CN-121982147-A - Diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing method, system, platform and storage medium

CN121982147ACN 121982147 ACN121982147 ACN 121982147ACN-121982147-A

Abstract

The invention discloses a method, a system, a platform and a storage medium for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model; the high-fidelity reconstruction and generation of the calcareous sand microstructure with extremely complex morphology are realized by integrating the principal component analysis, orientation alignment and v prediction diffusion model, wherein the generated two-dimensional microstructure image is not only highly matched with a real CT slice in vision, but also shows high consistency on the statistical distribution of 14 key morphology parameters such as porosity, fractal dimension, skeleton complexity and the like, and the physical authenticity of the two-dimensional microstructure image is verified. The method has the dual capacities of unconditional generation and conditional reconstruction, not only can synthesize a large number of statistically reasonable digital samples from zero to expand a scarce data set, but also can intelligently enhance low-quality experimental images. The method provides a stable, efficient and extensible digital material source for discrete element numerical simulation, permeability characteristic prediction and microstructure-oriented constitutive model development of the calcareous sand.

Inventors

HUANG LINCHONG
LI CHENGHAO
Lai Zhengshou
HUANG SHUAI
LIN YUEXIANG
LIU MINGJING
ZHANG FU

Assignees

中山大学

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. The method for reconstructing and processing the two-dimensional porous microstructure of the calcareous sand based on the diffusion model is characterized by comprising the following steps of: Creating and acquiring first data corresponding to calcareous sand particles, preprocessing the first data and generating second data corresponding to single calcareous sand particles, wherein the first data are microscopic CT scanning three-dimensional image data of the calcareous sand particles; The second data are processed through space orientation alignment, third data corresponding to a directional section are generated and acquired, and a first data set corresponding to calcareous sand is constructed and generated based on the third data, wherein the third data are standardized two-dimensional slice image data of the section along a specific direction after alignment processing, and the third data are calcareous sand microstructure training data sets; Creating and generating a corresponding first model based on a U-Net architecture, training the first model according to the first data set, and constructing and generating a corresponding second model, wherein the first model is a denoising diffusion probability model; And reconstructing and generating fourth data corresponding to calcareous sand through the second model by combining a reverse denoising process, wherein the fourth data is two-dimensional porous microstructure image data.
2. The method for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model according to claim 1, wherein the spatially aligning the second data, generating and acquiring third data corresponding to a directional cross section, and constructing and generating a first data set corresponding to calcareous sand based on the third data, further comprises: Constructing a covariance matrix corresponding to the particle voxel points, decomposing and generating fifth data corresponding to the particle voxel points based on the covariance matrix, and creating and generating a corresponding two-dimensional slice plane by taking a plane of the fifth data as a reference plane, wherein the fifth data is a main feature vector; Creating and generating a two-dimensional slice image corresponding to the calcareous sand particles, constructing and generating a corresponding standard-size image by combining a zero-value filling mode, and linearly normalizing the pixel value of the standard-size image to a preset numerical interval.
3. The method for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model according to claim 1 or 2, wherein the creating and generating a corresponding first model based on a U-Net architecture, training and processing the first model according to the first data set, and constructing and generating a corresponding second model, further comprises: and constructing and generating a first model by adopting a v prediction parameterization method, wherein a model prediction target is defined as a linear combination of a source signal and injection noise, and the specific expression is as follows: (7) In the formula, In the form of a clean microstructure image, In the event of a noise occurrence, And For and diffuse time step The model training objective function is the predicted speed Mean square error with the target velocity.
4. A method for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model according to claim 3, wherein said creating based on a U-Net architecture generates a corresponding first model, trains the first model according to the first data set, and constructs a corresponding second model, further comprising: controlling noise adding intensities of different time steps in a forward diffusion process through cosine scheduling, and carrying out sampling processing in a reverse denoising process according to DDPM or a DDIM algorithm; Model training is carried out by combining the time step embedded vector, and the time step embedded vector is fused into a convolution block of a U-Net architecture, wherein the U-Net architecture comprises an encoding path, a decoding path and jump connection for connecting the encoding path and the decoding path.
5. The method for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model according to claim 1, wherein the reconstructing by the second model in combination with a reverse denoising process generates fourth data corresponding to the calcareous sand, further comprising: Based on the second model, starting from random noise conforming to Gaussian distribution, and combining a reverse diffusion track, creating and generating sixth data corresponding to calcareous sand, wherein the sixth data is brand new calcareous sand two-dimensional porous microstructure image data.
6. The method for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model according to claim 1 or 5, wherein the reconstructing by the second model in combination with a reverse denoising process generates fourth data corresponding to the calcareous sand, further comprising: Generating and acquiring seventh data corresponding to calcareous sand, wherein the seventh data is microstructure image data of the calcareous sand containing noise; and reconstructing fourth data which corresponds to calcareous sand and has high definition according to the second model and in combination with a conditional reverse denoising process based on the seventh data.
7. A diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing system, characterized in that the system is applied to the diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing method according to any one of claims 1 to 6, the system comprising: the system comprises a data creation generation unit, a data generation unit and a data generation unit, wherein the data creation generation unit is used for creating and acquiring first data corresponding to calcareous sand particles, preprocessing the first data and generating second data corresponding to single calcareous sand particles, wherein the first data is microscopic CT scanning three-dimensional image data of the calcareous sand particles; the data processing generation unit is used for carrying out space orientation alignment on the second data, generating and acquiring third data corresponding to the directional section, and constructing and generating a first data set corresponding to calcareous sand based on the third data, wherein the third data is standardized two-dimensional slice image data of the section in a specific direction after the alignment treatment; the model creation generation unit is used for creating and generating a corresponding first model based on a U-Net architecture, training the first model according to the first data set, and constructing and generating a corresponding second model, wherein the first model is a denoising diffusion probability model; and the data reconstruction generating unit is used for reconstructing and generating fourth data corresponding to the calcareous sand through the second model and combining a reverse denoising process, wherein the fourth data is two-dimensional porous microstructure image data.
8. The diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing system of claim 7, wherein the data processing generation unit further comprises: The first processing module is used for constructing a covariance matrix corresponding to the particle voxel points, generating fifth data corresponding to the particle voxel points based on the covariance matrix decomposition, and creating a corresponding two-dimensional slice plane by taking the plane of the fifth data as a reference plane, wherein the fifth data is a main feature vector; the first generation module is used for creating and generating a two-dimensional slice image corresponding to the calcareous sand particles, constructing and generating an image with a corresponding standard size by combining a zero-value filling mode, and linearly normalizing the pixel value of the image with the standard size to a preset numerical value interval; And/or, the model creation generating unit further includes: The second generation module is used for constructing and generating a first model by adopting a v prediction parameterization method, wherein a model prediction target is defined as a linear combination of a source signal and injection noise, and a specific expression is as follows: (7) In the formula, In the form of a clean microstructure image, In the event of a noise occurrence, And For and diffuse time step The model training objective function is the predicted speed Mean square error with the target velocity; the second processing module is used for controlling the noise adding strength of different time steps in the forward diffusion process through cosine scheduling, and carrying out sampling processing in the reverse denoising process according to DDPM or a DDIM algorithm; the model training module is used for carrying out model training by combining the time step embedded vector, and integrating the time step embedded vector into a convolution block of a U-Net architecture, wherein the U-Net architecture comprises an encoding path, a decoding path and jump connection for connecting the encoding path and the decoding path; And/or, the data reconstruction generating unit further comprises: the third generation module is used for generating sixth data corresponding to the calcareous sand based on the second model, starting from random noise conforming to Gaussian distribution and combining with a reverse diffusion track, wherein the sixth data is brand new calcareous sand two-dimensional porous microstructure image data; The fourth generation module is used for generating and acquiring seventh data corresponding to the calcareous sand, wherein the seventh data is microstructure image data of the calcareous sand containing noise; And the fifth generation module is used for reconstructing and generating fourth data which corresponds to the calcareous sand and has high definition according to the seventh data, the second model and the reverse denoising process in combination with conditions.
9. The diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing platform is characterized by comprising a processor, a memory and a diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing platform control program, wherein the diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing platform control program is executed by the processor and stored in the memory, and the diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing platform control program realizes the diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing method according to any one of claims 1 to 6.
10. A computer readable storage medium, wherein the computer readable storage medium stores a control program of a diffusion model-based two-dimensional porous microstructure reconstruction processing platform of calcareous sand, and the control program of the diffusion model-based two-dimensional porous microstructure reconstruction processing platform of calcareous sand realizes the diffusion model-based two-dimensional porous microstructure reconstruction processing method of calcareous sand according to any one of claims 1 to 6.

Description

Diffusion model-based calcareous sand two-dimensional porous microstructure reconstruction processing method, system, platform and storage medium Technical Field The invention belongs to the technical field of geotechnical engineering digitization and artificial intelligence, and particularly relates to a method, a system, a platform and a storage medium for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model. Background The calcareous sand is a biological causative particle material widely distributed in tropical and subtropical sea areas, and is a key building material for ocean engineering such as artificial island reefs, seabed foundations and the like. Unlike Liu Yuandan quartz sand, the calcareous sand has extremely irregular particle morphology, rough surface and multi-scale pores developed inside, so that the calcareous sand shows unique engineering mechanical behaviors such as high compressibility, easy breaking, complex hydraulic characteristics and the like. The micromechanics mechanisms that reveal these macroscopic properties must rely on the precise characterization of particle-scale microtopography (including external geometry and internal pore structure). The X-ray microcomputer tomography (micro CT) technology is a main means for realizing nondestructive observation of the three-dimensional internal structure of the calcareous sand at present. However, this technique is costly, has long data processing cycles, and has a limited single scan field of view, resulting in a serious inadequacy in the size of high quality, representative calcareous sand microstructure data sets that can be acquired. The current status of data starvation greatly restricts the development of statistical analysis based on a large amount of data, calibration of Discrete Element Method (DEM) numerical simulation, and data-driven constitutive models. To break through the above-mentioned bottleneck, researchers have attempted to reconstruct the microstructure of porous media using numerical methods. While the traditional process simulation method is difficult to reproduce a complex pore system specific to biological debris particles, the deep learning method based on generation of a countermeasure network (GAN) and the like often faces the problems of unstable training, mode collapse and the like, and is difficult to reconstruct microscopic geometric details (such as fine throats and non-convex pores) which are vital to mechanical and hydraulic characteristics in a fidelity manner. The denoising diffusion probability model is a type of depth generation model which is rising in recent years, and fits complex data distribution by learning a reverse gradual denoising process. The model has the advantages of stable training, high generation quality and capability of effectively capturing complex geometric forms and space variability, and has remarkable success in the field of image generation. However, how to systematically apply the diffusion model to the microstructure reconstruction of the calcareous sand of the natural granular material with the most complex morphology, no intensive research and viable technical solutions are currently available. Therefore, in order to overcome the technical problems and disadvantages, there is an urgent need to design and develop a method, a system, a platform and a storage medium for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model. Disclosure of Invention In order to overcome the defects and difficulties in the prior art, the invention aims to provide a method, a system, a platform and a storage medium for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model, so as to solve the problem that a large number of representative calcareous sand microstructure samples are difficult to obtain due to the scarcity of CT scanning data in the prior art. The invention provides a method for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model, a system for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model, a platform for reconstructing a two-dimensional porous microstructure of calcareous sand based on a diffusion model and a computer readable storage medium. The first object of the invention is achieved in that the method comprises: Creating and acquiring first data corresponding to calcareous sand particles, preprocessing the first data and generating second data corresponding to single calcareous sand particles, wherein the first data are microscopic CT scanning three-dimensional image data of the calcareous sand particles; The second data are processed through space orientation alignment, third data corresponding to a directional section are generated and acquired, and a first data set corresponding to calcareous sand is constructed and generated base