Method and device for generating dynamic evolution process of three-dimensional pore structure of soil aggregates

By simulating the dynamic evolution of the three-dimensional pore structure of soil agglomerates through the generation adversarial network, the problem of destructiveness and low efficiency of the three-dimensional pore structure generation method of soil agglomerates in the prior art is solved, and efficient and non-destructive soil pore structure dynamics are achieved. Evolutionary modeling and visualization.

CN118886343BActive Publication Date: 2024-11-26JILIN AGRICULTURAL UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411388646.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-08
Publication Date
2024-11-26
Estimated Expiration
2044-10-08

AI Technical Summary

Technical Problem

The existing method of generating the dynamic evolution process of three-dimensional pore structures of soil agglomerates has problems such as destructiveness and inability to observe the dynamic changes and low efficiency of soil pore structures in real time.

Method used

Using a generative adversarial network (GAN) method, the three-dimensional pore generation adversarial network model of soil agglomerates is constructed, combining interval random sampling, differential strategy and spherical linear interpolation to simulate the dynamic evolution process of the three-dimensional pore structure of soil agglomerates, providing a Efficient, non-destructive solution.

Benefits of technology

It realizes efficient modeling and visualization of the dynamic changes of soil agglomerated pore structures, and can automatically and quickly generate clear and realistic three-dimensional pore structures of soil agglomerates, providing new methods and tools to study the dynamic evolution of soil structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118886343B_ABST
    Figure CN118886343B_ABST
Patent Text Reader

Abstract

The method and device for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates belong to the field of agricultural information technology, and solve the problems that the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates is destructive, cannot observe the dynamic changes of the soil pore structure in real time, and is inefficient. The method of the present invention includes: obtaining a three-dimensional pore substructure data set of soil aggregates; training a three-dimensional pore generative adversarial network model of soil aggregates based on the pore substructure; using a differential strategy to obtain an input noise vector corresponding to the starting point pore characteristics and the end point pore characteristics; using spherical linear interpolation to obtain an intermediate noise vector; inputting the noise vector into the generation model to obtain the dynamic evolution process of the pore three-dimensional structure. The present invention is suitable for real-time and accurate reconstruction and prediction of the change process of the soil pore structure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of agricultural information technology, and in particular to the generation of a dynamic evolution process of a three-dimensional pore structure of soil aggregates. Background Art

[0002] Soil aggregates are the basic units of soil structure, and they have very complex pore structures inside. The characteristic differences of these structures have an important impact on the migration of soil water, gas and nutrients, and play a key role in regulating soil functions. Good soil aggregate pore structure can promote plant root growth, improve the efficiency of water and nutrient utilization, and thus improve soil quality and maintain soil health. Therefore, studying soil pore structure and its dynamic evolution process is of great significance for understanding soil functions.

[0003] Traditional methods for studying soil pore structure mainly include experimental measurement and image processing technology. Although experimental measurement methods such as gas adsorption and mercury intrusion can obtain some quantitative information about soil pores, these methods are usually destructive and cannot observe the dynamic changes of soil pore structure in real time. Image processing techniques such as CT scanning and MRI imaging can provide three-dimensional structural information of soil, but these technologies usually require expensive equipment and complex post-processing processes, and are inefficient when processing large-scale data. Summary of the invention

[0004] The purpose of the present invention is to solve the problems that the existing method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates is destructive, cannot observe the dynamic changes of the soil pore structure in real time and is inefficient, and provides a method and equipment for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates.

[0005] The present invention is achieved through the following technical solutions. In one aspect, the present invention provides a method for generating a dynamic evolution process of a three-dimensional pore structure of a soil aggregate, the method comprising:

[0006] S1: Obtain the pore part in the original soil aggregate image and construct a voxel matrix dataset of the three-dimensional pore structure of soil aggregates;

[0007] S2: performing random sampling on the soil aggregate three-dimensional pore structure voxel matrix dataset to obtain a soil aggregate three-dimensional pore substructure dataset;

[0008] S3: constructing a soil aggregate three-dimensional pore generative adversarial network model, wherein the soil aggregate three-dimensional pore generative adversarial network model includes a generative model and a discriminative model;

[0009] The generative model is used to learn the distribution of three-dimensional pores of real soil aggregates and generate pseudo-pore structures;

[0010] The discriminant model is used to learn to distinguish true and false pore structures and calculate the probability that the pore structure comes from a three-dimensional pore substructure data set of soil aggregates;

[0011] S4: Training a soil aggregate 3D pore generative adversarial network model based on the soil aggregate 3D pore substructure dataset until the soil aggregate 3D pore generative adversarial network model generates a soil aggregate 3D pore structure that matches the distribution of the real 3D pore substructure characteristics;

[0012] S5: Defining the starting point pore characteristics of the 3D pore structure evolution process of soil aggregates and endpoint pore characteristics ;

[0013] S6: Use differential strategy to obtain pore characteristics and The corresponding soil aggregate 3D pore generation adversarial network model input noise vector and ;

[0014] S7: Use spherical linear interpolation to obtain the noise vector and The intermediate noise vector between ;

[0015] S8: The noise vector The data is input into the soil aggregate three-dimensional pore generative adversarial network model to obtain the dynamic evolution process of the three-dimensional pore structure.

[0016] Furthermore, in S3, the objective loss function of the soil aggregate three-dimensional pore generation adversarial network model is:

[0017]

[0018] In the formula, To generate the model, is the discriminant model, is the real pore structure. is the mathematical expectation of the real pore structure, Generate model for input A normally distributed random variable, is the mathematical expectation for generating the pore structure.

[0019] Further, in S3, the generation model is composed of The layer is composed of deconvolution layers;

[0020] The generative model is The deconvolution output of the first layer is batch normalized and ReLU nonlinearly activated. The generative model performs Tanh nonlinear activation on the deconvolution output of the last layer to generate the three-dimensional pore substructure of soil aggregates.

[0021] The discriminant model is composed of The convolutional layers are composed in the reverse order of the generative model;

[0022] The discriminant model is The convolution output of the layer is spectrally normalized and LeakyReLU nonlinearly activated; the discriminant model performs Sigmoid nonlinear activation on the convolution output of the last layer to calculate the probability that the input pore structure is classified as true; the expression of the Sigmoid function is as follows:

[0023]

[0024] In the formula, It is the convolution output of the last layer of the discriminative model.

[0025] Furthermore, in S4, the hyperparameters of the soil aggregate three-dimensional pore generative adversarial network model are: batch size is 64, loss function optimizer is AdamW, learning rate is 0.0001, Beta1 is 0.6, and number of iterations is 500; the batch size is the number of samples selected for one training; Beta1 is the exponential decay rate of the first-order moment estimate.

[0026] Furthermore, S5 includes:

[0027] For the simulated The three-dimensional pore characteristics of soil aggregates are defined respectively as the pore characteristic calculation function , traverse the soil aggregate 3D pore substructure dataset The three-dimensional pore substructure of each soil aggregate , calculate the soil aggregate three-dimensional pore substructure data set The upper bound of each pore feature and the lower bound ,in:

[0028]

[0029]

[0030] According to the starting pore characteristics and the end pore characteristics of the pore characteristic evolution process to be simulated, Select the starting pore features and endpoint pore characteristics .

[0031] Furthermore, S6 includes:

[0032] S6.1: Random sampling from a normal distribution indivual dimensional random noise ;

[0033] S6.2: Noise Input them into the soil aggregate three-dimensional pore generation adversarial network model to obtain the pore structure ;

[0034] S6.3: Calculate pore structure Pore ​​characteristics , respectively calculate Pore ​​characteristics and dynamic evolution of initial pore structure or pore characteristics of dynamically evolving pore structures Euclidean distance , where S and E correspond to the starting point and ending point of the pore structure during the dynamic evolution process, respectively;

[0035] S6.4: From Randomly select 3 noises from , and , substituting into the following formula to obtain the new random noise :

[0036]

[0037] Where, F is a random number between [0,1];

[0038] S6.5: Yes The second norm of , then use replace Otherwise it is not correct Modify the value of

[0039] in

[0040]

[0041]

[0042] in, represents the gamma function;

[0043] S6.6: Input into the soil aggregate three-dimensional pore generation adversarial network model to obtain the output pore structure ,calculate The pore characteristic value and or Euclidean distance ;

[0044] S6.7: If Less than , then Replace with ,Will Replace with ;

[0045] S6.8: Traverse each , repeat S6.4 to S6.7;

[0046] S6.9: Repeat S6.8 MaxIter times, output The noise vector corresponding to the minimum value in , denoted as the noise vector of the initial pore structure of dynamic evolution Or the noise vector of the dynamically evolving pore structure .

[0047] Furthermore, S7 includes:

[0048] For the noise vector and , the expression for defining spherical linear interpolation is:

[0049]

[0050] in, , is a real number between [0,1];

[0051] set up ,Will As the input of spherical linear interpolation, we get the noise vector and The intermediate noise vector between .

[0052] Furthermore, S1 includes:

[0053] S1.1: Use ImageJ software to capture the middle area of ​​the soil aggregate CT scan image as the feature domain;

[0054] S1.2: Import the 2D image stack into ImageJ software and perform pre-processing for image segmentation, including:

[0055] S1.2.1: Use the Enhance Contrast command to normalize the image to eliminate the brightness difference between different CT slice images;

[0056] S1.2.2: Use the Despeckle command to remove image noise;

[0057] S1.2.3: Use Median operations to achieve smooth operations;

[0058] S1.3: When performing image segmentation operations, use the Threshold operation to perform image threshold segmentation to separate the pore parts in the image; generate a binary image and convert it into a voxel matrix whose elements are all 0 or 1.

[0059] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the steps of a method for generating a dynamic evolution process of a three-dimensional pore structure of soil aggregates as described above are executed.

[0060] In a third aspect, the present invention provides a computer-readable storage medium, in which a plurality of computer instructions are stored, and the plurality of computer instructions are used to enable a computer to execute a method for generating a dynamic evolution process of a three-dimensional pore structure of a soil aggregate as described above.

[0061] Beneficial effects of the present invention:

[0062] The method of the present invention simulates the dynamic evolution of soil pore three-dimensional structure based on a generative model, aiming to provide an efficient and non-destructive solution for studying the dynamic evolution process of soil structure.

[0063] The method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates provided by the present invention provides an efficient and non-destructive solution for studying and understanding the evolution process of the three-dimensional pore structure of soil aggregates; it can effectively model the dynamic change law of pore characteristics in the pores of soil aggregates, and provides a reference method for studying the evolution of soil aggregates; it provides a method for visualizing the dynamic evolution process of the three-dimensional pore structure of soil aggregates, and can automatically and quickly generate a large number of relatively clear, visually realistic, and continuously changing three-dimensional pore structures of soil aggregates.

[0064] The present invention uses a 3D-DCGAN model (3D deep convolutional generative adversarial network) to generate the 3D pore structure of soil aggregates, and applies the model to the 3D pore structure data set of soil aggregates. It can capture the underlying pore distribution law of the 3D pore structure of soil aggregates by unsupervised learning rather than manually constructing features, and can apply the distribution law learned by the generative model to the subsequent dynamic evolution generation process.

[0065] The present invention proposes a meta-heuristic optimization algorithm to optimize the input noise of the generative model. Thus, for soil aggregates with target pore characteristics, the meta-heuristic optimization algorithm can search for a noise vector Z, input Z into the generative model, and the Euclidean distance between the pore characteristics of the pore structure X output by the generative model and the target pore characteristics is less than a given constant. At the same time, the search process of the proposed meta-heuristic optimization algorithm is also optimized for the distribution of the noise space, so that each search point in the search process is consistent with the distribution of the generative model input. The input noise vector of a given pore characteristic can be searched unbiasedly. And for subsequent spherical linear interpolation, a starting point noise vector and an end point noise vector are provided.

[0066] The generative model of the present invention has shown great potential in soil simulation and reconstruction of complex structures. Based on Generative Adversarial Networks (GAN) as a generative model, realistic data is generated through an adversarial training process. The generative model is applied to the simulation and dynamic evolution research of soil pore structure. It can reconstruct and predict the change process of soil pore structure in real time and accurately under non-destructive conditions, providing new methods and tools for soil science research. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0068] Figure 1 It is a flow chart of the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates;

[0069] Figure 2 The basic framework structure diagram of the generative model in the soil aggregate three-dimensional pore generative adversarial network model;

[0070] Figure 3 The basic framework structure diagram of the discriminant model in the three-dimensional pore generation adversarial network model of soil aggregates;

[0071] Figure 4 Schematic diagram of the process of obtaining the noise vector using the differential strategy;

[0072] Figure 5 Schematic diagram of the process of generating the dynamic evolution of the three-dimensional pore structure of soil aggregates. DETAILED DESCRIPTION

[0073] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0074] Embodiment 1: A method for generating a dynamic evolution process of a three-dimensional pore structure of a soil aggregate, the method comprising:

[0075] S1: Obtain the pore part in the original soil aggregate image and construct a voxel matrix dataset of the three-dimensional pore structure of soil aggregates;

[0076] S2: performing random sampling on the soil aggregate three-dimensional pore structure voxel matrix dataset to obtain a soil aggregate three-dimensional pore substructure dataset;

[0077] S3: constructing a soil aggregate three-dimensional pore generative adversarial network model, wherein the soil aggregate three-dimensional pore generative adversarial network model includes a generative model and a discriminative model;

[0078] The generative model is used to learn the distribution of three-dimensional pores of real soil aggregates and generate pseudo-pore structures;

[0079] The discriminant model is used to learn to distinguish true and false pore structures and calculate the probability that the pore structure comes from a three-dimensional pore substructure data set of soil aggregates;

[0080] S4: Training a soil aggregate 3D pore generative adversarial network model based on the soil aggregate 3D pore substructure dataset until the soil aggregate 3D pore generative adversarial network model generates a soil aggregate 3D pore structure that matches the distribution of the real 3D pore substructure characteristics;

[0081] S5: Defining the starting point pore characteristics of the 3D pore structure evolution process of soil aggregates and endpoint pore characteristics ;

[0082] S6: Use differential strategy to obtain pore characteristics and The corresponding soil aggregate 3D pore generation adversarial network model input noise vector and ;

[0083] S7: Use spherical linear interpolation to obtain the noise vector and The intermediate noise vector between ;

[0084] S8: The noise vector The data is input into the soil aggregate three-dimensional pore generative adversarial network model to obtain the dynamic evolution process of the three-dimensional pore structure.

[0085] This implementation adopts a generative model-based simulation of the dynamic evolution of soil pore three-dimensional structure, providing an efficient and non-destructive solution for studying the dynamic evolution of soil structure. It can effectively model the dynamic change law of pore characteristics in soil aggregate pores, providing a reference method for studying the evolution of soil aggregates; it provides a method for visualizing the dynamic evolution process of soil aggregate three-dimensional pore structure, which can automatically and quickly generate a large number of relatively clear, visually realistic, and continuously changing soil aggregate three-dimensional pore structures.

[0086] Embodiment 2: This embodiment further defines the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates described in Embodiment 1. In this embodiment, S3 is further defined, specifically including:

[0087] In S3, the objective loss function of the soil aggregate three-dimensional pore generation adversarial network model is:

[0088]

[0089] In the formula, To generate the model, is the discriminant model, is the real pore structure. is the mathematical expectation of the real pore structure, Generate model for input A normally distributed random variable, is the mathematical expectation for generating the pore structure.

[0090] The generative model is composed of The layer is composed of deconvolution layers;

[0091] The generative model is The deconvolution output of the first layer is batch normalized and ReLU nonlinearly activated. The generative model performs Tanh nonlinear activation on the deconvolution output of the last layer to generate the three-dimensional pore substructure of soil aggregates.

[0092] The discriminant model is composed of The convolutional layers are composed in the reverse order of the generative model;

[0093] The discriminant model is The convolution output of the layer is spectrally normalized and LeakyReLU nonlinearly activated; the discriminant model performs Sigmoid nonlinear activation on the convolution output of the last layer to calculate the probability that the input pore structure is classified as true; the expression of the Sigmoid function is as follows:

[0094]

[0095] In the formula, It is the convolution output of the last layer of the discriminative model.

[0096] In this implementation, the 3D-DCGAN (three-dimensional deep convolutional generative adversarial network) model is applied to the three-dimensional generation of the three-dimensional pore structure of soil aggregates, and a deep learning model that captures the underlying distribution law of the three-dimensional structure of soil aggregates is constructed.

[0097] Embodiment 3: This embodiment further defines the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates as described above. In this embodiment, S4 is further defined, specifically including:

[0098] In S4, the hyperparameters of the soil aggregate three-dimensional pore generative adversarial network model are: batch size is 64, loss function optimizer is AdamW, learning rate is 0.0001, Beta1 is 0.6, and number of iterations is 500; the batch size is the number of samples selected for one training; Beta1 is the exponential decay rate of the first-order moment estimate.

[0099] This embodiment provides specific parameter settings for a soil aggregate three-dimensional pore generative adversarial network model.

[0100] Embodiment 4: This embodiment further defines the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates as described above. In this embodiment, S5 is further defined, specifically including:

[0101] S5 includes:

[0102] For the simulated The three-dimensional pore characteristics of soil aggregates are defined respectively as the pore characteristic calculation function , traverse the soil aggregate 3D pore substructure dataset The three-dimensional pore substructure of each soil aggregate , calculate the soil aggregate three-dimensional pore substructure data set The upper bound of each pore feature and the lower bound ,in:

[0103]

[0104]

[0105] According to the starting pore characteristics and the end pore characteristics of the pore characteristic evolution process to be simulated, Select the starting pore features and endpoint pore characteristics .

[0106] This embodiment provides a method for constructing a starting point pore feature vector and an end point pore feature vector of a dynamic evolution generation model using three pore features: porosity, specific surface area, and connectivity. The method uses three pore features: porosity, specific surface area, and connectivity to construct a starting point pore feature vector and an end point pore feature vector of a dynamic evolution generation model.

[0107] Embodiment 5: This embodiment further defines the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates as described above. In this embodiment, S6 is further defined, specifically including:

[0108] S6 includes:

[0109] S6.1: Random sampling from a normal distribution indivual dimensional random noise ;

[0110] S6.2: Noise Input them into the soil aggregate three-dimensional pore generation adversarial network model to obtain the pore structure ;

[0111] S6.3: Calculate pore structure Pore ​​characteristics , respectively calculate Pore ​​characteristics and dynamic evolution of initial pore structure or pore characteristics of dynamically evolving pore structures Euclidean distance , where S and E correspond to the starting point and ending point of the pore structure during the dynamic evolution process, respectively;

[0112] S6.4: From Randomly select 3 noises from , and , substituting into the following formula to obtain the new random noise :

[0113]

[0114] Where, F is a random number between [0,1];

[0115] S6.5: Yes The second norm of , then use replace Otherwise it is not correct Modify the value of

[0116] in

[0117]

[0118]

[0119] in, represents the gamma function;

[0120] S6.6: Input into the soil aggregate three-dimensional pore generation adversarial network model to obtain the output pore structure ,calculate The pore characteristic value and or Euclidean distance ;

[0121] S6.7: If Less than , then Replace with ,Will Replace with ;

[0122] S6.8: Traverse each , repeat S6.4 to S6.7;

[0123] S6.9: Repeat S6.8 MaxIter times, output The noise vector corresponding to the minimum value in , denoted as the noise vector of the initial pore structure of dynamic evolution Or the noise vector of the dynamically evolving pore structure .

[0124] This embodiment proposes a meta-heuristic optimization algorithm to optimize the input noise of the generative model. Thus, for soil aggregates with target pore characteristics, the meta-heuristic optimization algorithm can search for a noise vector Z, input Z into the generative model, and the Euclidean distance between the pore characteristics of the pore structure X output by the generative model and the target pore characteristics is less than a given constant. At the same time, the search process of the proposed meta-heuristic optimization algorithm is also optimized for the distribution of the noise space, so that each search point in the search process is consistent with the distribution of the generative model input.

[0125] Embodiment 6: This embodiment further limits the method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates as described above. In this embodiment, S7 is further limited, specifically including:

[0126] S7 includes:

[0127] For the noise vector and , the expression for defining spherical linear interpolation is:

[0128]

[0129] in, , is a real number between [0,1];

[0130] set up ,Will As the input of spherical linear interpolation, we get the noise vector and The intermediate noise vector between .

[0131] This embodiment provides a method of using spherical linear interpolation to obtain a noise vector and The intermediate noise vector between method.

[0132] Embodiment 7: This embodiment is Example 1 of a method for generating a dynamic evolution process of a three-dimensional pore structure of a soil aggregate as described above, and specifically includes:

[0133] A method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates, such as Figure 1 As shown, the method includes:

[0134] S1: Preprocess the original soil aggregate CT scan image data, separate the pore part in the image, and construct the soil aggregate three-dimensional pore structure voxel matrix data set;

[0135] S2: Perform random sampling of the soil aggregate 3D pore structure dataset to obtain the soil aggregate 3D pore substructure dataset;

[0136] S3: Construct a three-dimensional pore generative adversarial network model for soil aggregates;

[0137] S4: Training a soil aggregate 3D pore generative adversarial network model based on the soil aggregate 3D pore substructure dataset until the soil aggregate 3D pore generative adversarial network model generates a soil aggregate 3D pore structure that matches the distribution of the real 3D pore substructure characteristics;

[0138] S5: Defining the starting point pore characteristics of the 3D pore structure evolution process of soil aggregates and endpoint pore characteristics ;

[0139] S6: Use differential strategy to obtain pore characteristics and The corresponding soil aggregate 3D pore generation adversarial network model input noise vector and ;

[0140] S7: Use spherical linear interpolation to obtain the noise vector and The intermediate noise vector between ;

[0141] S8: The noise vector The data is input into the soil aggregate three-dimensional pore generative adversarial network model to obtain the dynamic evolution process of the three-dimensional pore structure.

[0142] Furthermore, a preferred embodiment is provided, wherein S1 comprises:

[0143] ImageJ software was used to capture the middle area of ​​the soil aggregate CT scan image ( voxels) as the feature domain. When selecting the feature domain, in order to reduce the influence of external factors such as edge effect and radiation hardening, obvious cracks and sample edge areas were avoided. Then the two-dimensional image stack was imported into ImageJ software, and the following operations were performed for image segmentation preprocessing. First, the Enhance Contrast command was used to normalize the image to eliminate the brightness difference between different CT slice images; then the Despeckle command was used to eliminate the image noise; finally, the Median operation was used to perform smoothing operations, making the pore structure in the image clearer and more accurate. When performing image segmentation operations, the Threshold operation was used to perform image threshold segmentation to separate the pore part in the image. A binary image was generated, in which the white area represented the pores and the black area represented the soil matrix, and they were converted into a voxel matrix with elements of 0 or 1. This voxel matrix is ​​the voxel matrix dataset of the three-dimensional pore structure of soil aggregates.

[0144] Furthermore, a preferred embodiment is provided, wherein S2 comprises:

[0145] In the soil aggregate 3D pore structure voxel matrix dataset, a soil aggregate 3D pore structure voxel matrix is ​​randomly selected. The size of the voxel matrix is . Randomly intercept a voxel matrix with a size of The voxel sub-matrix of soil aggregates is obtained by An element in .

[0146] in, The selection of refers to the basic volume cutting theory, so that the voxel submatrix of this specification is sufficient to effectively capture and preserve the geometric characteristics of the pore structure and the interactions between the pore structures.

[0147] Repeat the above process The soil aggregate 3D pore substructure dataset is completed. 's construction.

[0148] Furthermore, a preferred embodiment is provided, in which the soil aggregate three-dimensional pore generation adversarial network model in S3 includes:

[0149] The soil aggregate three-dimensional pore generative adversarial network model is divided into two parts: a generative model and a discriminative model;

[0150] like Figure 2 As shown, the generative model is used to learn the potential distribution of three-dimensional pores of real soil aggregates and generate pseudo-pore structures that are indistinguishable from real pores;

[0151] like Figure 3As shown, the discriminant model is used to learn to distinguish true and false pore structures and calculate the probability that the pore structure comes from the real training set;

[0152] The objective loss function of the soil aggregate three-dimensional pore generation adversarial network model is:

[0153]

[0154] In the formula, To generate the model, is the discriminant model, is the real pore structure. is the mathematical expectation of the real pore structure, Generate model for input A normally distributed random variable, To generate the mathematical expectation of the pore structure;

[0155] The generative model is composed of The layer is composed of deconvolution layers;

[0156] The generative model is The deconvolution output of the first layer is batch normalized and ReLU nonlinearly activated, and the generative model performs Tanh nonlinear activation on the deconvolution output of the last layer to generate the three-dimensional pore substructure of soil aggregates.

[0157] The discriminant model is composed of The convolutional layers are composed in the reverse order of the generative model;

[0158] The discriminant model is The convolution output of the layer is spectrally normalized and LeakyReLU nonlinearly activated; the discriminant model performs Sigmoid nonlinear activation on the convolution output of the last layer to calculate the probability that the input pore structure is classified as true; the expression of the Sigmoid function is as follows:

[0159]

[0160] In the above formula, It is the convolution output of the last layer of the discriminative model.

[0161] Furthermore, a preferred method is provided, wherein the effect of improving the three-dimensional pore substructure generated by the three-dimensional pore generation adversarial network model of soil aggregates in S4, and optimizing the network framework structure and hyperparameters include:

[0162] After completing the construction of the soil aggregate three-dimensional pore generative adversarial network model, it is necessary to further optimize the hyperparameters of the model to achieve the best effect in the generation of the three-dimensional pore structure of soil aggregates; S4 The hyperparameters of the optimized soil aggregate three-dimensional pore generative adversarial network model are: batch size is 64, loss function optimizer is AdamW, learning rate is 0.0001, Beta1 is 0.6, and number of iterations is 500; the batch size is the number of samples selected for one training; Beta1 is the exponential decay rate of the first-order moment estimate.

[0163] Furthermore, a preferred embodiment is provided, wherein S5 includes:

[0164] For those who want to simulate The three-dimensional pore characteristics of soil aggregates are defined respectively as the pore characteristic calculation function , traverse the soil aggregate 3D pore substructure dataset The three-dimensional pore substructure of each soil aggregate , calculate the soil aggregate three-dimensional pore substructure data set The upper bound of each pore feature and the lower bound ,in:

[0165]

[0166]

[0167] According to the starting pore characteristics and the end pore characteristics of the pore characteristic evolution process to be simulated, Select the starting pore features and endpoint pore characteristics .

[0168] Further, a preferred method is also provided, such as Figure 4 As shown, S6 includes:

[0169] Step 1: Randomly sample from a normal distribution indivual dimensional random noise ;

[0170] Step 2: Noise The pore structures of soil aggregates are input into the three-dimensional pore generation adversarial network model to obtain the output pore structure. ;

[0171] Step 3: Calculate the pore structure Pore ​​characteristics , respectively calculate and (or ) ;

[0172] Step 4: Traverse each , repeat Step 5 to Step 8;

[0173] Step 5: From Randomly select 3 noises from , and , substituting into the following formula to obtain the new random noise :

[0174]

[0175] Step 6: The second norm of , then use replace ,in

[0176]

[0177]

[0178] in, represents the gamma function.

[0179] Step 7: Input into the soil aggregate three-dimensional pore generation adversarial network model to obtain the output pore structure ,calculate The pore characteristic value and (or ) ;

[0180] Step 8: If Less than , then Replace with ,Will Replace with ;

[0181] Step 9: Repeat Step 4 MaxIter times and output The noise vector corresponding to the minimum value in , denoted as (or ).

[0182] Furthermore, a preferred embodiment is provided, wherein S7 includes:

[0183] For the noise vector and , the expression for defining spherical linear interpolation is:

[0184]

[0185] in, , is a real number between [0,1].

[0186] set up ,Will As the input of spherical linear interpolation, the output noise vector is obtained .

[0187] Further, a preferred method is also provided, such as Figure 5 As shown, S8 includes:

[0188] The noise vector The input is used in the soil aggregate three-dimensional pore generation adversarial network model to obtain the dynamic evolution process of soil aggregate three-dimensional pore output by the model. .

[0189] Embodiment 8: This embodiment is Example 2 of a method for generating a dynamic evolution process of a three-dimensional pore structure of a soil aggregate as described above, and specifically includes:

[0190] A method for generating a dynamic evolution process of a three-dimensional pore structure of a soil aggregate, the method comprising:

[0191] S1: Preprocess the original soil aggregate CT scan image data, separate the pore part in the image, and construct the soil aggregate three-dimensional pore structure voxel matrix data set;

[0192] S2: Perform random sampling of the soil aggregate 3D pore structure dataset to obtain the soil aggregate 3D pore substructure dataset;

[0193] S3: Construct a three-dimensional pore generative adversarial network model for soil aggregates;

[0194] S4: Training a soil aggregate 3D pore generative adversarial network model based on the soil aggregate 3D pore substructure dataset until the soil aggregate 3D pore generative adversarial network model generates a soil aggregate 3D pore structure that matches the distribution of the real 3D pore substructure characteristics;

[0195] S5: Defining the starting point pore characteristics of the 3D pore structure evolution process of soil aggregates and endpoint pore characteristics ;

[0196] S6: Use differential strategy to obtain pore characteristics and The corresponding soil aggregate 3D pore generation adversarial network model input noise vector and ;

[0197] S7: Use spherical linear interpolation to obtain the noise vector and The intermediate noise vector between ;

[0198] S8: The noise vector The data is input into the soil aggregate three-dimensional pore generative adversarial network model to obtain the dynamic evolution process of the three-dimensional pore structure.

[0199] Furthermore, a preferred embodiment is provided, wherein S1 comprises:

[0200] ImageJ software was used to intercept 9 middle parts (256×256×256 voxels) of the soil aggregate CT scan image as feature domains. When selecting the feature domain, in order to reduce the influence of external factors including edge effect and radiation hardening, obvious cracks and sample edge areas were avoided. The two-dimensional image stack was then imported into ImageJ software, and the following operations were performed for image segmentation preprocessing. First, the Enhance Contrast command was used to normalize the image to eliminate the brightness difference between different CT slice images; then the Despeckle command was used to eliminate the image noise; finally, the Median operation was used to perform a smoothing operation to make the pore structure in the image clearer and more accurate. When performing image segmentation operations, the Threshold operation was used to perform image threshold segmentation to separate the pore part in the image. 9 groups of 256 binary images were generated, in which the white area represents the pores and the black area represents the soil matrix, and they were converted into 9 Voxel matrix, which is the voxel matrix dataset of soil aggregate three-dimensional pore structure.

[0201] Furthermore, a preferred embodiment is provided, wherein S2 comprises:

[0202] In the soil aggregate 3D pore structure voxel matrix dataset, a soil aggregate 3D pore structure voxel matrix is ​​randomly selected. The size of the voxel matrix is . Randomly intercept a voxel matrix with a size of The voxel sub-matrix of soil aggregates is obtained by An element in .

[0203] in, The selection of refers to the basic volume cutting theory, so that the voxel submatrix of this specification is sufficient to effectively capture and preserve the geometric characteristics of the pore structure and the interactions between the pore structures.

[0204] Repeat the above process The soil aggregate 3D pore substructure dataset is completed. 's construction.

[0205] Based on the representative basic volume cutting theory, the size is selected The voxel matrix is ​​used as the substructure unit for analysis. The voxel matrix of this specification is sufficient to effectively capture and preserve the geometric characteristics of the substructures and the interactions between the substructures. The voxel matrix specifications were used to intercept the soil aggregate three-dimensional pore structure voxel matrix dataset, and 8192 sub-matrices were extracted as the soil aggregate three-dimensional pore substructure dataset using the interval random sampling method.

[0206] Furthermore, a preferred embodiment is provided, in which the soil aggregate three-dimensional pore generation adversarial network model in S3 includes:

[0207] The soil aggregate three-dimensional pore generative adversarial network model is divided into two parts: a generative model and a discriminative model;

[0208] The generative model is used to learn the potential distribution of three-dimensional pores of real soil aggregates and generate pseudo-pore structures that are indistinguishable from real pores;

[0209] The discriminant model is used to learn to distinguish true and false pore structures and calculate the probability that the pore structure comes from the real training set;

[0210] The objective loss function of the soil aggregate three-dimensional pore generation adversarial network model is:

[0211]

[0212] In the formula, To generate the model, is the discriminant model, is the real pore structure. is the mathematical expectation of the real pore structure, Generate model for input A normally distributed random variable, To generate the mathematical expectation of the pore structure;

[0213] The generative model consists of 5 deconvolution layers;

[0214] The number of channels in the input layer of the generative model is 512; from deconvolution layer 1 to 5, the number of channels in each layer output is 512, 256, 128, 64 and 1 respectively; the convolution kernel size of the first deconvolution layer is , filled with 0, with a step size of 1; the convolution kernel size of the 2nd to 5th deconvolution layers is , fill is 1, stride is 2;

[0215] The generative model performs batch normalization and ReLU nonlinear activation on the deconvolution outputs of the first four layers, and performs Tanh nonlinear activation on the deconvolution outputs of the last layer to generate The three-dimensional pore substructure of soil aggregates.

[0216] The discriminative model consists of 5 convolutional layers in the reverse order of the generative model;

[0217] The input of the first convolutional layer is: The three-dimensional pore substructure of soil aggregates;

[0218] The number of channels in the input layer of the discriminant model is 1; from convolutional layers 1 to 5, the number of channels in each layer output is 64, 128, 256, 512, and 1 respectively; the convolution kernel size of the 1st to 4th convolutional layers is , the padding is 1, the stride is 2; the convolution kernel size of the fifth convolution layer is , filled with 0, step size is 1;

[0219] The discriminant model performs batch normalization and LeakyReLU nonlinear activation on the convolution outputs of the first four layers; the discriminant model performs Sigmoid nonlinear activation on the convolution output of the last layer to calculate the probability that the input pore structure is classified as true; the expression of the Sigmoid function is as follows:

[0220]

[0221] In the above formula, It is the convolution output of the last layer of the discriminative model.

[0222] Furthermore, a preferred method is provided, wherein the effect of improving the three-dimensional pore substructure generated by the three-dimensional pore generation adversarial network model of soil aggregates in S4, and optimizing the network framework structure and hyperparameters include:

[0223] After completing the construction of the soil aggregate three-dimensional pore generative adversarial network model, it is necessary to further optimize the hyperparameters of the model to achieve the best effect in the generation of the three-dimensional pore structure of soil aggregates; S4 The hyperparameters of the optimized soil aggregate three-dimensional pore generative adversarial network model are: batch size is 64, loss function optimizer is AdamW, learning rate is 0.0001, Beta1 is 0.6, and number of iterations is 500; the batch size is the number of samples selected for one training; Beta1 is the exponential decay rate of the first-order moment estimate.

[0224] Furthermore, a preferred embodiment is provided, wherein S5 includes:

[0225] For those who want to simulate The three-dimensional pore characteristics of soil aggregates: porosity, specific surface area and connectivity, respectively define the pore characteristic calculation function ,in:

[0226]

[0227] in, The specific calculation process is as follows:

[0228] Iterate over the voxel matrix For each element in , check the values ​​of the elements around it, find all the pore elements (value is 1), and for each pore element, check the element values ​​in the six directions of up, down, left, right, front and back. If the adjacent element is a soil element (value is 0), then the pore element has a boundary in that direction. Calculate the number of boundaries of each pore element in each direction and add them up to get the total surface area contribution of the pore element. By adding up the surface area contributions of all pore elements, you can get the surface area of ​​the entire three-dimensional binary image matrix. ;

[0229] in, The specific calculation process is as follows:

[0230] from Starting from a pore element in the , we gradually search for the pore elements around it and mark them as the same connected component. By traversing all pore elements, we can calculate the number of connected components. .

[0231] in, The specific calculation process is as follows:

[0232] from Starting from the soil element in , we gradually search for the soil elements around it and mark them as the same hole. By traversing all soil elements, we can calculate the number of holes. .

[0233] Traversing the 3D pore substructure dataset of soil aggregates The three-dimensional pore substructure of each soil aggregate , calculate the soil aggregate three-dimensional pore substructure data set The upper bound of each pore feature and the lower bound ,in:

[0234]

[0235]

[0236] According to the starting pore characteristics and the end pore characteristics of the pore characteristic evolution process to be simulated, Select the starting pore features and endpoint pore characteristics .

[0237] Furthermore, a preferred embodiment is provided, wherein S6 includes:

[0238] Step 1: Randomly sample from a normal distribution indivual dimensional random noise ;

[0239] Step 2: Noise The pore structures of soil aggregates are input into the three-dimensional pore generation adversarial network model to obtain the output pore structure. ;

[0240] Step 3: Calculate the pore structure Pore ​​characteristics , respectively calculate and (or ) ;

[0241] Step 4: Traverse each , repeat Step 5 to Step 8;

[0242] Step 5: From Randomly select 3 noises from , and , substituting into the following formula to obtain the new random noise :

[0243]

[0244] Step 6: The second norm of , then use replace ,in

[0245]

[0246]

[0247] in, represents the gamma function.

[0248] Step 7: Input into the soil aggregate three-dimensional pore generation adversarial network model to obtain the output pore structure ,calculate The pore characteristic value and (or ) ;

[0249] Step 8: If Less than , then Replace with ,Will Replace with ;

[0250] Step 9: Repeat Step 4 MaxIter times and output The noise vector corresponding to the minimum value in , denoted as (or ).

[0251] Furthermore, a preferred embodiment is provided, wherein S7 includes:

[0252] For the noise vector and , the expression for defining spherical linear interpolation is:

[0253]

[0254] in, , is a real number between [0,1].

[0255] set up ,Will As the input of spherical linear interpolation, the output noise vector is obtained .

[0256] Furthermore, a preferred embodiment is provided, wherein S8 comprises:

[0257] The noise vector The input is used in the soil aggregate three-dimensional pore generation adversarial network model to obtain the dynamic evolution process of soil aggregate three-dimensional pore output by the model. ,in .

Claims

1. A method for generating a dynamic evolution process of a three-dimensional pore structure of soil aggregates, characterized in that: The method comprises: S1: Obtain the pore part in the original soil aggregate image and construct a voxel matrix dataset of the three-dimensional pore structure of soil aggregates; S2: performing random sampling on the soil aggregate three-dimensional pore structure voxel matrix dataset to obtain a soil aggregate three-dimensional pore substructure dataset; S3: constructing a soil aggregate three-dimensional pore generative adversarial network model, wherein the soil aggregate three-dimensional pore generative adversarial network model includes a generative model and a discriminative model; The generative model is used to learn the distribution of three-dimensional pores of real soil aggregates and generate pseudo-pore structures; The discriminant model is used to learn to distinguish true and false pore structures and calculate the probability that the pore structure comes from a three-dimensional pore substructure data set of soil aggregates; S4: Training a soil aggregate 3D pore generative adversarial network model based on the soil aggregate 3D pore substructure dataset until the soil aggregate 3D pore generative adversarial network model generates a soil aggregate 3D pore structure that matches the distribution of the real 3D pore substructure characteristics; S5: Defining the starting point pore characteristics of the 3D pore structure evolution process of soil aggregates and endpoint pore characteristics ; S6: Use differential strategy to obtain pore characteristics and The corresponding soil aggregate 3D pore generation adversarial network model input noise vector and ; S7: Use spherical linear interpolation to obtain the noise vector and The intermediate noise vector between ; S8: The noise vector The data is input into the soil aggregate three-dimensional pore generative adversarial network model to obtain the dynamic evolution process of the three-dimensional pore structure.

2. The method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates according to claim 1, characterized in that: In S3, the objective loss function of the soil aggregate three-dimensional pore generation adversarial network model is: In the formula, To generate the model, is the discriminant model, is the real pore structure. is the mathematical expectation of the real pore structure, Generate model for input A normally distributed random variable, is the mathematical expectation for generating the pore structure.

3. The method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates according to claim 2, characterized in that: In S3, the generative model is composed of The layer is composed of deconvolution layers; The generative model is The deconvolution output of the first layer is batch normalized and ReLU nonlinearly activated. The generative model performs Tanh nonlinear activation on the deconvolution output of the last layer to generate the three-dimensional pore substructure of soil aggregates. The discriminant model is composed of The convolutional layers are composed in the reverse order of the generative model; The discriminant model is The convolution output of the layer is spectrally normalized and LeakyReLU nonlinearly activated; the discriminant model performs Sigmoid nonlinear activation on the convolution output of the last layer to calculate the probability that the input pore structure is classified as true; the expression of the Sigmoid function is as follows: In the formula, It is the convolution output of the last layer of the discriminative model.

4. The method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates according to claim 1, characterized in that: In S4, the hyperparameters of the soil aggregate three-dimensional pore generative adversarial network model are: batch size is 64, loss function optimizer is AdamW, learning rate is 0.0001, Beta1 is 0.6, and number of iterations is 500; the batch size is the number of samples selected for one training; Beta1 is the exponential decay rate of the first-order moment estimate.

5. The method for generating a dynamic evolution process of a soil aggregate three-dimensional pore structure according to claim 1, characterized in that S5 include: For the simulated The three-dimensional pore characteristics of soil aggregates are defined respectively as the pore characteristic calculation function , traverse the soil aggregate 3D pore substructure dataset The three-dimensional pore substructure of each soil aggregate , calculate the soil aggregate three-dimensional pore substructure data set The upper bound of each pore feature and lower bound ,in: According to the starting pore characteristics and the end pore characteristics of the pore characteristic evolution process to be simulated, Select the starting pore features and endpoint pore characteristics .

6. The method for generating a dynamic evolution process of a soil aggregate three-dimensional pore structure according to claim 1, characterized in that S6 include: S6.1: Random sampling from a normal distribution indivual dimensional random noise ; S6.2: Noise Input them into the soil aggregate three-dimensional pore generation adversarial network model to obtain the pore structure ; S6.3: Calculate pore structure Pore ​​characteristics , respectively calculate Pore ​​characteristics and dynamic evolution of initial pore structure or pore characteristics of dynamically evolving pore structures Euclidean distance , where S and E correspond to the starting point and ending point of the pore structure during the dynamic evolution process, respectively; S6.4: From Randomly select 3 noises from , and , substituting into the following formula to obtain the new random noise : Where F is a random number between [0,1]; S6.5: Yes The second norm of , then use replace Otherwise it is not correct Modify the value of in in, represents the gamma function; S6.6: Input into the soil aggregate three-dimensional pore generation adversarial network model to obtain the output pore structure ,calculate The pore characteristic value and or Euclidean distance ; S6.7: If Less than , then Replace with ,Will Replace with ; S6.8: Traverse each , repeat S6.4 to S6.7; S6.9: Repeat S6.8 MaxIter times, output The noise vector corresponding to the minimum value in , denoted as the noise vector of the initial pore structure of dynamic evolution Or the noise vector of the dynamically evolving pore structure .

7. The method for generating a dynamic evolution process of a soil aggregate three-dimensional pore structure according to claim 1, characterized in that S7 include: For the noise vector and , the expression for defining spherical linear interpolation is: in, , is a real number between [0,1]; set up ,Will As the input of spherical linear interpolation, we get the noise vector and The intermediate noise vector between .

8. The method for generating the dynamic evolution process of the three-dimensional pore structure of soil aggregates according to claim 1, characterized in that: S1 includes: S1.1: Use ImageJ software to capture the middle area of ​​the soil aggregate CT scan image as the feature domain; S1.2: Import the 2D image stack into ImageJ software and perform pre-processing for image segmentation, including: S1.2.1: Use the Enhance Contrast command to normalize the image to eliminate the brightness difference between different CT slice images; S1.2.2: Use the Despeckle command to remove image noise; S1.2.3: Use Median operation to achieve smooth operation; S1.3: When performing image segmentation operations, use the Threshold operation to perform image threshold segmentation to separate the pore parts in the image; generate a binary image and convert it into a voxel matrix whose elements are all 0 or 1.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor runs the computer program stored in the memory, the steps of the method according to any one of claims 1 to 8 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of computer instructions, and the plurality of computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 8.