A digital rock reconstruction method, device, equipment and medium

Abstract

The application discloses a digital rock reconstruction method and device, equipment and medium, and relates to the field of digital rock reconstruction, and comprises the following steps: extracting micro-texture features in two-dimensional scanning electron microscope image data to obtain corresponding texture feature vectors, and performing integration processing on the texture feature vectors to obtain target texture feature vectors; extracting structural feature information from three-dimensional computer tomography image data, splicing the target texture feature vectors and target noise, and mapping the spliced vectors into target style vectors; fusing the structural feature information and the target style vectors to obtain a three-dimensional feature map, capturing geometric feature responses corresponding to local geometric structures of rocks in the three-dimensional feature map, and generating spatial modulation weights matched with the local geometric structures of the rocks; performing voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weights to obtain a target feature map, and generating target three-dimensional digital rock body data based on the target feature map.

Description

Technical Field

[0001] This application relates to the field of digital rock reconstruction, and in particular to a digital rock reconstruction method, apparatus, equipment and medium. Background Technology

[0002] Three-dimensional digital rock reconstruction is a core foundational technology for rock physics analysis, seepage mechanism simulation, and oil and gas reservoir evaluation. Its core objective is to construct a digital volume model that can highly reproduce the real pore topology, spatial connectivity characteristics, and micro-texture details of rocks based on limited experimental observation data. High-quality reconstruction models are key to improving the accuracy of rock property prediction and ensuring the reliability of multi-physics numerical simulations. It has irreplaceable application value in fields such as oil and gas resource exploration and development and unconventional energy development.

[0003] With the development of deep learning technology, using deep learning models such as Generative Adversarial Networks (GANs) for 3D digital rock reconstruction has become a mainstream research direction in the field. Existing techniques attempt to combine 3D computed tomography (CT) image data, which characterizes the macroscopic skeleton of rocks, with 2D scanning electron microscope (SEM) image data, which characterizes microscopic textures. Through cross-scale and cross-dimensional data complementarity, this approach balances the spatial continuity of the reconstruction model with the resolution of microscopic representations.

[0004] However, existing multimodal fusion-based 3D digital rock reconstruction technologies still have significant technical shortcomings in practical applications, making it difficult to meet the requirements for constructing high-precision digital rock models. These shortcomings are mainly reflected in two aspects: First, existing cross-dimensional multimodal feature fusion mechanisms often lack the ability to extract and semantically align SEM texture information along the three-dimensional orthogonal axis, making it easy for high-resolution textures to exhibit decoupling biases in 3D spatial mapping, making it difficult to realistically reproduce the complex microscopic heterogeneity and anisotropic distribution inside the rock. Second, in existing digital rock reconstruction research, the introduction methods are mostly based on global or channel-level feature modulation, lacking explicit adaptive modulation mechanisms for differences in 3D spatial location. In scenarios with complex pore structures, it is difficult to dynamically guide the spatial distribution of texture information based on the local geometric features of the rock, resulting in limited collaborative representation capabilities between microscopic texture details and the macroscopic 3D skeleton, affecting the structural consistency and realism of the reconstructed model. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a digital rock reconstruction method, apparatus, device, and medium that achieves deep adaptive coupling of multimodal features, thereby improving the fidelity of digital rock reconstruction. The specific scheme is as follows:

[0006] In a first aspect, this application provides a digital rock reconstruction method, comprising:

[0007] Acquire multimodal data; the multimodal data includes three-dimensional computed tomography image data for characterizing the macroscopic framework of rocks, and two-dimensional scanning electron microscope image data from different orthogonal directions for characterizing the microscopic texture of rocks;

[0008] Microscopic texture features from different orthogonal directions are extracted from the two-dimensional scanning electron microscope image data to obtain the corresponding texture feature vectors. The texture feature vectors are then integrated to obtain the target texture feature vector.

[0009] Structural feature information is extracted from the three-dimensional computed tomography image data. The target texture feature vector is concatenated with the target noise, and the concatenated vector is mapped to the target style vector based on a style mapping network. The style mapping network is a mapping network constructed based on a multi-layer fully connected network.

[0010] The structural feature information and the target style vector are fused to obtain a three-dimensional feature map. The three-dimensional feature map is then input into a spatial adaptive feature modulation module to capture the geometric feature responses corresponding to the local geometry of the rock at different levels in the three-dimensional feature map. The geometric feature responses are analyzed based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock.

[0011] The three-dimensional feature map is subjected to voxel-by-voxel weighted modulation according to the spatial modulation weight to obtain the target feature map. The target feature map is then upsampled layer by layer to generate the target three-dimensional digital rock mass data.

[0012] Optionally, the digital rock reconstruction method further includes:

[0013] The acquired two-dimensional scanning electron microscope image data is randomly divided into blocks for sampling, so as to crop out several image blocks from the two-dimensional scanning electron microscope image data;

[0014] The acquired 3D computed tomography (CT) image data is subjected to data normalization and size alignment processing in order to extract structural feature information from the processed 3D CT image data; the size alignment processing includes spatial dimension alignment and physical resolution alignment.

[0015] Optionally, the step of extracting microscopic texture features from different orthogonal directions in the two-dimensional scanning electron microscope image data to obtain corresponding texture feature vectors, and integrating the texture feature vectors to obtain the target texture feature vector, includes:

[0016] The image blocks with different orthogonal directions are input into a pre-set two-dimensional feature encoding branch, and the image blocks are encoded based on the two-dimensional feature encoding branch through a shared or parallel feature extraction network to obtain texture feature vectors corresponding to each orthogonal direction; the feature extraction network includes convolutional layers and nonlinear mapping units;

[0017] When encoding the image patch using a shared feature extraction network, the image patch shares the same set of feature extraction networks. When encoding the image patch using a parallel feature extraction network, a set of feature extraction networks with the same structure is configured for image patches in different orthogonal directions, and multiple sets of feature extraction networks work in parallel and independently.

[0018] The texture feature vectors from different orthogonal directions are integrated to obtain the target texture feature vector; the target texture feature vector is used to describe the texture statistical characteristics of rock material in different spatial directions.

[0019] Optionally, the step of concatenating the target texture feature vector with the target noise, and mapping the concatenated vector to a target style vector based on a style mapping network, includes:

[0020] The target texture feature vector is concatenated with a random noise vector, and the concatenated vector is input into a style mapping network to map the concatenated vector into a target style vector.

[0021] Optionally, the step of using the spatial adaptive feature modulation module to capture geometric feature responses corresponding to the local geometry of the rock at different levels in the three-dimensional feature map, and analyzing the geometric feature responses based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock, includes:

[0022] The multi-scale spatial perception operator built into the spatial adaptive feature modulation module is used to capture the geometric feature responses corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map.

[0023] The multi-scale spatial perception operator performs feature analysis and numerical calculation on the captured geometric feature responses based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock.

[0024] Optionally, the step of performing voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weights to obtain the target feature map includes:

[0025] The spatial modulation weights are upsampled to the same target resolution as the three-dimensional feature map, and the upsampled spatial modulation weights are multiplied with the three-dimensional feature map on a voxel-by-voxel basis to obtain the target feature map.

[0026] The target feature map is obtained by multiplying the value of each voxel in the three-dimensional feature map by the weight coefficient at the corresponding position.

[0027] Optionally, the digital rock reconstruction method further includes:

[0028] The generated target three-dimensional digital rock mass data is compared with real rock sample data or preset statistical constraints, and the corresponding loss function is calculated. The loss function includes adversarial loss and statistical constraint loss. The adversarial loss is used to constrain the consistency between the generated result and the real sample distribution, and the statistical constraint loss is used to constrain the consistency between the pore structure characteristics of the generated result and the macroscopic skeleton of the rock.

[0029] The network parameters of the style mapping network, the two-dimensional feature encoding branch, and the generator network are updated based on the loss function.

[0030] Secondly, this application provides a digital rock reconstruction apparatus, comprising:

[0031] The data acquisition module is used to acquire multimodal data, including three-dimensional computed tomography image data for characterizing the macroscopic skeleton of rocks, and two-dimensional scanning electron microscope image data from different orthogonal directions for characterizing the microscopic texture of rocks.

[0032] The first vector determination module is used to extract microscopic texture features from different orthogonal directions in the two-dimensional scanning electron microscope image data, obtain the corresponding texture feature vectors, and integrate the texture feature vectors to obtain the target texture feature vector.

[0033] The second vector determination module is used to extract structural feature information from the three-dimensional computed tomography image data, concatenate the target texture feature vector with the target noise, and map the concatenated vector to the target style vector based on a style mapping network; the style mapping network is a mapping network constructed based on a multilayer fully connected network.

[0034] The weight generation module is used to fuse the structural feature information and the target style vector to obtain a three-dimensional feature map. The three-dimensional feature map is input into the spatial adaptive feature modulation module to capture the geometric feature responses corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map. The geometric feature responses are analyzed based on adaptive pooling and convolution mapping techniques to generate spatial modulation weights that match the local geometric structure of the rock.

[0035] The digital rock reconstruction module is used to perform voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weight to obtain the target feature map, and to perform layer-by-layer upsampling on the target feature map to generate target three-dimensional digital rock body data.

[0036] Thirdly, this application provides an electronic device, comprising:

[0037] Memory, used to store computer programs;

[0038] A processor is used to execute the computer program to implement the aforementioned digital rock reconstruction method.

[0039] Fourthly, this application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned digital rock reconstruction method.

[0040] In this application, multimodal data is acquired; the multimodal data includes three-dimensional computed tomography (CT) image data for characterizing the macroscopic framework of rocks, and two-dimensional scanning electron microscope (SEM) image data from different orthogonal directions for characterizing the microscopic texture of rocks; microscopic texture features from different orthogonal directions are extracted from the two-dimensional SEM image data to obtain corresponding texture feature vectors, and the texture feature vectors are integrated to obtain a target texture feature vector; structural feature information is extracted from the three-dimensional CT image data, the target texture feature vector is concatenated with target noise, and the concatenated vector is mapped to a target style vector based on a style mapping network; the style mapping network... A mapping network based on a multi-layer fully connected network is used. The structural feature information and the target style vector are fused to obtain a three-dimensional feature map. The three-dimensional feature map is input into a spatial adaptive feature modulation module to capture the geometric feature responses corresponding to the local geometry of the rock at different levels in the three-dimensional feature map. The geometric feature responses are analyzed based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock. The three-dimensional feature map is then subjected to voxel-by-voxel weighted modulation according to the spatial modulation weights to obtain a target feature map. The target feature map is then upsampled layer by layer to generate target three-dimensional digital rock mass data. As can be seen from the above, on the one hand, this application obtains two-dimensional SEM image data in different orthogonal directions, extracts micro-texture features from them to obtain corresponding texture feature vectors, integrates the texture feature vectors into a target texture feature vector, and then concatenates the target texture feature vector with target noise and maps it to a target style vector through a style mapping network constructed by a multi-layer fully connected network. This achieves accurate extraction and unified semantic mapping of SEM texture features in different orthogonal axes, ensuring semantic consistency of texture information in each direction within the three-dimensional generation framework. It effectively solves the decoupling bias problem of high-resolution textures in three-dimensional spatial mapping, and realistically restores the micro-heterogeneity and anisotropic distribution inside the rock. On the other hand, this application fuses structural feature information with target style vectors to obtain a three-dimensional feature map, and then inputs it into a spatial adaptive feature modulation module. This module captures the geometric feature responses corresponding to the local geometric structure of the rock at different levels of the three-dimensional feature map, and generates matching spatial modulation weights based on adaptive pooling and convolutional mapping techniques. Then, it uses these weights to perform voxel-by-voxel weighted modulation on the three-dimensional feature map, realizing the sensitive response of the generation process to the differences in three-dimensional spatial position. It can dynamically guide the spatial distribution of texture information according to the local geometric features of the rock, improve the collaborative representation accuracy of micro-texture details and macro-three-dimensional skeleton, and ensure the structural consistency and realism of the reconstructed model. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0042] Figure 1 This is a flowchart of a digital rock reconstruction method disclosed in this application;

[0043] Figure 2 This is a schematic diagram illustrating the data flow and operation process of a spatial adaptive feature modulation module disclosed in this application;

[0044] Figure 3 This is a schematic diagram of the structure of a digital rock reconstruction device disclosed in this application;

[0045] Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in this application. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0047] Existing cross-dimensional multimodal feature fusion mechanisms lack the ability to extract and semantically align SEM texture information along three-dimensional orthogonal axes, making it difficult to realistically reproduce the complex microscopic heterogeneity and anisotropic distribution within rocks. Furthermore, current digital rock reconstruction research lacks explicit adaptive modulation mechanisms for differences in three-dimensional spatial location. In scenarios with complex porous structures, it is difficult to dynamically guide the spatial distribution of texture information based on local rock geometric features, resulting in limited collaborative representation capabilities between microscopic texture details and the macroscopic three-dimensional skeleton. Therefore, this application provides a digital rock reconstruction method that, by constructing a spatially aware modulation mechanism, achieves deep adaptive coupling of multimodal features, thereby improving the fidelity of digital rock reconstruction.

[0048] See Figure 1 As shown in the figure, this application discloses a digital rock reconstruction method, including:

[0049] Step S11: Acquire multimodal data; the multimodal data includes three-dimensional computed tomography image data for characterizing the macroscopic skeleton of rocks, and two-dimensional scanning electron microscope image data from different orthogonal directions for characterizing the microscopic texture of rocks.

[0050] In this embodiment, firstly, multi-source heterogeneous input data for digital rock reconstruction is acquired. This multi-source heterogeneous input data includes at least a set of three-dimensional CT image data for characterizing the macroscopic structure information of the rock, and a set of two-dimensional SEM image data for characterizing the microscopic texture information of the rock. The SEM image data can come from different orthogonal directions.

[0051] Furthermore, the acquired two-dimensional scanning electron microscope (SEM) image data can be randomly segmented and sampled to extract several image blocks. The acquired three-dimensional computed tomography (CT) image data undergoes data normalization and size alignment processing to extract structural feature information; the size alignment processing includes, but is not limited to, spatial dimension alignment and physical resolution alignment.

[0052] Understandably, preprocessing operations are performed on 3D CT image data and 2D SEM image data to address the differences in resolution and data dimensionality between different modalities. Specifically, for 3D CT image data, volume data normalization and size alignment are performed to form the 3D structure input for subsequent generation stages. For 2D SEM image data, random block sampling is performed according to a preset sampling strategy, cropping the original high-resolution SEM image into several representative image blocks to reduce the input scale and enhance the diversity of texture samples.

[0053] Step S12: Extract microscopic texture features from different orthogonal directions from the two-dimensional scanning electron microscope image data to obtain the corresponding texture feature vectors, and integrate the texture feature vectors to obtain the target texture feature vector.

[0054] In this embodiment, microscopic texture features from different orthogonal directions are extracted from two-dimensional scanning electron microscope image data to obtain corresponding texture feature vectors. These texture feature vectors are then integrated to obtain a target texture feature vector. This process can include: first, inputting image blocks from different orthogonal directions into a pre-set two-dimensional feature encoding branch; then encoding the image blocks using a shared or parallel feature extraction network based on the two-dimensional feature encoding branch to obtain texture feature vectors corresponding to each orthogonal direction; the feature extraction network includes, but is not limited to, convolutional layers and nonlinear mapping units; when encoding image blocks using a shared feature extraction network, the image blocks share the same set of feature extraction networks; when encoding image blocks using parallel feature extraction networks, a set of feature extraction networks with the same structure is configured for image blocks from different orthogonal directions, and multiple sets of feature extraction networks work independently in parallel. Then, the texture feature vectors from different orthogonal directions are integrated to obtain a unified target texture feature vector; the target texture feature vector is used to describe the texture statistical characteristics of rock materials in different spatial directions.

[0055] Step S13: Extract structural feature information from the three-dimensional computed tomography image data, concatenate the target texture feature vector with the target noise, and map the concatenated vector to the target style vector based on the style mapping network; the style mapping network is a mapping network constructed based on a multi-layer fully connected network.

[0056] In this embodiment, 3D CT image data can be input into the structural path of the generation network in voxel form to extract structural features used to constrain the 3D generation process, such as the overall topological framework and spatial connectivity information of rock pores. These structural features participate in the calculation as 3D tensors during the generation process and are continuously used across multiple generation levels to ensure that the generated result maintains spatial consistency with the original CT data.

[0057] Furthermore, the target texture feature vector can be concatenated with the target noise, and the concatenated vector can be mapped to the target style vector using a style mapping network. Specifically, this can include concatenating the target texture feature vector with a random noise vector, and then inputting the concatenated vector into a style mapping network to map the concatenated vector to the target style vector.

[0058] The above method concatenates the target texture feature vector with a random noise vector to form a joint conditional vector; the random noise vector is used to introduce randomness into the generation process. The joint conditional vector is then input into a style mapping network, which maps the input joint conditional vector to an intermediate style space and outputs the corresponding target style vector. This target style vector is used to uniformly represent data features from different modalities and serves as an important parameter controlling the 3D generation process.

[0059] During the mapping process, normalization can be performed on the input or intermediate features to maintain the numerical stability of the style space.

[0060] Step S14: The structural feature information and the target style vector are fused to obtain a three-dimensional feature map. The three-dimensional feature map is then input into a spatial adaptive feature modulation module to capture the geometric feature responses corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map. The geometric feature responses are analyzed based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometric structure of the rock.

[0061] In this embodiment, after obtaining the target style vector, it can be fused with the 3D structural features obtained in step S13 within the generative network. Specifically, in multiple layers of the generative network, the target style vector participates in feature modulation through linear mapping or affine transformation, while the 3D structural features serve as spatial inputs for convolution or deconvolution operations. This approach ensures that the generation process maintains 3D structural consistency while being jointly controlled by multimodal conditional features.

[0062] Furthermore, the spatial adaptive feature modulation module is used to capture geometric feature responses corresponding to the local geometry of the rock at different levels in the 3D feature map. Based on adaptive pooling and convolutional mapping techniques, the geometric feature responses are analyzed to generate spatial modulation weights that match the local geometry of the rock. This process can include: first, using the multi-scale spatial awareness operator built into the spatial adaptive feature modulation module to capture geometric feature responses corresponding to the local geometry of the rock at different levels in the 3D feature map; then, using the multi-scale spatial awareness operator based on adaptive pooling and convolutional mapping techniques, feature analysis and numerical calculations are performed on the captured geometric feature responses to generate spatial modulation weights that match the local geometry of the rock.

[0063] like Figure 2As shown, in the 3D generation stage, this embodiment introduces a spatial adaptive feature modulation module to dynamically optimize the intermediate layer feature map. This module uses the 3D feature map output by the generation network as input, first extracting spatial features at different levels (scale 1, scale 2, and scale 3). These multi-scale spatial features are then fed into a multi-scale spatial perception processing stage. The module's built-in multi-scale spatial perception operator performs feature analysis and numerical calculations based on adaptive pooling and convolutional mapping techniques, generating spatial adaptive modulation parameters, i.e., spatial modulation weights, that match the local geometry of the rock. Then, a spatial feature fusion stage integrates the multi-scale spatial features with the spatial adaptive modulation parameters. Next, feature modulation is performed on the original 3D intermediate feature map, upsampling the spatial adaptive modulation parameters to the target resolution and then performing voxel-by-voxel weighted modulation with the 3D intermediate feature map. Finally, the modulated target feature map with spatial adaptive dynamic optimization is output, achieving a sensitive response to 3D spatial positional differences in the generation process. By deploying the above modulation operations at multiple key levels of the generation architecture, this embodiment ensures that, under the unified constraint of the global style vector, the generation process can adaptively and dynamically decouple and enhance the feature distribution according to the complexity of the 3D spatial structure.

[0064] Step S15: Perform voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weight to obtain the target feature map, and perform layer-by-layer upsampling on the target feature map to generate target three-dimensional digital rock mass data.

[0065] In this embodiment, the target feature map is obtained by performing voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weight. This may include: upsampling the spatial modulation weight to the same target resolution as the three-dimensional feature map, and performing voxel-by-voxel multiplication on the upsampled spatial modulation weight and the three-dimensional feature map to obtain the target feature map; wherein, the value of each voxel in the three-dimensional feature map is multiplied by the weight coefficient at the corresponding position to obtain the target feature map after spatial adaptive modulation.

[0066] Understandably, after obtaining the target feature map, a generative network can be used to perform 3D deconvolution or convolution operations, upsampling the target feature map layer by layer, and finally generating the target 3D digital rock mass data. This target 3D digital rock mass data can be output in 3D voxel format for subsequent rock physics analysis or numerical simulation processing.

[0067] In this embodiment, the generated target 3D digital rock mass data can be compared with real rock sample data or preset statistical constraints to calculate the corresponding loss function. The loss function includes adversarial loss and statistical constraint loss. The adversarial loss is used to constrain the consistency between the generated result and the distribution of the real sample, while the statistical constraint loss is used to constrain the consistency between the pore structure features of the generated result and the macroscopic skeleton of the rock. Then, the network parameters of the style mapping network, the 2D feature encoding branch, and the generation network can be updated based on the loss function.

[0068] As shown above, this embodiment solves the technical problems of texture misalignment and non-adaptive modulation in multimodal fusion through a collaborative mechanism of multi-axial texture feature encoding and spatial adaptive feature modulation. Specifically, by performing parallel feature extraction and integration on two-dimensional scanning electron microscope images in different orthogonal directions, a unified texture feature representation is formed, avoiding decoupling bias when high-resolution textures are mapped to three-dimensional space. Simultaneously, a spatial adaptive feature modulation module is introduced, utilizing multi-scale spatial awareness operators to capture local geometric structural features at different levels in the three-dimensional feature map. Based on adaptive pooling and convolutional mapping techniques, spatial modulation weights matching the local geometric structure are generated in real time. After upsampling the weights, voxel-by-voxel weighted modulation is performed with the three-dimensional feature map, achieving a sensitive response to spatial position differences in the generation process. This embodiment maintains the continuity of the macroscopic structure in three-dimensional computed tomography while achieving effective collaboration between microscopic texture and three-dimensional skeleton, improving the structural consistency and texture realism of the reconstruction results in complex porous structure scenarios, and providing a high-fidelity digital rock model for rock physics numerical simulation.

[0069] See Figure 3 As shown in the embodiments, this application also discloses a digital rock reconstruction device, comprising:

[0070] The data acquisition module 11 is used to acquire multimodal data; the multimodal data includes three-dimensional computed tomography image data for characterizing the macroscopic skeleton of rocks, and two-dimensional scanning electron microscope image data from different orthogonal directions for characterizing the microscopic texture of rocks.

[0071] The first vector determination module 12 is used to extract microscopic texture features from different orthogonal directions in the two-dimensional scanning electron microscope image data, obtain the corresponding texture feature vectors, and integrate the texture feature vectors to obtain the target texture feature vector.

[0072] The second vector determination module 13 is used to extract structural feature information from the three-dimensional computed tomography image data, concatenate the target texture feature vector with the target noise, and map the concatenated vector to the target style vector based on a style mapping network; the style mapping network is a mapping network constructed based on a multilayer fully connected network.

[0073] The weight generation module 14 is used to fuse the structural feature information and the target style vector to obtain a three-dimensional feature map, input the three-dimensional feature map into the spatial adaptive feature modulation module, so as to use the spatial adaptive feature modulation module to capture the geometric feature response corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map, and analyze the geometric feature response based on adaptive pooling and convolution mapping technology to generate spatial modulation weights that match the local geometric structure of the rock.

[0074] The digital rock reconstruction module 15 is used to perform voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weight to obtain a target feature map, and to perform layer-by-layer upsampling on the target feature map to generate target three-dimensional digital rock body data.

[0075] In some specific embodiments, the digital rock reconstruction device further includes:

[0076] The sampling unit is used to randomly divide the acquired two-dimensional scanning electron microscope image data into blocks for cropping out several image blocks from the two-dimensional scanning electron microscope image data.

[0077] The data processing unit is used to perform data normalization and size alignment processing on the acquired three-dimensional computed tomography (CT) image data in order to extract structural feature information from the processed three-dimensional CT image data; the size alignment processing includes spatial dimension alignment and physical resolution alignment.

[0078] In some specific embodiments, the first vector determination module 12 includes:

[0079] An encoding unit is used to input the image blocks in different orthogonal directions into a pre-set two-dimensional feature encoding branch, so as to encode the image blocks based on the two-dimensional feature encoding branch through a shared or parallel feature extraction network to obtain texture feature vectors corresponding to each orthogonal direction; the feature extraction network includes convolutional layers and nonlinear mapping units;

[0080] When encoding the image patch using a shared feature extraction network, the image patch shares the same set of feature extraction networks. When encoding the image patch using a parallel feature extraction network, a set of feature extraction networks with the same structure is configured for image patches in different orthogonal directions, and multiple sets of feature extraction networks work in parallel and independently.

[0081] The vector integration unit is used to integrate the texture feature vectors from different orthogonal directions to obtain the target texture feature vector; the target texture feature vector is used to describe the texture statistical characteristics of rock material in different spatial directions.

[0082] In some specific embodiments, the second vector determination module 13 includes:

[0083] The vector mapping unit is used to concatenate the target texture feature vector with a random noise vector, and input the concatenated vector into the style mapping network so as to map the concatenated vector into a target style vector using the style mapping network.

[0084] In some specific embodiments, the weight generation module 14 includes:

[0085] The response acquisition unit is used to capture the geometric feature responses corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map by utilizing the multi-scale spatial perception operator built into the spatial adaptive feature modulation module.

[0086] The weight generation unit is used to perform feature analysis and numerical calculation on the captured geometric feature response based on adaptive pooling and convolution mapping techniques by the multi-scale spatial perception operator, and generate spatial modulation weights that match the local geometric structure of the rock.

[0087] In some specific embodiments, the digital rock reconstruction module 15 includes:

[0088] The feature map determination unit is used to upsample the spatial modulation weights to the same target resolution as the three-dimensional feature map, and to perform voxel-by-voxel multiplication on the upsampled spatial modulation weights and the three-dimensional feature map to obtain the target feature map;

[0089] The target feature map is obtained by multiplying the value of each voxel in the three-dimensional feature map by the weight coefficient at the corresponding position.

[0090] In some specific embodiments, the digital rock reconstruction device further includes:

[0091] The function calculation unit is used to compare the generated target three-dimensional digital rock body data with real rock sample data or preset statistical constraints, and calculate the corresponding loss function. The loss function includes adversarial loss and statistical constraint loss. The adversarial loss is used to constrain the consistency between the generated result and the real sample distribution, and the statistical constraint loss is used to constrain the consistency between the pore structure characteristics of the generated result and the macroscopic skeleton of the rock.

[0092] The data update unit is used to update the network parameters of the style mapping network, the two-dimensional feature encoding branch, and the generator network based on the loss function.

[0093] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0094] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the digital rock reconstruction method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0095] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0096] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0097] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the digital rock reconstruction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0098] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed digital rock reconstruction method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0099] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0100] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0101] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0102] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0103] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A digital rock reconstruction method, characterized in that, include: Acquire multimodal data; the multimodal data includes three-dimensional computed tomography image data for characterizing the macroscopic framework of rocks, and two-dimensional scanning electron microscope image data from different orthogonal directions for characterizing the microscopic texture of rocks; Microscopic texture features from different orthogonal directions are extracted from the two-dimensional scanning electron microscope image data to obtain the corresponding texture feature vectors. The texture feature vectors are then integrated to obtain the target texture feature vector. Structural feature information is extracted from the three-dimensional computed tomography image data. The target texture feature vector is concatenated with the target noise, and the concatenated vector is mapped to the target style vector based on a style mapping network. The style mapping network is a mapping network constructed based on a multi-layer fully connected network. The structural feature information and the target style vector are fused to obtain a three-dimensional feature map. The three-dimensional feature map is then input into a spatial adaptive feature modulation module to capture the geometric feature responses corresponding to the local geometry of the rock at different levels in the three-dimensional feature map. The geometric feature responses are analyzed based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock. The three-dimensional feature map is subjected to voxel-by-voxel weighted modulation according to the spatial modulation weight to obtain the target feature map. The target feature map is then upsampled layer by layer to generate the target three-dimensional digital rock mass data.

2. The digital rock reconstruction method according to claim 1, characterized in that, Also includes: The acquired two-dimensional scanning electron microscope image data is randomly divided into blocks for sampling, so as to crop out several image blocks from the two-dimensional scanning electron microscope image data; The acquired 3D computed tomography (CT) image data is normalized and sized to extract structural feature information from the processed 3D CT image data. The size alignment process includes spatial dimension alignment and physical resolution alignment.

3. The digital rock reconstruction method according to claim 2, characterized in that, The step of extracting microscopic texture features from different orthogonal directions in the two-dimensional scanning electron microscope image data to obtain corresponding texture feature vectors, and integrating the texture feature vectors to obtain the target texture feature vector, includes: The image blocks with different orthogonal directions are input into a pre-set two-dimensional feature encoding branch, and the image blocks are encoded based on the two-dimensional feature encoding branch through a shared or parallel feature extraction network to obtain texture feature vectors corresponding to each orthogonal direction; the feature extraction network includes convolutional layers and nonlinear mapping units; When encoding the image patch using a shared feature extraction network, the image patch shares the same set of feature extraction networks. When encoding the image patch using a parallel feature extraction network, a set of feature extraction networks with the same structure is configured for image patches in different orthogonal directions, and multiple sets of feature extraction networks work in parallel and independently. The texture feature vectors from different orthogonal directions are integrated to obtain the target texture feature vector; the target texture feature vector is used to describe the texture statistical characteristics of rock material in different spatial directions.

4. The digital rock reconstruction method according to claim 1, characterized in that, The step of concatenating the target texture feature vector with the target noise, and mapping the concatenated vector to a target style vector based on a style mapping network, includes: The target texture feature vector is concatenated with a random noise vector, and the concatenated vector is input into a style mapping network to map the concatenated vector into a target style vector.

5. The digital rock reconstruction method according to claim 1, characterized in that, The process involves using the spatial adaptive feature modulation module to capture geometric feature responses corresponding to the local geometry of the rock at different levels in the 3D feature map, and analyzing these geometric feature responses based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock, including: The multi-scale spatial perception operator built into the spatial adaptive feature modulation module is used to capture the geometric feature responses corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map. The multi-scale spatial perception operator performs feature analysis and numerical calculation on the captured geometric feature responses based on adaptive pooling and convolutional mapping techniques to generate spatial modulation weights that match the local geometry of the rock.

6. The digital rock reconstruction method according to claim 5, characterized in that, The step of performing voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weights to obtain the target feature map includes: The spatial modulation weights are upsampled to the same target resolution as the three-dimensional feature map, and the upsampled spatial modulation weights are multiplied with the three-dimensional feature map on a voxel-by-voxel basis to obtain the target feature map. The target feature map is obtained by multiplying the value of each voxel in the three-dimensional feature map by the weight coefficient at the corresponding position.

7. The digital rock reconstruction method according to any one of claims 1 to 6, characterized in that, Also includes: The generated target three-dimensional digital rock mass data is compared with real rock sample data or preset statistical constraints, and the corresponding loss function is calculated. The loss function includes adversarial loss and statistical constraint loss. The adversarial loss is used to constrain the consistency between the generated result and the real sample distribution, and the statistical constraint loss is used to constrain the consistency between the pore structure characteristics of the generated result and the macroscopic skeleton of the rock. The network parameters of the style mapping network, the two-dimensional feature encoding branch, and the generator network are updated based on the loss function.

8. A digital rock reconstruction device, characterized in that, include: The data acquisition module is used to acquire multimodal data, including three-dimensional computed tomography image data for characterizing the macroscopic skeleton of rocks, and two-dimensional scanning electron microscope image data from different orthogonal directions for characterizing the microscopic texture of rocks. The first vector determination module is used to extract microscopic texture features from different orthogonal directions in the two-dimensional scanning electron microscope image data, obtain the corresponding texture feature vectors, and integrate the texture feature vectors to obtain the target texture feature vector. The second vector determination module is used to extract structural feature information from the three-dimensional computed tomography image data, concatenate the target texture feature vector with the target noise, and map the concatenated vector to the target style vector based on a style mapping network; the style mapping network is a mapping network constructed based on a multilayer fully connected network. The weight generation module is used to fuse the structural feature information and the target style vector to obtain a three-dimensional feature map. The three-dimensional feature map is input into the spatial adaptive feature modulation module to capture the geometric feature responses corresponding to the local geometric structure of the rock at different levels in the three-dimensional feature map. The geometric feature responses are analyzed based on adaptive pooling and convolution mapping techniques to generate spatial modulation weights that match the local geometric structure of the rock. The digital rock reconstruction module is used to perform voxel-by-voxel weighted modulation on the three-dimensional feature map according to the spatial modulation weight to obtain the target feature map, and to perform layer-by-layer upsampling on the target feature map to generate target three-dimensional digital rock body data.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the digital rock reconstruction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the digital rock reconstruction method as described in any one of claims 1 to 7.

Images