated according to the initial core image data, the extension direction information, and the target extended image data. For example, the initial core image data and the target extended image data can be spliced according to the specified extension direction to obtain the target core image data.

[0053] In the above embodiment, through the image extension generation technology in the target Stable Diffusion model, the dependence of digital cores on high-cost scanning is reduced, and the cost of analyzing rock characteristics of digital cores is significantly reduced. This not only enables more research projects to be carried out within a limited budget but also provides economic feasibility for large-scale research on unconventional oil and gas reservoir formations, promoting the development of this field. Further, the original Stable Diffusion model is fine-tuned using the low-rank adapter algorithm, enabling the resulting target Stable Diffusion model to quickly adapt to the characteristics of core image data. Without changing most of the parameters of the original model, the unique features and patterns of core images are effectively learned through the core image dataset to obtain the parameters of the low-rank adapter algorithm. This approach improves the adaptability and generalization ability of the target Stable Diffusion model, which is not only applicable to the current core image data but also has good processing effects on core images from different sources and of different types, providing possibilities for more diverse core research in the future. In addition, the technical achievements of this solution not only have important applications in the research of unconventional oil and gas reservoir formations but can also be extended to other fields. For example, in geological exploration, it can generate more detailed underground rock structure images to help more accurately judge the formation structure and mineral distribution.

[0054] In some embodiments of this specification, the target expansion model may further include a tiled diffusion variational autoencoder, which is used to extract features from the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector may include key features of the initial core image data, and the key features may include at least one of the following: pore structure, permeability, and spatial relationship; correspondingly, the input of the target stable diffusion model may include the latent vector and the expansion direction information, and the target stable diffusion model is used to expand and generate the latent vector along the specified expansion direction to obtain preliminary expanded image data.

[0055] Specifically, a Variational Autoencoder (VAE) is a neural network structure used to learn the latent representation of data. When applied to SD, its main role is to encode and decode images. The encoding process converts the input image into a low-dimensional latent vector representation, which contains key feature information of the image, such as pore distribution, particle structure, etc. The decoding process generates a reconstructed image based on the latent vector. Using VAE in SD can improve the quality and efficiency of the generated images, and at the same time can effectively compress and represent the features of the images, enabling the model to better learn and generate images that meet the requirements. For example, when processing core images, VAE can encode the complex structure and features of the core image into a latent vector, and then when generating an expanded image, generate a new image with similar features by decoding the latent vector. In this way, VAE helps the model understand and capture the internal structure of the core image, providing a feature basis for expansion generation.

[0056] In some embodiments of this specification, the target expansion model may further include a local redrawing model; the target stable diffusion model is used to expand and generate the initial core image data along the specified expansion direction to obtain preliminary expanded image data; the input of the local redrawing model may include the preliminary expanded image data; the local redrawing model is used to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain the target expanded image data.

[0057] Specifically, the local redrawing model (e.g., ControlNet) precisely controls and redraws the local part of the preliminary extended image data according to the input specific conditions (such as the local features of the initial core image). In core image extension, a partial area of the initial core image (such as the edge, specific structure, etc.) is used as a control condition to guide the model to generate content that naturally connects with the initial core image data in the extended area. The local redrawing model can ensure a smooth transition between the extended part and the original part, avoiding obvious seams or unnatural areas, making the finally generated target core image data more complete and realistic as a whole, and ensuring the quality and coherence of the image.

[0058] In some embodiments of this specification, the target extension model may further include: a tiled diffusion variational autoencoder and a local redrawing model; using the target extension model to perform extension generation on the initial core image data along the specified extension direction to obtain the corresponding target extended image data may include: using the tiled diffusion variational autoencoder to extract features from the initial core image data to obtain the latent vector corresponding to the initial core image data; the latent vector may include the key features of the initial core image data, and the key features may include at least one of the following: pore structure, permeability, and spatial relationship; using the target stable diffusion model to perform extension generation on the initial core image data along the specified extension direction based on the latent vector to obtain the preliminary extended image data; using the local redrawing model to modify and redraw the preliminary extended image data based on the extension direction information and the initial core image data to obtain the target extended image data.

[0059] In this embodiment, the target extension model may include: a tiled diffusion variational autoencoder, a target stable diffusion model, and a local redrawing model. The initial core image data can be input into the target extension model to obtain the target extended image data. Specifically, the variational autoencoder part includes an encoder and a decoder. The encoder is responsible for mapping the image data after tiling and diffusion processing into the latent space to generate the latent vector. The decoder can then reconstruct the latent vector into an image to evaluate the quality of the features extracted by the encoder. By continuously adjusting the parameters of the encoder and decoder, the reconstructed image is made as close as possible to the original image, so that the encoder can accurately extract the key features of the image. The key features of the initial core image contained in the latent vector are crucial for subsequent image extension.

[0060] The pore structure of the core is one of its important microscopic features. Information such as the size, shape, and distribution of pores is encoded into the latent vector. For example, large pores and small pores have different grayscale values and spatial positions in the image, and the tiled diffusion variational autoencoder can identify these differences and convert them into specific dimensional information in the latent vector. Through the accurate extraction of the pore structure, the subsequently extended generated images can maintain pore characteristics similar to the initial core image, which is of great significance for studying the reservoir performance of the core.

[0061] The permeability of the core reflects the ability of the core to allow fluid to pass through, and is closely related to the pore structure and connectivity of the core. The tiled diffusion variational autoencoder can encode permeability-related features into the latent vector by analyzing information such as the connectivity between pores and the tortuosity of pores in the image. Although the permeability cannot be directly observed visually from the image, the model can learn the image feature patterns related to permeability and thus be reflected in the latent vector.

[0062] The spatial relationship between different structures in the core image is also one of the key features. For example, the relative position between pores and particles, the distribution relationship between different types of pores, etc. The tiled diffusion variational autoencoder can capture this spatial relationship information and integrate it into the latent vector. The accurate extraction of this spatial relationship helps the subsequently extended generated images to maintain consistency in structure with the original image and avoid unreasonable structure splicing.

[0063] The target stable diffusion model generates images by performing a reverse denoising process in the latent space. During the training phase, it learns the feature distribution and generation rules of a large number of images. For the core image extension task, the low-rank adapter algorithm is used to fine-tune the model to better adapt to the characteristics of core images. After obtaining the latent vector corresponding to the initial core image data, the latent vector and the specified extension direction information are input into the target stable diffusion model. The specified extension direction information can be concatenated with the latent vector in an encoded manner as the input to the target stable diffusion model. The target stable diffusion model performs reverse denoising operations in the latent space according to the input information. Starting from a random noise latent vector, combining the key features of the core image and the extension direction information contained in the latent vector, the noise is gradually removed to generate a latent vector that is similar in features to the initial core image data and conforms to the specified extension direction. Then, the generated latent vector is decoded into preliminary extended image data by the decoder of the tiled diffusion variational autoencoder. During the generation process, the target stable diffusion model will utilize the learned core image features and patterns to attempt to continue the structure and features of the initial core image in the specified extension direction. For example, if the specified extension direction is upward, the model will generate an extended part with similar features above the initial core image according to the information such as pore structure, permeability, and spatial relationship recorded in the latent vector.

[0064] The local redrawing model is used to perform fine processing on the preliminary extended image data generated by the target stable diffusion model. It can accurately modify and redraw specific areas in the preliminary extended image according to the extended direction information and the initial core image data. The local redrawing model receives the extended direction information, the initial core image data and the preliminary extended image data. The extended direction information clarifies the area to be redrawn and the redrawing direction requirements; the initial core image data is used as a reference to provide the real characteristics of the initial core image; and the preliminary extended image data is the object to be modified. The local redrawing model will match the features of the preliminary extended image with the initial core image to find out the inconsistencies or unnatural places between the two. For example, there may be problems such as texture discontinuity and structural incoordination at the junction of the extended part and the initial image. The local redrawing model adjusts these inconsistent areas based on the extended direction information and the initial core image as a template. The extended part and the initial image data can be made to transition more naturally in color, texture and structure by adjusting pixel values, changing texture patterns, etc.

[0065] In the above embodiment, the collaborative work of the tiled diffusion variational autoencoder, the target stable diffusion model and the local redrawing model can effectively expand and generate the initial core image data, obtain high-quality target extended image data, and provide richer image information for core research.

[0066] In some embodiments of the present specification, the extension direction information may include multiple specified extension directions; accordingly, using the target extension model to expand and generate the initial core image data along the specified extension directions to obtain corresponding target extended image data may include: using the target extension model to expand and generate the initial core image data along each of the multiple specified extension directions, respectively, to obtain target extended image data corresponding to each specified extension direction; accordingly, generating target core image data based on the initial core image data, the extension direction information and the target extended image data may include: generating target core image data based on the initial core image data, the extension direction information and the target extended image data corresponding to each specified extension direction.

[0067] Specifically, when the expansion direction information includes multiple specified expansion directions, the process of generating an image expansion using the target expansion model can be carried out sequentially in different directions. In actual operation, the expansion generation for multiple specified expansion directions can be performed in parallel to improve processing efficiency. However, since the expansions in different directions may overlap or interact with each other in areas such as the edges or corners of the image, a certain coordination mechanism is required. For example, priorities or weights can be set, and higher priorities can be given to some key directions (such as the expansion direction consistent with the main structure direction of the core) to ensure the expansion quality in these directions. At the same time, when processing the overlapping areas, a fusion algorithm can be adopted to reasonably fuse the results generated by the expansions in different directions to avoid conflicts or unnatural splicing. After using the target expansion model to expand and generate the initial core image data along each specified expansion direction among the multiple specified expansion directions to obtain the target expansion image data corresponding to each specified expansion direction, the target core image data can be generated based on the initial core image data, the expansion direction information, and the target expansion image data corresponding to each specified expansion direction. After generating the target core image data, its overall quality needs to be evaluated. Multiple evaluation metrics can be used, such as the peak signal-to-noise ratio (PSNR), the structural similarity index (SSIM), etc., to measure the quality of the image. At the same time, from a geological professional perspective, the rationality and consistency of the core features in the image can also be evaluated, such as the continuity of the pore structure and the rationality of the permeability characteristics. If the evaluation results are not satisfactory, the generation process needs to be optimized. The previous steps can be returned to adjust the parameters of the target expansion model, such as the encoding parameters of the tiled diffusion variational autoencoder, the fine-tuning parameters of the target stable diffusion model, or the redrawing strategy of the local redrawing model. The rationality of the expansion direction information can also be re-evaluated to determine whether certain expansion directions need to be added or reduced to improve the quality of the finally generated target core image data. Through the above methods, the initial image data can be expanded along multiple expansion directions to obtain a larger-sized target core image data, providing richer and more accurate image materials for core research.

[0068] In some embodiments of the present specification, generating the target core image data based on the initial core image data, the expansion direction information, and the target expansion image data may include: determining the splicing position where the target expansion image is spliced with the initial core image data according to the expansion direction information; splicing and fusing the initial core image data and the target expansion image data according to the splicing position to obtain the target core image data.

[0069] Specifically, according to the expansion direction information, determine the splicing position where the target expanded image is spliced with the initial core image data. The expansion direction information plays a crucial guiding role in this process. Different expansion directions (such as up, down, left, right, upper left, upper right, lower left, lower right, etc.) directly determine the specific position of the target expanded image around the initial core image. For example, when the expansion direction is "up", the target expanded image should be spliced above the initial core image; if the expansion direction is "lower right", the target expanded image needs to be placed at the lower right corner of the initial core image. To more accurately determine the splicing position, factors such as the size and proportion of the images also need to be considered. If the size of the target expanded image does not exactly match that of the initial core image, appropriate scaling or cropping operations may be required on the target expanded image to ensure the rationality of the spliced image in terms of visual appearance and data structure. At the same time, for some core image analysis tasks with special requirements, it may also be necessary to fine-tune the splicing position according to the specific structural characteristics or research focus of the core, so that the spliced image can better display the required information. Next, perform splicing processing and fusion processing on the initial core image data and the target expanded image data according to the splicing position to obtain the target core image data. The splicing processing is to accurately place the target expanded image at the determined splicing position to achieve the preliminary combination of the images. In this process, the alignment accuracy between the images should be ensured to avoid misalignment or deviation, otherwise it may affect the subsequent analysis and interpretation of the core image. The fusion processing is to make the spliced image more natural and coherent and eliminate the splicing traces. Since there may be some differences between the target expanded image and the initial core image during the generation process, such as color, brightness, texture, etc., the fusion processing can be achieved through various algorithms. Common fusion algorithms include the weighted average fusion algorithm, which assigns appropriate weights to the pixels of the target expanded image and the initial core image according to different regions and features of the images, and obtains the fused pixel values through weighted calculation, thereby achieving a smooth transition of the image. In addition, a fusion algorithm based on multi-resolution analysis, such as the wavelet transform fusion algorithm, can be used to perform fusion processing on the image at different resolution levels, better retain the detail information of the image, and improve the quality of the fused image. After the fusion processing, some post-processing operations can also be performed on the generated target core image data, such as adjusting the contrast, brightness, color saturation, etc. of the image to further optimize the visual effect of the image and make it more in line with the requirements of core image analysis. At the same time, quality assessment of the target core image data is also an essential step. The quality of the image can be measured by calculating relevant indexes of the image (such as peak signal-to-noise ratio, structural similarity index, etc.) to ensure that the generated target core image data can meet the requirements of subsequent research and applications. Through the above methods, the purpose of inferring the whole from the part can be achieved, and finally high-quality image expansion generation can be realized, and it has good adaptability and scalability.

[0070] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized. Specifically, reference can be made to the description of the relevant embodiments of the foregoing related processing, and details will not be repeated here.

[0071] The specific embodiments of this specification have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The above method will be described below in conjunction with a specific embodiment. However, it should be noted that this specific embodiment is only for better explaining this specification and does not constitute an improper limitation to this specification.

[0073] In this specific embodiment, a method for generating an extended digital core image is provided. By combining the StableDiffusion model and digital core technology, based on the core image obtained by a high-precision scanning instrument, the purpose of image extension is achieved, thereby generating a core image with a larger size and more information, and achieving the goal of reducing the core cost. The method may include the following steps.

[0074] Step 1, obtain a core image using a high-precision scanning instrument.

[0075] Set up the SD software. Apply the tiled diffusion variational autoencoder (VAE) to SD, select Realistic Vision V2.0 as the basic model, and use a large number of core images to train the low-rank adaptation (LoRA) algorithm, combined with the local redrawing model ControlNet model.

[0076] Combine the trained LoRA and ControlNet. The usage parameters of LoRA are shown in Table 1 below.

[0077] Table 1

[0078] Parameter Value Image Size 256×256 Data Volume 400 Maximum Number of Training Epochs (Max epochs) 20 Batch Size 1 Learning Rate <![CDATA[1.00×10 -4 > Learning Rate Scheduler Cosine with restarts Learning Rate Warmup Steps 0 Number of Cycles of the Learning Rate Scheduler 1 Optimizer Type Adam W8bit CLIP Model Usage Frequency (Clip skip) 2 Mixed Precision Fp16 Network Dimension 32 Network Alpha 32

[0079] Step 2: Take the scanned core image (the image that needs to be image-expanded) as the input picture, and use the set-up SD software to sequentially expand the picture in eight directions. Finally, combine the 9 pictures into one final image.

[0080] The method in the above embodiment solves the problems caused by environmental factors and high scanning costs. The stable diffusion model trained with a large number of core pictures is used to generate the unscanned core area, greatly reducing the cost of digital core analysis of rock characteristics.

[0081] The above solution will be described below in conjunction with specific embodiments. The method in this embodiment includes the following content.

[0082] Step 1: Obtain the original core image through a high-precision scanning instrument, as Figure 2 shown. Divide the original core image into 9 parts.

[0083] Step 2: Set up the SD software. Among them, the basic model Realistic Vision V2.0 is usually used for the processing of generating realistic images, especially portrait images. By comparing with the effects of other basic models, the basic model Realistic Vision V2.0 has the best effect when applied to core images after fine-tuning, and is usually used together with VAE to improve the quality of the generated images and eliminate blue artifacts. In addition, the local redrawing model of ControlNet combines the redrawn area with the whole, effectively avoiding obvious seams and ensuring a smooth transition at the boundaries.

[0084] Step 3: Take the middle local image (black frame) of the original core image as the input image (i.e., the initial core image data in the above embodiment), as Figure 3 shown, and expand it sequentially in eight directions, a total of 8 times. The expansion process is as Figure 4 shown.

[0085] Step 3: Integrate the generated images to obtain the final expansion result, as Figure 5 shown.

[0086] During the process of training the model, the model recognizes and understands the complex structure of the images from a large number of core images. However, after comparison Figure 2 and Figure 5The specific details will lead to a conclusion that there is a significant gap between the image of the extended part and the original core image of the corresponding area. However, the attributes of the extended part image are very similar to those of the original core image, such as porosity, permeability, and spatial relationship, which are exactly the laws required in the field of oil and gas analysis. The model in SD uses the learned knowledge to generate content similar to the law of the original core image, thus achieving the purpose of inferring the whole from the part and finally realizing high-quality image extension. This method can effectively solve the problem of high cost of digital core scanning.

[0087] In the above embodiments, a method for generating an extended digital core image is provided. Combining the stable diffusion model and digital core technology, based on the core image obtained by a high-precision scanning instrument, the purpose of image extension is achieved, so as to generate a core image with a larger size and more information, and achieve the goal of reducing the core cost. The technical solution is as follows: obtaining a core image by using a high-precision scanning instrument; setting up SD software. Applying the tiled diffusion variational autoencoder to SD, selecting RealisticVision V2.0 as the basic model, combining the local redrawing model ControlNet as the basic model, and training LoRA with a large number of core images; combining the trained LoRA and ControlNet; using the initially scanned core image (the image that needs to be extended) as the input picture, and using the set-up SD software to perform picture extension in eight directions in sequence, and finally combining the 9 pictures into one final image.

[0088] Based on the same inventive concept, an apparatus for generating an extended digital core image is also provided in the embodiments of this specification, as described in the following embodiments. Since the principle of solving problems by the apparatus for generating an extended digital core image is similar to that of the method for generating an extended digital core image, the implementation of the apparatus for generating an extended digital core image can refer to the implementation of the method for generating an extended digital core image, and the repeated parts will not be described again. As used below, the term "unit" or "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the apparatuses described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated. Figure 6 is a structural block diagram of the apparatus for generating an extended digital core image according to the embodiments of this specification, as Figure 6 shown, including: an acquisition module 601, an extension module 602, and a generation module 603. The following describes this structure.

[0089] The acquisition module 601 is used to acquire the initial core image data of the target reservoir and the extension direction information; the extension direction information is used to characterize the specified extension direction in which the initial core image data is to be extended.

[0090] The expansion module 602 is used to generate corresponding target expanded image data by expanding the initial core image data along the specified expansion direction using a target expansion model; the target expansion model includes a target stable diffusion model, and the target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model and fine-tuning it using the low-rank adapter algorithm, and the parameters of the low-rank adapter algorithm are obtained by training using a core image dataset.

[0091] The generation module 603 is used to generate target core image data according to the initial core image data, the expansion direction information, and the target expanded image data.

[0092] In some embodiments of the present specification, the target expansion model further includes a tiled diffusion variational autoencoder, and the tiled diffusion variational autoencoder is used to extract features from the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship; correspondingly, the input of the target stable diffusion model includes the latent vector and the expansion direction information, and the target stable diffusion model is used to expand and generate the latent vector along the specified expansion direction to obtain preliminary expanded image data.

[0093] In some embodiments of the present specification, the target expansion model further includes a local redrawing model; the target stable diffusion model is used to expand and generate the initial core image data along the specified expansion direction to obtain preliminary expanded image data; the input of the local redrawing model includes the preliminary expanded image data; the local redrawing model is used to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain target expanded image data.

[0094] In some embodiments of the present specification, the target expansion model further includes: a tiled diffusion variational autoencoder and a local redrawing model; the expansion module is specifically used for: using the tiled diffusion variational autoencoder to extract features from the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship; using the target stable diffusion model to expand and generate the initial core image data along the specified expansion direction based on the latent vector to obtain preliminary expanded image data; using the local redrawing model to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain target expanded image data.

[0095] In some embodiments of this specification, the extension direction information includes multiple specified extension directions; correspondingly, the extension module is specifically configured to: use a target extension model to respectively extend the initial core image data along each of the multiple specified extension directions to generate target extension image data corresponding to each specified extension direction; correspondingly, the generation module is specifically configured to: generate target core image data according to the initial core image data, the extension direction information, and the target extension image data corresponding to each specified extension direction.

[0096] In some embodiments of this specification, the generation module is specifically configured to: determine the splicing position where the target extension image is spliced with the initial core image data according to the extension direction information; perform splicing processing and fusion processing on the initial core image data and the target extension image data according to the splicing position to obtain the target core image data.

[0097] The embodiments of this specification also provide a computer device, which can specifically refer to Figure 7 the schematic structural diagram of the computer device based on the digital core image extension generation method provided in the embodiments of this specification as shown. The computer device can specifically include an input device 71, a processor 72, and a memory 73. Among them, the memory 73 is used to store instructions executable by the processor. When the processor 72 executes the instructions, it implements the steps of the digital core image extension generation method described in any of the above embodiments.

[0098] In this embodiment, the input device may specifically be one of the main devices for information exchange between the user and the computer system. The input device may include a keyboard, a mouse, a camera, a scanner, a light pen, a handwriting input board, a voice input device, etc.; the input device is used to input raw data and the programs for processing these data into the computer. The input device may also acquire and receive data transmitted from other modules, units, and devices. The processor may be implemented in any suitable manner. For example, the processor may take the form of, for example, a microprocessor or a processor, and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. The memory may specifically be a memory device for storing information in modern information technology. The memory may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function without a physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory module, a TF card, etc.

[0099] In this embodiment, the functions and effects specifically implemented by this computer device can be explained in comparison with other embodiments and will not be elaborated here.

[0100] This specification embodiment also provides a computer storage medium based on the digital core image extension generation method. The computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the digital core image extension generation method described in any of the above embodiments are implemented.

[0101] In this embodiment, the above storage medium includes but is not limited to a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk drive (HDD), or a memory card. The memory may be used to store computer program instructions. The network communication unit may be set according to the standards specified by the communication protocol and is an interface for network connection communication.

[0102] In this embodiment, the functions and effects specifically implemented by the program instructions stored in this computer storage medium can be explained in comparison with other embodiments and will not be elaborated here.

[0103] Obviously, those skilled in the art should understand that the various modules or steps of the above-described embodiments of the present specification can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed over a network composed of multiple computing devices. Optionally, they can be implemented with program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a sequence different from that here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. In this way, the embodiments of the present specification are not limited to any specific combination of hardware and software.

[0104] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and many applications other than the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this specification should not be determined with reference to the above description, but should be determined with reference to the full scope of the foregoing claims and the equivalents thereof.

[0105] The above is only the preferred embodiment of this specification and is not used to limit this specification. For those skilled in the art, various changes and modifications can be made to the embodiments of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the protection scope of this specification.

Claims

1. A method for generating an extended digital core image, characterized in that Including: Obtain the initial core image data and the expansion direction information of the target reservoir; The expansion direction information is used to characterize the specified expansion direction in which the initial core image data is to be expanded; Using a target expansion model, expand and generate the initial core image data along the specified expansion direction to obtain corresponding target expansion image data; the target expansion model includes a target stable diffusion model, and the target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model for fine-tuning using the low-rank adapter algorithm, and the parameters of the low-rank adapter algorithm are obtained by training using a core image data set; Generate target core image data according to the initial core image data, the expansion direction information, and the target expansion image data.

2. The digital core image expansion generation method according to claim 1, wherein The target expansion model further includes a tiled diffusion variational autoencoder, and the tiled diffusion variational autoencoder is used to extract features of the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship; Correspondingly, the input of the target stable diffusion model includes the latent vector and the expansion direction information, and the target stable diffusion model is used to expand and generate the latent vector along the specified expansion direction to obtain preliminary expansion image data.

3. The method for generating a digital core image extension according to claim 1, wherein The target expansion model further includes a local redrawing model; the target stable diffusion model is used to expand and generate the initial core image data along the specified expansion direction to obtain preliminary expansion image data; the input of the local redrawing model includes the preliminary expansion image data; the local redrawing model is used to modify and redraw the preliminary expansion image data based on the expansion direction information and the initial core image data to obtain target expansion image data.

4. The method for generating an extended digital core image according to claim 1, wherein The target expansion model further includes: a tiled diffusion variational autoencoder and a local redrawing model; Using a target expansion model, expand and generate the initial core image data along the specified expansion direction to obtain corresponding target expansion image data, including: Using the tiled diffusion variational autoencoder to extract features of the initial core image data to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship; Using the target stable diffusion model to expand and generate the initial core image data along the specified expansion direction based on the latent vector to obtain preliminary expansion image data; Using the local redrawing model to modify and redraw the preliminary expansion image data based on the expansion direction information and the initial core image data to obtain target expansion image data.

5. The method for generating a digital core image expansion according to claim 1, wherein The expansion direction information includes a plurality of specified expansion directions; Correspondingly, using a target expansion model, expand and generate the initial core image data along the specified expansion direction to obtain corresponding target expansion image data, including: Using the target expansion model, expand the initial core image data along each of the plurality of specified expansion directions respectively to generate the target expansion image data corresponding to each specified expansion direction; Correspondingly, generating target core image data according to the initial core image data, the expansion direction information, and the target expansion image data includes: Generating target core image data according to the initial core image data, the expansion direction information, and the target expansion image data corresponding to each specified expansion direction.

6. The method for generating a digital core image extension according to claim 1, wherein Generating target core image data according to the initial core image data, the expansion direction information, and the target expansion image data includes: Determining the splicing position where the target expansion image is spliced with the initial core image data according to the expansion direction information; Performing splicing processing and fusion processing on the initial core image data and the target expansion image data according to the splicing position to obtain the target core image data.

7. A digital core image extension generation device, characterized in that, Including: An acquisition module for acquiring the initial core image data and expansion direction information of the target reservoir; The expansion direction information is used to characterize the specified expansion direction in which the initial core image data is to be expanded; An expansion module for using the target expansion model to expand and generate the initial core image data along the specified expansion direction to obtain the corresponding target expansion image data; The target expansion model includes a target stable diffusion model, and the target stable diffusion model is obtained by adding a low-rank matrix to relevant layers of the original stable diffusion model for fine-tuning using the low-rank adapter algorithm, and the parameters of the low-rank adapter algorithm are obtained by training using a core image data set; A generation module for generating target core image data according to the initial core image data, the expansion direction information, and the target expansion image data.

8. The digital core image expansion generation device according to claim 7, wherein The target expansion model further includes: a tiled diffusion variational autoencoder and a local redrawing model; Correspondingly, the expansion module is specifically configured to: extract features of the initial core image data using the tiled diffusion variational autoencoder to obtain a latent vector corresponding to the initial core image data; the latent vector includes key features of the initial core image data, and the key features include at least one of the following: pore structure, permeability, and spatial relationship; use the target stable diffusion model to expand and generate the initial core image data along the specified expansion direction based on the latent vector to obtain preliminary expanded image data; use the local redrawing model to modify and redraw the preliminary expanded image data based on the expansion direction information and the initial core image data to obtain the target expansion image data.

9. A computer device, characterized in that, Including a processor and a memory for storing processor-executable instructions, and when the processor executes the instructions, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that, When the instructions are executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images