Dolomite multi-scale digital core reconstruction method and device, electronic equipment and medium

Abstract

The present application provides a dolomite multi-scale digital core reconstruction method, device, electronic equipment and medium, the method comprising: inputting the CT scanning two-dimensional slice image of the dolomite sample to be processed into the reconstruction model to obtain multi-view SEM style image and view angle parameter; inputting the style image and view angle parameter into the NeRF model to obtain a three-dimensional high-resolution image; the reconstruction model is obtained based on the following steps: adjusting the CT scanning two-dimensional slice image and SEM image of the dolomite sample according to a unified standard; inputting the CT scanning two-dimensional slice image and SEM image of the unified standard into the FastGAN network for image enhancement augmentation preprocessing to obtain a standard simulation image; training the generator in the CycleGAN network based on the standard simulation image to obtain the reconstruction model. The present application can solve the problem that the traditional digital core technology is limited by image noise and equipment cost, and it is difficult to realize the fusion of nanometer and micrometer multi-scale.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil and gas exploration, and particularly relates to a dolomite multi-scale digital core reconstruction method and device, electronic equipment and medium. BACKGROUND

[0002] Dolomite reservoirs are very important in the field of oil and gas exploration, but there are still many difficult questions about their genetic mechanism and reservoir pore evolution law. On the one hand, the formation of dolomite is affected by many factors such as sedimentary environment, diagenetic fluid source, burial depth and microbial activity, resulting in long-term controversy over its genetic mechanism. On the other hand, the present reservoir property of dolomite is the result of the combined action of various diagenetic processes such as dolomitization, dissolution, cementation and fracture on the pre-existing pores of the original sediment, and its reservoir space is mainly secondary pore, with various pore types and complex micro-pore throat structure characteristics. The pores, cavities and fractures in the reservoir match and superimpose each other due to their different causes and distribution laws, forming pore type, cavity type, fracture type and composite type. The strong heterogeneity of dolomite reservoirs makes it difficult to quantitatively analyze the micro-pore throat structure and three-dimensional connectivity by traditional characterization techniques such as rock thin section and mercury injection method, greatly increasing the difficulty of identifying the advantages and disadvantages of underground dolomite reservoirs, and thus affecting the evaluation of oil and gas reservoir properties. Exploring the genetic mechanism of dolomite reservoir pores requires comprehensive consideration of the comprehensive influence of tectonic-sedimentary-diagenetic factors. Conventional reservoir characterization techniques such as petrology and rock thin section observation, low-pressure gas adsorption method and high-pressure mercury injection method cannot quantitatively characterize the complex pore throat structure characteristics of dolomite and cannot provide quantitative calculation results of rock physical properties and three-dimensional pore throat information.

[0003] With the development and improvement of experimental instruments and testing techniques, the characterization of dolomite reservoir micro-pore structure has realized the transformation from qualitative to quantitative, from two-dimensional to three-dimensional, and from single scale to multi-scale, making it possible to characterize the internal structure and physical properties of the reservoir more finely. The resolution of CT scanning technology can reach microns, with the advantages of fast testing speed and no damage to the sample. It can not only obtain a three-dimensional pore network model of dolomite, but also calculate characterization parameters such as total porosity, pore throat radius, pore throat number, pore throat length and spatial connectivity of dolomite. Digital core technology has received widespread attention in the academic community due to its non-destructive, high-efficiency and high-precision characteristics. This technology scans the core sample by CT or electron microscope and converts it into digital data to simulate the real core structure, thereby quickly and intuitively analyzing the pore space structure of the rock. By extracting the pore network model and combining numerical simulation and artificial intelligence software, the micro-pore structure, fracture distribution, permeability characteristics and other reservoir parameters of the reservoir can be qualitatively and quantitatively analyzed.

[0004] However, the current CT scanning and digital core technology are affected by the limitations of scanning equipment and the instability of imaging conditions. The obtained rock images are disturbed by blur and noise, and the image resolution is low, which poses a challenge to the accuracy of subsequent image processing and analysis. The resolution of existing CT scanning technology is limited (micron level), and high-resolution CT equipment (such as synchrotron radiation source CT) is costly. Therefore, the traditional digital core technology is limited by image noise and equipment cost, and it is difficult to realize the multi-scale fusion of nanometers and microns in the field of digital core. SUMMARY

[0005] Therefore, it is necessary to provide a dolomite multi-scale digital core reconstruction method, device, electronic equipment and medium, to solve the technical problem that the traditional digital core technology is limited by image noise and equipment cost, and it is difficult to realize the multi-scale fusion of nanometers and microns in the field of digital core.

[0006] To solve the above problems, in a first aspect, the present application provides a dolomite multi-scale digital core reconstruction method, comprising:

[0007] inputting the to-be-processed CT scanning two-dimensional slice image of the dolomite sample into a reconstruction model to obtain a multi-view SEM style image and corresponding view angle parameters;

[0008] inputting the multi-view SEM style image and the corresponding view angle parameters into a preset NeRF model to obtain a three-dimensional high-resolution image;

[0009] The reconstruction model is obtained based on the following steps:

[0010] adjusting the CT scanning two-dimensional slice image and the SEM image of the dolomite sample according to a unified standard to obtain a unified standard CT scanning two-dimensional slice image and a SEM image;

[0011] inputting the unified standard CT scanning two-dimensional slice image and the SEM image into a trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image;

[0012] cyclically training a generator in a CycleGAN network based on the standard simulation image to obtain a reconstruction model.

[0013] In a possible implementation, before adjusting the CT scanning two-dimensional slice image and the SEM image of the dolomite sample according to a unified standard, it includes:

[0014] obtaining CT scanning three-dimensional data and SEM images of the dolomite sample;

[0015] The CT scan three-dimensional data is cut into two-dimensional images, and the two-dimensional images are normatively named to add perspective information to the two-dimensional images, to obtain a CT scan two-dimensional slice image of the dolomite sample.

[0016] In a possible implementation, the FastGAN network comprises a generator and a discriminator.

[0017] The generator of the FastGAN network is a ResNet structure, comprising a convolutional layer, a residual module and a deconvolutional layer.

[0018] In a possible implementation, the generator of the CycleGAN network comprises an encoder, a decoder, a skip connection layer arranged between the encoder and the decoder, and a preprocessing layer arranged at the front end of the encoder.

[0019] The preprocessing layer is configured to convert input image data into a tensor of a set dimension.

[0020] The encoder is configured to extract features from the tensor of the set dimension through a plurality of convolutional layers and down-sampling layers, and to connect and transmit the extracted features to corresponding layers of the decoder through the skip connection layer.

[0021] In a possible implementation, the loss function used in the training process of the generator of the CycleGAN network is a cycle consistency loss function.

[0022] In a possible implementation, a to-be-processed CT scan two-dimensional slice image of a dolomite sample is input into a reconstruction model to obtain a multi-perspective SEM style image and corresponding perspective parameters, comprising:

[0023] The to-be-processed CT scan two-dimensional slice image of the dolomite sample is input into the reconstruction model to obtain the multi-perspective SEM style image.

[0024] Based on the naming specification of the to-be-processed CT scan two-dimensional slice image, the multi-perspective SEM style image is named to obtain perspective parameters corresponding to the multi-perspective SEM style image.

[0025] In a possible implementation, the multi-perspective SEM style image and the corresponding perspective parameters are input into a preset NeRF model to obtain a three-dimensional high-resolution image, comprising:

[0026] The multi-perspective SEM style image and the corresponding perspective parameters are input into the preset NeRF model, and the multi-perspective SEM style image and the corresponding perspective parameters are calculated by means of neural radiation field to construct a three-dimensional density field.

[0027] After the three-dimensional density field is aligned with the two-dimensional slice image to be processed of the CT scan in space based on the voxel size and spatial range of the two-dimensional slice image to be processed of the CT scan, the three-dimensional density field and the two-dimensional slice image to be processed of the CT scan are weighted and fused in a voxel-by-voxel manner to obtain fused voxel data.

[0028] The fused voxel data is combined to obtain a final three-dimensional high-resolution image.

[0029] In a second aspect, the present application provides a dolomite multi-scale digital core reconstruction device, comprising:

[0030] An image conversion module is configured to input the two-dimensional slice image to be processed of the CT scan of the dolomite sample into a reconstruction model to obtain a multi-view SEM style image and corresponding view angle parameters.

[0031] An image fusion module is configured to input the multi-view SEM style image and the corresponding view angle parameters into a preset NeRF model to obtain a three-dimensional high-resolution image.

[0032] The reconstruction model is trained based on the following steps:

[0033] The CT scan two-dimensional slice image and the SEM image of the dolomite sample are adjusted according to a unified standard to obtain a unified standard CT scan two-dimensional slice image and a unified standard SEM image.

[0034] The unified standard CT scan two-dimensional slice image and the unified standard SEM image are input into the trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image.

[0035] The generator in the CycleGAN network is cyclically trained based on the standard simulation image to obtain a reconstruction model.

[0036] In a third aspect, the present application provides an electronic device comprising a memory and a processor, wherein

[0037] The memory is configured to store a program.

[0038] The processor is coupled to the memory and is configured to execute the program stored in the memory to implement the steps of the dolomite multi-scale digital core reconstruction method according to any one of the above aspects.

[0039] In a fourth aspect, the present application provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the dolomite multi-scale digital core reconstruction method according to any one of the above aspects.

[0040] The beneficial effects of the above implementation manner are that: the reconstruction model provided in the application first performs image enhancement augmentation preprocessing through the FastGAN network to obtain a standard simulation image in the training process, and then performs cyclic training on the generator in the CycleGAN network based on the standard simulation image, which performs image data augmentation, feature enhancement and cross-modal image style conversion based on the advantages of the generative adversarial network, so that the image data is more standard, the problems of poor image quality and insufficient data quantity are solved, the training time and the calculation resource consumption are significantly reduced under the premise of ensuring the image generation quality, and then the use of high-cost training equipment can be avoided.

[0041] In addition, the CycleGAN is used to realize the conversion of the low-resolution CT two-dimensional slice image into the high-resolution SEM style image, which can achieve the dual purposes of cross-modal style migration and resolution improvement, and solve the problem that the digital core technology is limited by image noise. In the process of dolomite multi-scale digital core reconstruction, the 3D reconstruction function of NeRF is used to directly realize multi-view fusion and efficiently generate high-resolution three-dimensional high-resolution images.

[0042] In the field of digital core technology, the pore structure features in the CT two-dimensional slice image belong to the micron level, and the pore structure features in the SEM style image belong to the nanometer level. The application can fuse the features of the SEM style image by reconstructing the CT two-dimensional slice image to obtain a high-resolution three-dimensional high-resolution image, realize multi-scale fusion of nanometers and microns, and thus solve the technical problem that the traditional digital core technology is limited by image noise and equipment cost and is difficult to realize multi-scale fusion of nanometers and microns in the field of digital core. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0044] Figure 1 The flowchart of one embodiment of the dolomite multi-scale digital core reconstruction method provided by the application;

[0045] Figure 2 The schematic diagram before and after image augmentation provided by the application;

[0046] Figure 3 The flowchart of another embodiment of the dolomite multi-scale digital core reconstruction method provided by the application;

[0047] Figure 4A generator structure schematic diagram of the FastGAN provided by the present application is shown in FIG. 1.

[0048] Figure 5 A generator structure schematic diagram of the CycleGAN provided by the present application is shown in FIG. 2.

[0049] Figure 6 A principle block diagram of one embodiment of the dolomite multi-scale digital core reconstruction device provided by the present application is shown in FIG. 3.

[0050] Figure 7 A structure schematic diagram of one embodiment of the electronic device provided by the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0052] In the description of the embodiments of the present application, the meaning of “a plurality of” is two or more, unless otherwise specified.

[0053] In the embodiments of the present application, the terms “comprising” and “having” and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product or equipment comprising a series of steps or modules does not have to be limited to the clearly listed steps or modules, but can include other steps or modules that are not clearly listed or inherent to these processes, methods, products or equipment.

[0054] The naming or numbering of the steps appearing in the embodiments of the present application does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The flow steps that have been named or numbered can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0055] In this document, the term “embodiment” means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0056] The present application provides a dolomite multi-scale digital core reconstruction method, device, electronic equipment and medium, which are described below respectively.

[0057] As Figure 1 shown, the present application provides a dolomite multi-scale digital core reconstruction method, comprising:

[0058] S101, input the CT scanning two-dimensional slice image of the dolomite sample to be processed into the reconstruction model, to obtain a multi-view SEM style image and corresponding view angle parameters.

[0059] It can be understood that CT scanning (computed tomography) is a medical imaging examination method that generates three-dimensional images of internal structures of the human body using X-rays and computer technology. The SEM style image, i.e. the scanning electron microscope image (SEM image), is a microscopic morphology image collected by a scanning electron microscope (SEM).

[0060] S102, input the multi-view SEM style image and the corresponding view angle parameters into the preset NeRF model to obtain a three-dimensional high-resolution image.

[0061] The reconstruction model is trained based on the following steps:

[0062] The CT scanning two-dimensional slice image and the SEM image of the dolomite sample are adjusted according to a unified standard to obtain a unified standard CT scanning two-dimensional slice image and a SEM image;

[0063] The unified standard CT scanning two-dimensional slice image and the SEM image are input into the trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image; the comparison chart before and after image augmentation is shown in Figure 2 ;

[0064] The generator in the CycleGAN network is cyclically trained based on the standard simulation image to obtain the reconstruction model.

[0065] It can be understood that the standard simulation image includes the preprocessed CT scanning two-dimensional slice image and the SEM image. The preprocessed CT scanning two-dimensional slice image is used as a training sample, and the SEM image is used as a training sample label. The generator in the CycleGAN network is cyclically trained.

[0066] Since the scanning electron microscope (SEM) has a resolution of nanometers (even sub-nanometers), it has the characteristics of clear imaging and diversified information. The present application combines the generation of adversarial neural networks (GAN) and neural radiation fields (NeRF), combines the advantages of high resolution of SEM, and proposes a dolomite multi-scale digital core reconstruction method.

[0067] First, image feature enhancement and image augmentation are performed using FastGAN, and then the generated images are used to train CycleGAN to convert low-resolution CT images (i.e., CT scan two-dimensional slice images) into high-resolution SEM images, and then NeRF (Instant-NGP) is used to fuse multi-view SEM style images to reconstruct high-resolution three-dimensional models. The generated high-resolution three-dimensional images are consistent with the global structure of the real CT, and the image resolution is increased by 8-16 times. This method can break through the limitations of resolution and scale in traditional digital core reconstruction, providing an efficient and low-cost solution for digital core modeling, and providing new ideas for current deep-ultra-deep carbonate reservoir pore network modeling, percolation mechanism analysis and reserve estimation.

[0068] The method provided by the application can be applied to a terminal device or a server side, and the method is realized by executing an application program on the terminal device or the server side.

[0069] The application performs image data augmentation, feature enhancement and cross-modal image style conversion on images based on the advantages of the generative adversarial network, standardizes the image data, and solves the problems of poor image quality and insufficient data.

[0070] In addition, the application directly realizes multi-view fusion and efficiently generates high-resolution three-dimensional high-resolution images by means of the 3D reconstruction function of NeRF (Instant-NGP).

[0071] Moreover, the application provides a low-cost, efficient and high-precision dolomite multi-scale digital core reconstruction method, which breaks through the resolution and scale limitations of traditional technology, realizes nanometer-micrometer-macro multi-level digital core modeling, and provides fine data support for reservoir evaluation and oil and gas development.

[0072] The method provided by the application mainly includes the following parts:

[0073] 1. Data preprocessing: including standardized collection and processing of existing dolomite low-resolution CT scan images (i.e., CT scan two-dimensional slice images) and high-resolution SEM image data;

[0074] 2. Image feature enhancement and augmentation: image feature enhancement processing and image data set expansion are performed on existing images by means of FastGAN network;

[0075] 3. Cross-modal image style conversion: low-resolution CT scan images are converted into high-resolution SEM style images by using CycleGAN network.

[0076] 4. Three-dimensional image reconstruction: using NeRF (Instant-NGP) to fuse multi-view SEM style images to generate three-dimensional high-resolution images, and then to construct multi-scale digital cores.

[0077] In some embodiments, before adjusting the CT scan two-dimensional slice images and SEM images of the dolomite sample according to the unified standard, comprising:

[0078] Obtaining CT scan three-dimensional data of the dolomite sample, and SEM images;

[0079] The CT scan three-dimensional data is divided into two-dimensional images, and the two-dimensional images are standardized named to add perspective information to the two-dimensional images, to obtain the CT scan two-dimensional slice images of the dolomite sample.

[0080] It can be understood that professional image analysis software is needed to first divide the three-dimensional CT scan data into two-dimensional images, and to standardize the naming of the images, which can clearly reflect the corresponding perspective information, facilitating the later three-dimensional reconstruction. The perspective information is the information that presents the influencing characteristics from different angles.

[0081] In some embodiments, the FastGAN network comprises a generator and a discriminator.

[0082] The generator of the FastGAN network is a ResNet structure, comprising a convolutional layer, a residual module and a deconvolutional layer.

[0083] It can be understood that the ResNet structure, which is the full name of residual network (Residual Networks), is a kind of deep convolutional neural network architecture. After inputting the low-resolution CT scan two-dimensional slice or high-resolution SEM image, the initial convolutional layer will preliminarily extract the features of the image, map the image from the spatial dimension to the feature dimension, preliminarily encode the features of the original image, and capture some basic edge, texture, hole and other features; then a series of residual blocks are followed, each residual block contains two convolutional layers, which retain part of the features of the original image through a jump connection, help to alleviate the gradient vanishing problem, and improve the feature extraction capability; then the feature map size is gradually restored to the original image size through the deconvolutional layer, and finally the converted image is output.

[0084] In some embodiments, the generator of the CycleGAN network comprises an encoder, a decoder, a jump connection layer arranged between the encoder and the decoder, and a preprocessing layer arranged at the front end of the encoder.

[0085] The preprocessing layer is used to convert the input image data into a tensor of a set dimension.

[0086] The encoder is configured to extract features from a tensor with a set dimension through a plurality of convolutional layers and down-sampling layers, and pass the extracted features to corresponding layers of the decoder through a skip connection layer.

[0087] It can be understood that the generator of the CycleGAN network adopts a U-Net structure, and the encoding path of the U-Net extracts features from the input through four convolutional layers and down-sampling layers, and passes the features extracted from each layer to corresponding layers of the decoding path through a skip connection.

[0088] In some embodiments, the loss function used in the training process of the generator of the CycleGAN network is a cycle consistency loss function.

[0089] It can be understood that the use of the cycle consistency loss function in the model training process enables the generated image to be restored to the original image, and the key information of the original image is not lost during style conversion, and the structure information of the original CT is preserved and SEM style details are injected.

[0090] In some embodiments, the CT scan two-dimensional slice image to be processed of the dolomite sample is input into the reconstruction model to obtain a multi-view SEM style image and corresponding view parameters, including:

[0091] The CT scan two-dimensional slice image to be processed of the dolomite sample is input into the reconstruction model to obtain a multi-view SEM style image.

[0092] Based on the naming specification of the CT scan two-dimensional slice image to be processed, the multi-view SEM style image is named to obtain the view parameters corresponding to the multi-view SEM style image.

[0093] It can be understood that the CT scan two-dimensional slice image to be processed (two-dimensional image in X / Y / Z direction) is input into the generator G L&H of the reconstruction model, and combined with the internally learned feature mapping relationship, an image with high-resolution SEM style features is generated. The generated image has significant improvement in details, clarity, and texture compared to the original low-resolution CT scan two-dimensional image, achieving style conversion with improved image resolution. The generated high-resolution SEM style image is retrieved from the computing device, and the generated high-resolution SEM style image is named according to the naming specification of the original CT scan slice, and the corresponding view parameters are set, providing a data basis for the NeRF model to reconstruct a three-dimensional density field from two-dimensional images.

[0094] In some embodiments, the multi-view SEM style image and the corresponding view parameters are input into a preset NeRF model to obtain a three-dimensional high-resolution image, including:

[0095] input the multi-view SEM style image and the corresponding view angle parameter into a preset NeRF model, calculate the multi-view SEM style image and the corresponding view angle parameter in a manner of neural radiance field, so as to construct a three-dimensional density field;

[0096] After the three-dimensional density field is spatially aligned with the to-be-processed CT scan two-dimensional slice image based on the voxel size and spatial range of the to-be-processed CT scan two-dimensional slice image, the three-dimensional density field and the to-be-processed CT scan two-dimensional slice image are weighted and fused voxel by voxel to obtain fused voxel data;

[0097] The fused voxel data are combined to obtain a final three-dimensional high-resolution image.

[0098] It can be understood that the three-dimensional density field data (the three-dimensional density field represents the density information of different positions in the scene, which corresponds to the microstructure reflected by the multi-view SEM style image) is extracted, and then the three-dimensional density field is adjusted (scaled, translated) according to the voxel size and spatial range of the original CT data (i.e. the to-be-processed CT scan two-dimensional slice image), so that it is spatially aligned with the original CT data, so as to ensure that they can be accurately matched when fused. Finally, the three-dimensional density field and the original CT data are weighted and fused voxel by voxel (to reflect the microstructure details, the weight of the three-dimensional density field can be appropriately increased). The fused voxel data are combined into new three-dimensional image data to generate a final three-dimensional high-resolution image.

[0099] In some embodiments, the method provided by the present application, as shown in Figure 3 includes the following specific implementation steps:

[0100] Step 1: Existing CT scan and SEM images of dolomite samples from the study area were collected at uniform size and contrast to ensure data consistency and usability. First, CT scan images were collected with a resolution of 200 μm and SEM images with a resolution of 50 nm. The dimensions of both the length and width of the images were adjusted to a uniform size of 256 pixels × 256 pixels to meet the FastGAN network's requirement for consistent input image size. This resulted in the collection of 60 standard CT scan images and 100 standard SEM images. For the X / Y / Z directions, specialized image analysis software was used to first segment the 3D CT scan data into 2D images (i.e., CT scan 2D slice images), and the images were then properly named to clearly reflect their corresponding viewpoint information, facilitating subsequent 3D reconstruction. Second, preliminary data augmentation was performed before inputting the data into the FastGAN network to enrich the data diversity. By performing operations such as random cropping, horizontal flipping, and random rotation of the images, we finally obtained 1800 CT scan images (i.e., CT scan 2D slice images) and 2000 SEM images. 80% of the data was used as the training set and 20% as the validation set.

[0101] Step 2: Input the low-resolution CT scan images (i.e., 2D slice images from CT scans) and high-resolution SEM images into the FastGAN network for image feature enhancement and image data augmentation. First, the unpaired training set data is used by the FastGAN network's generator (G... L G H ) and discriminator (D L D H Multiple rounds of training are performed until the FastGAN network model can generate graphs that achieve the expected results and run stably. Figure 4 As shown, the generator adopts a ResNet structure, which consists of multiple convolutional layers, residual blocks, and deconvolutional layers. After inputting a low-resolution CT scan 2D slice or a high-resolution SEM image, the initial convolutional layers perform preliminary feature extraction, mapping the image from a spatial dimension to a feature dimension, and initially encoding the features of the original image, capturing some basic features such as edges, textures, and holes. This is followed by a series of residual blocks, each containing two convolutional layers. Skip connections preserve some features of the original image, helping to alleviate the gradient vanishing problem and improve feature extraction capabilities. Then, deconvolutional layers gradually restore the feature map size to the original image size, finally outputting the transformed image. A discriminator is used to determine whether the input image is a real image or a generated image. Next, the images to be enhanced (low-resolution CT scan image and high-resolution SEM image) are input into the trained generator G, respectively. L and generator G H In the middle, generator GL with G H combined with the internal learning of the feature mapping relationship, an image with enhanced features is generated. Further, to increase the amount and diversity of data, the trained generator network is used for data augmentation operation. The image to be augmented is input into the generator multiple times, and different enhanced images are continuously generated according to different random noise vectors. The generated image retains the basic features of the original image while differing in details, textures, etc., thereby achieving data augmentation. Finally, 50,000 standard simulation images are simulated, i.e., standard simulation images.

[0102] Step 3: The generated standard simulation images (low-resolution CT scan images and high-resolution SEM images) are input into the CycleGAN network for generator G L&H cycle training (as shown in Figure 5 , the generator adopts a U-Net structure, which mainly consists of an encoder (Encoder), a decoder (Decoder), and a skip connection (Skip Connection). U-Net is a symmetrical network structure, with the left side being a downsampling process and the right side being an upsampling process. The features in the encoder are connected to the features in the decoder through a skip connection to retain feature information at different levels. The encoding path of U-Net extracts features from the input through four layers of convolution and downsampling, and transmits the features extracted at each layer to the corresponding layer of the decoding path through a skip connection. On the encoding path of U-Net, the preprocessing layer converts the image into a tensor with dimensions of (0, w 0, h 0, f 0). After preprocessing, the width w 0and height h 0of the tensor are halved in each downsampling block, while the feature dimension f 0is doubled. A cycle consistency loss function is also used in the model, which can restore the generated image back to the original image. While performing style conversion, the key information of the original image is not lost, and the structure information of the original CT is retained and SEM style details are injected. First, the CT scan two-dimensional slice image (two-dimensional image in X / Y / Z direction) to be converted (resolution improved) is input into the generator G L&HIn the specific implementation, the generated image with high-resolution SEM style features is retrieved from the computing device, and the generated high-resolution SEM style image is named according to a naming specification of the original CT scan two-dimensional slice image, and a corresponding viewing angle parameter is set. The NeRF model is provided with a data basis for reconstructing a three-dimensional density field from the two-dimensional image. In addition, the peak signal-to-noise ratio (PSNR) and the structural similarity (SSIM) are used as quantitative evaluation indexes to verify the effectiveness of the resolution improvement.

[0103] The calculation formula of the PSNR is as follows:

[0104] (1)

[0105] wherein, MAX CT is a maximum pixel value in the CT, MSE is a mean square error. The higher the PSNR value is, the higher the similarity between the pseudo CT image and the real CT image is.

[0106] The calculation formula of the SSIM is as follows:

[0107] (2)

[0108] wherein, x and y are average values of the real CT two-dimensional slice image and the real SEM image, respectively, L&H after the generator G μ x and μ y x and y σ x and σ y x and y σ xy x and y c 1 and c ​​​​​​2 are two constants used to maintain stability. The final image fidelity and resolution evaluation index values in this embodiment are shown in Table 1. The SSIM value ranges from [-1, 1], and the larger the value, the higher the similarity between the two images.

[0109] Table 1: Final image fidelity and resolution evaluation index values

[0110]

[0111] Step 4: Input the generated multi-view SEM style images (different angle virtual slices generated by CycleGAN) and their corresponding view parameters into the NeRF (Instant-NGP) model, start the training process of Instant-NGP, and accelerate the training through the hash coding of Instant-NGP. The model will gradually build a three-dimensional density field according to the input image and view information through the calculation of neural radiance field. When the NeRF (Instant-NGP) training is completed, extract the three-dimensional density field data from the trained model (the three-dimensional density field represents the density information of different positions in the scene, corresponding to the microstructure reflected by the multi-view SEM style image). Then adjust (scale, translate) the three-dimensional density field according to the voxel size and spatial range of the original CT data, so that it is spatially aligned with the original CT data, to ensure accurate matching when the two are fused. Finally, perform weighted fusion calculation on the three-dimensional density field and the original CT data voxel by voxel (to reflect the microstructure details, the weight of the three-dimensional density field can be appropriately increased). Combine the fused voxel data into a new three-dimensional image data to generate the final three-dimensional high-resolution image. This method provides an efficient and low-cost solution for deep carbonate reservoir digital core modeling.

[0112] (3)

[0113] V final : final fused three-dimensional high-resolution image (or voxel data); α : fusion weight, value range [0, 1], α = 0.3 means more emphasis on NeRF details, but retains the global structure of CT; V CT : original or preprocessed low-resolution three-dimensional CT volume data; VNeRF: three-dimensional model reconstructed from multi-view SEM style images by neural radiance field (NeRF).

[0114] In summary, the key points of the present application include:

[0115] 1. A lightweight and efficient generative adversarial network (FastGAN network) is used to train a low-resolution CT scan and a high-resolution SEM image to generate a diversified synthetic image expansion dataset, which significantly reduces the training time and computational resource consumption under the premise of ensuring the generation quality;

[0116] 2. CycleGAN is used to realize the process of converting low-resolution CT images into high-resolution SEM style images, achieving the dual purposes of cross-modal style transfer and resolution enhancement;

[0117] 3. Based on the 3D reconstruction function of NeRF (Instant-NGP), the multi-view converted high-resolution SEM style image is ingeniously accelerated and fused to generate a three-dimensional high-resolution image;

[0118] 4. CycleGAN and NeRF are combined to realize the closed-loop reconstruction of "low-resolution CT→high-resolution SEM→three-dimensional model", and complete the cross-modal and cross-scale joint modeling.

[0119] As shown in Figure 6 The present application also provides a dolomite multi-scale digital core reconstruction device 600, which comprises:

[0120] An image conversion module 601 is used to input the CT scan two-dimensional slice image of the dolomite sample to be processed into a reconstruction model to obtain multi-view SEM style images and corresponding viewing angle parameters;

[0121] An image fusion module 602 is used to input the multi-view SEM style images and corresponding viewing angle parameters into a preset NeRF model to obtain a three-dimensional high-resolution image;

[0122] The reconstruction model is trained based on the following steps:

[0123] The CT scan two-dimensional slice image and the SEM image of the dolomite sample are adjusted according to a unified standard to obtain a unified standard CT scan two-dimensional slice image and a SEM image;

[0124] The unified standard CT scan two-dimensional slice image and the SEM image are input into the trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image;

[0125] The generator in the CycleGAN network is cyclically trained based on the standard simulation image to obtain a reconstruction model.

[0126] The dolostone multi-scale digital core reconstruction device provided by the above embodiment can implement the technical solutions described in the dolostone multi-scale digital core reconstruction method embodiment, and the principles of implementation of the above modules or units can be referred to the corresponding content in the dolostone multi-scale digital core reconstruction method embodiment, which will not be repeated here.

[0127] As shown in Figure 7 The present application also provides an electronic device 700 accordingly. The electronic device 700 includes a processor 701, a memory 702 and a display 703. Figure 7 Only part of the components of the electronic device 700 are shown, but it should be understood that all the components shown are not required, and more or less components can be implemented instead.

[0128] The memory 702 can be an internal storage unit of the electronic device 700 in some embodiments, such as a hard disk or a memory of the electronic device 700. The memory 702 can also be an external storage device of the electronic device 700 in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0129] Further, the memory 702 can include both an internal storage unit and an external storage device of the electronic device 700. The memory 702 is used to store application software and various data installed in the electronic device 700.

[0130] The processor 701 can be a central processing unit (CPU), a microprocessor or other data processing chip in some embodiments, used to run program codes or process data stored in the memory 702, such as the dolostone multi-scale digital core reconstruction method in the present application.

[0131] The display 703 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc. in some embodiments. The display 703 is used to display information of the electronic device 700 and to display a visualized user interface. The components 701-703 of the electronic device 700 communicate with each other through a system bus.

[0132] In some embodiments of the present application, when the processor 701 executes the dolostone multi-scale digital core reconstruction program in the memory 702, the following steps can be implemented:

[0133] The CT scanning two-dimensional slice image of the dolomite sample to be processed is input into the reconstruction model to obtain a multi-view SEM style image and corresponding view angle parameters;

[0134] The multi-view SEM style image and the corresponding view angle parameters are input into a preset NeRF model to obtain a three-dimensional high-resolution image.

[0135] The reconstruction model is obtained based on the following steps:

[0136] The CT scanning two-dimensional slice image and the SEM image of the dolomite sample are adjusted according to a unified standard to obtain a unified standard CT scanning two-dimensional slice image and a unified standard SEM image.

[0137] The unified standard CT scanning two-dimensional slice image and the unified standard SEM image are input into the trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image.

[0138] The generator in the CycleGAN network is cyclically trained based on the standard simulation image to obtain the reconstruction model.

[0139] It should be understood that, in addition to the above functions, the processor 701 can also implement other functions when executing the dolomite multi-scale digital core reconstruction program in the memory 702. For details, refer to the description of the corresponding method embodiments.

[0140] Further, the type of the electronic device 700 is not specifically limited, and the electronic device 700 can be a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop, or the like. Exemplary embodiments of the portable electronic device include, but are not limited to, a portable electronic device running an IOS, android, microsoft, or other operating system. The above-mentioned portable electronic device can also be other portable electronic devices, such as a laptop with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present application, the electronic device 700 can not be a portable electronic device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0141] In another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the dolomite multi-scale digital core reconstruction method provided by the above-mentioned methods, and the method comprises:

[0142] The CT scanning two-dimensional slice image of the dolomite sample to be processed is input into the reconstruction model to obtain a multi-view SEM style image and corresponding view angle parameters;

[0143] The multi-view SEM style image and the corresponding view angle parameters are input into a preset NeRF model to obtain a three-dimensional high-resolution image;

[0144] The reconstruction model is obtained based on the following steps:

[0145] The CT scanning two-dimensional slice image and the SEM image of the dolomite sample are adjusted according to a unified standard to obtain a unified standard CT scanning two-dimensional slice image and a unified standard SEM image;

[0146] The unified standard CT scanning two-dimensional slice image and the unified standard SEM image are input into the trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image;

[0147] The generator in the CycleGAN network is cyclically trained based on the standard simulation image to obtain the reconstruction model.

[0148] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. The computer readable storage medium includes a magnetic disk, an optical disk, a read-only memory, a random access memory, etc.

[0149] The dolomite multi-scale digital core reconstruction method, device, electronic equipment and medium provided by the present application are described in detail above, and the principles and implementation modes of the present application are described by applying specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed; in summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A dolostone multiscale digital core reconstruction method, characterized in that, The application relates to a method for reconstructing a dolomite sample, comprising the following steps: inputting a CT scanning two-dimensional slice image of a dolomite sample to be processed into a reconstruction model to obtain a multi-view SEM style image and corresponding view parameters; inputting the multi-view SEM style image and the corresponding view parameters into a preset NeRF model to obtain a three-dimensional high-resolution image; wherein the reconstruction model is obtained by training based on the following steps: adjusting a CT scanning two-dimensional slice image and a SEM image of a dolomite sample according to a unified standard to obtain a CT scanning two-dimensional slice image and a SEM image of the dolomite sample according to the unified standard; inputting the CT scanning two-dimensional slice image and the SEM image of the dolomite sample according to the unified standard into a trained FastGAN network for image enhancement and augmentation preprocessing to obtain a standard simulation image; cyclically training a generator in a CycleGAN network based on the standard simulation image to obtain the reconstruction model; in the training process of the generator of the CycleGAN network, a loss function used is a cycle consistency loss function; the generator of the CycleGAN network comprises an encoder, a decoder, a skip connection layer arranged between the encoder and the decoder, and a preprocessing layer arranged at the front end of the encoder; the preprocessing layer is used for converting input image data into a tensor of a set dimension; the encoder is used for extracting features of the tensor of the set dimension through a plurality of convolution layers and a down-sampling layer, and connecting and transmitting the extracted features to corresponding layers of the decoder through the skip connection layer to generate an image with high-resolution SEM style features in combination with an internally learned feature mapping relationship; the resolution of the image with high-resolution SEM style features is verified by using a peak signal-to-noise ratio and a structural similarity quantitative evaluation index.

2. The dolostone multi-scale digital core reconstruction method of claim 1, wherein, Before adjusting a CT scanning two-dimensional slice image and a SEM image of a dolomite sample according to a unified standard, the following steps are included: obtaining CT scanning three-dimensional data and a SEM image of the dolomite sample; cutting the CT scanning three-dimensional data into two-dimensional images, and performing standard naming on the two-dimensional images to add view information to the two-dimensional images to obtain a CT scanning two-dimensional slice image of the dolomite sample.

3. The dolostone multi-scale digital core reconstruction method of claim 1, wherein, the FastGAN network comprises a generator and a discriminator; the generator of the FastGAN network is of a ResNet structure and comprises a convolution layer, a residual module and a deconvolution layer.

4. The dolostone multi-scale digital core reconstruction method of claim 1, wherein, inputting a CT scanning two-dimensional slice image of a dolomite sample to be processed into a reconstruction model to obtain a multi-view SEM style image and corresponding view parameters, comprising: inputting a CT scanning two-dimensional slice image of a dolomite sample to be processed into a reconstruction model to obtain a multi-view SEM style image; naming the multi-view SEM style image based on a naming specification of the CT scanning two-dimensional slice image to be processed to obtain view parameters corresponding to the multi-view SEM style image.

5. The dolostone multi-scale digital core reconstruction method according to any one of claims 1-4, characterized in that, inputting the multi-view SEM style image and the corresponding view parameters into a preset NeRF model to obtain a three-dimensional high-resolution image, comprising: input the multi-view SEM style image and the corresponding view angle parameter into a preset NeRF model, calculate the multi-view SEM style image and the corresponding view angle parameter in the form of neural radiation field, so as to construct a three-dimensional density field; align the three-dimensional density field with the to-be-processed CT scan two-dimensional slice image in space based on the voxel size and spatial range of the to-be-processed CT scan two-dimensional slice image, and then perform weighted fusion calculation on the three-dimensional density field and the to-be-processed CT scan two-dimensional slice image voxel by voxel to obtain fused voxel data; combine the fused voxel data to obtain a final three-dimensional high-resolution image.

6. A dolostone multiscale digital core reconstruction device, characterized in that, comprise: an image conversion module configured to input a to-be-processed CT scan two-dimensional slice image of a dolomite sample into a reconstruction model to obtain a multi-view SEM style image and corresponding view angle parameters; an image fusion module configured to input the multi-view SEM style image and the corresponding view angle parameter into a preset NeRF model to obtain a three-dimensional high-resolution image; wherein the reconstruction model is obtained based on the following steps: adjusting CT scan two-dimensional slice images and SEM images of a dolomite sample according to a unified standard to obtain unified standard CT scan two-dimensional slice images and SEM images; inputting the unified standard CT scan two-dimensional slice images and SEM images into a trained FastGAN network for image enhancement and augmentation preprocessing to obtain standard simulation images; recurrently training a generator in a CycleGAN network based on the standard simulation images to obtain a reconstruction model; wherein the training process of the generator of the CycleGAN network adopts a cycle consistency loss function as a loss function; the generator of the CycleGAN network comprises an encoder, a decoder, a skip connection layer arranged between the encoder and the decoder, and a preprocessing layer arranged at the front end of the encoder; the preprocessing layer is configured to convert input image data into a tensor of a set dimension; the encoder is configured to extract features from the tensor of the set dimension through multiple convolution layers and down-sampling layers, and pass the extracted features to the corresponding layers of the decoder through the skip connection layer to combine the internal learned feature mapping relationship to generate an image with high-resolution SEM style features; the resolution of the image with high-resolution SEM style features is verified by using a peak signal-to-noise ratio and a structural similarity quantitative evaluation index.

7. An electronic device, comprising: comprise a memory and a processor, wherein the memory is configured to store a program; the processor is coupled to the memory and is configured to execute the program stored in the memory to implement the steps of the dolomite multi-scale digital core reconstruction method according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the dolomite multi-scale digital core reconstruction method according to any one of claims 1 to 5.

Images